PERSPECTIVES

have focused largely on a limited region of

OPINION
chemical space that is defined by established
molecular frameworks.
Expanding the medicinal chemistry This raises the question of the potential
impact of synthetic chemistry biases and
synthetic toolbox limitations on the chemical and structural
characteristics of compounds explored
so far in medicinal chemistry efforts. A
Jonas Boström, Dean G. Brown, Robert J. Young and György M. Keserü pioneering investigation of this issue was
reported in 2008 by Roughley and Jordan12.
Abstract | The key objectives of medicinal chemistry are to efficiently design and Analysing a relatively small data set of
synthesize bioactive compounds that have the potential to become safe and published structure–activity relationship
efficacious drugs. Most medicinal chemistry programmes rely on screening (SAR) studies from AstraZeneca,
compound collections populated by a range of molecules derived from a set of GlaxoSmithKline and Pfizer, they found that
known and robust chemistry reactions. Analysis of the role of synthetic organic medicinal chemistry programmes are based
on a limited number of reactions. This
chemistry in subsequent hit and lead optimization efforts suggests that only a few
finding was recently reinforced on a larger
reactions dominate. Thus, the uptake of new synthetic methodologies in drug data set 13, based on a literature analysis at
discovery is limited. Starting from the known limitations of reaction parameters, two points (1984 and 2014). Brown and
synthesis design tools, synthetic strategies and innovative chemistries, here we Boström confirmed that a few reactions
highlight opportunities for the expansion of the medicinal chemists’ synthetic dominate contemporary practice and,
toolbox. More intense crosstalk between synthetic and medicinal chemists in more importantly, that new synthetic
discoveries have a very slow uptake. Five
industry and academia should enable enhanced impact of new methodologies in major reactions are the most dominant and
future drug discovery. most frequently used13 (FIG. 1a) and represent
more than 80% of reactions used for drug
Medicinal chemistry encompasses the these characteristics by 2010; the top 50 discovery purposes12.
design and synthesis of novel bioactive frameworks still covered half (48–52%) It should be noted that most analyses of
compounds that have the potential to of approved and experimental drugs7. this type will show power-law distributions.
become drugs. It typically involves cycles Furthermore, Taylor and colleagues found Thus, a relatively small number of reactions
of iterative optimization of compounds that the 1,175 drugs marketed before 2013 dominate the lists, with long tails of
to enhance their efficacy, safety and contained 351 unique ring systems8, which infrequently used synthetic reactions.
pharmacokinetic characteristics, leading represent 2% of the possible combinations Nonetheless, the most frequently used
to the selection of a candidate drug that of known monocyclic and bicyclic ring reactions (as well as the top frameworks and
ultimately will need to be manufactured systems9. This study also revealed that four ring systems in drugs derived from these
on a large scale if it progresses towards out of five (83%) of the most frequently reaction types) in medicinal chemistry have
regulatory approval. used ring systems were first used in drugs not changed much over time. A similar
An astronomical number of developed before 1983 and, on average, six conclusion was drawn by Schneider and
small-molecule compounds are purportedly new ring systems appear each year, found in colleagues, who analysed a comprehensive
synthetically feasible1,2,3, so although less than one-third of the new drugs. set of 1.15 million unique reactions in the
the number of reported molecules has A similar conclusion was drawn by Pitt and pharmaceutical patent literature14. The
reached 135 million, this still represents colleagues10, who showed that very few authors concluded that the medicinal
only a tiny proportion of the feasible (7%) of the 24,000 reasonable monocyclic chemistry toolkit is biased towards a small
drug-like compounds4. In a pioneering and bicyclic ring systems collected in the set of standard reaction types, in particular
investigation into the characteristics of the VEHICLe database had been synthesized. to amide bond formations and Suzuki–
kind of compounds that have so far been More importantly, they also concluded that Miyaura couplings; 123 different reaction
developed as drugs in 1996, Bemis and the rate of publication of novel examples types accounted for 95% of the reactions,
Murcko reported that half of the 5,120 drug was 5−10 per year. More recently, Reymond which increased to 159 (27%) over 35 years,
molecules analysed could be described by and co‑workers enumerated the chemical but this growth has stagnated over the
only 32 of the most frequently occurring universe of ring systems and found that the past 10 years. Alkylation and acylation
molecular frameworks5. They also found known structures cover a fraction (1.4%) reactions plus protective group chemistry
that the top 20 side chains represent 73% of the chemically feasible ring systems11. are other common transformations
of the total used in marketed drugs6. Wang Overall, the message from these studies used in the discovery phase (FIG. 1b). The
and Hou found no significant change in is that medicinal chemistry efforts so far formation of new carbon–carbon bonds,

NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 1

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

the transformation of functional groups because the functional group tolerance of and thereby supporting the synthesis of
and reductions represented over 5% of the the reactions limits the scope of SAR the right compound at the right time more
reactions analysed, while oxidations and the studies. Furthermore, focusing only on effectively.
formation of heterocycles are rare. Similar known robust reactions may undermine the
trends were observed in process chemistry innovative spirit and personal creativity of Why is the toolbox limited?
applications15,16. researchers and delay the implementation Early drug discovery is viewed and practised
The use of a limited set of reactions in of new scientific and technological by many as a numbers game, in which the
most medicinal chemistry efforts potentially achievements. In this Perspective, we briefly more compounds made and tested, the
compromises the quality of drug candidates overview current practice and then discuss greater the chance for success. Although
or the opportunity to identify viable opportunities and challenges in extending alternative drug discovery strategies based
leads for challenging targets; for example, the synthetic toolbox of medicinal chemists on the screening of smaller libraries of

a
Other reaction types
Amide formation
O
O R2
HN R2
+ R1 N
Electrophilic reactions of amines 19% R1 OH R3
32% R3
R1 O R1 R3
N H + N
R2 H R3 R2
8%

9%

Amine Boc-deprotections 19%

13%
R1 O R1
N N H
R2 O R2 Suzuki–Miyaura reaction R2

B(OR)2 X
R1 + R2 R1
Aromatic nucleophilic substitution reaction (SNAr)
R2 X = Cl, Br, I, OTf
R2 N
HN R3
R1 N + R1 N
R3

b 45

40 Library synthesis (577 reactions) 2005–2009

Lead optimization (12,115 reactions) 2005–2009
35
Process chemistry (1,693 reactions) 1997–2005
Reaction frequency (%)

30

25

20

15

10

0
n ) g ps n n n n ic er
tio es in io io tio io all th
la id m ou rs at c at et
lky am fo
r gr nv
e
xid du rm m
O
n g Re fo
A fo nd in co O no
sul bo ect ter c le rg
a
s/ C ot in cy O
id
e C– Pr up
te
ro
(a
m g ro He
io
n n al
y lat ct
io
Ac n
Fu
Figure 1 | Common chemical reactions in drug discovery and develop- compounds in drug discovery programmes are given (data sourced from
ment. a | The uptake of new chemical reactions is slow, and a few have REF.13). b | Reaction types used in library synthesis (GlaxoSmithKline
Nature Reviews (GSK):
| Drug Discovery
consistently dominated the past decades. The percentages of the five most 2005–2009)189, lead optimization (AstraZeneca, GSK and Pfizer: 2005–
frequently used reaction types (amide formation, Suzuki–Miyaura reaction, 2009)12 and process chemistry for the synthesis of candidate drugs
aromatic nucleophilic substitutions, amine Boc-deprotection and electro- (AstraZeneca, GSK and Pfizer: 1997–2005)15,16. The most common reactions
philic reactions with amines) along with all other reaction types to produce show similar popularity in discovery and development settings.

2 | ADVANCE ONLINE PUBLICATION www.nature.com/nrd

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

carefully selected fragment compounds17,18
or virtual screening efforts19,20 relying
on low-throughput assays have achieved
success, numbers-oriented thinking still
provides the basis for large-scale screening.
Thus, a compound collection of structurally
diverse molecules is considered to be crucial
for the success of drug discovery efforts in
most companies pursuing novel molecular
drug targets21.
Nevertheless, current screening collections
are mostly populated by compounds made
from a limited set of reactions, frequently
using commercially available building blocks.
Although these reactions may be ideal for
the synthesis of large numbers of compounds
for screening, one analysis has shown that
the reliance of these reaction types may
lead to an overpopulation of certain types
of molecular frameworks, such as biaryls13.
Furthermore, the chemistry embedded in the
initial hit structures sets the stage for the type
of chemistry pursued through to the clinical
candidate22,23,24,25.
The limited set of reaction types used in
medicinal chemistry can be rationalized by
the use of several criteria in their selection.
The first criterion is the availability of
starting materials and reagents. The second
is the ease of synthesis (such as short reaction
times, moderate temperatures and high
yields with limited by‑products). Indeed,
analysis of the synthetic methodologies
used over the past 240 years in more than
6.5 million organic reactions26 indicated
that key parameters, such as reaction time,
temperature, pressure and solvent, are biased
by anthropogenic factors. For example,
half of the reactions were complete within
<3 hours, and >90% of reactions were run
at atmospheric pressure, typically between
−80 °C and +200 °C. Furthermore, medicinal
chemists are often reluctant to make
difficult-­to-synthesize molecules without
compelling precedence or predictions, and
the strength of computer-aided design is still
largely geared towards prioritizing lists of
compounds and designing libraries rather
than predicting a single optimal compound.
Consequently, chemists may prefer to focus
on simple reactions that can efficiently
cover a lot of ground27 and, indeed, this
is what is observed when analysing what
is actually made in medicinal chemistry
laboratories12,13,28. Access to highly efficient
purification technologies also has an impact
on synthetic choices. For example, chiral
syntheses are often avoided in favour of
chiral separations where needed, owing to
the increased cost and complexity of (chiral)
reagents and timelines.
The third criterion is broad functional
group tolerance in reagents (for example,
more diverse building blocks can be used
in the reactions). It should be noted that
publications disclosing new synthetic
transformations are often from academic
groups intent on achieving high yields with
some structural variations, rather than using
a diversity of functionalities, and often use a
limited set of examples that are less complex
and functionally similar 22,29. Furthermore,
polar functionalities are usually sampled less
often because available work‑up/purification
procedures favour more lipophilic
molecules30, which may be less desirable for
medicinal chemistry purposes. Based on
these data, chemists often cannot discern
whether more complex compounds with a
differing physicochemical make‑up (typically
equipped with polar functionalities30) will
successfully undergo the same reaction
without time-consuming optimization
or different protection strategies. This is
relevant to the final criterion — management
and cultural reasons, such as the common
focus on rapid cycle times in industry (linked
to performance metrics, deadlines and
bonuses), as well as structural conservatism,
variable training 31 and experience.
Most importantly, synthetic technologies
should fulfil the objectives of medicinal
chemists, delivering the right compound
at the right time, and so constraints in
resources, time and budget can strongly
affect the synthesis of new compounds.
Indeed, a recent analysis of the synthetic
aspects of medicinal chemistry 32 concluded
that medicinal chemists are inevitably biased
towards robust reactions — essentially,
chemical transformations that are applicable
to structurally diverse substrates, that tolerate
a range of functionalities and that can be
realized on simple equipment in a reasonable
time frame. Hartenfeller and colleagues
suggested similar soft criteria33 after
considering reactions that yielded products
structurally relevant for drug discovery from
readily available starting materials with broad
applicability and functional group tolerance.
Finally, it has become standard practice
for companies to outsource synthetic
chemistry to contract research organizations
(CROs). Paying CROs on a per-compound
basis has been common procedure in the
past, and this has also created a bias towards
the use of robust reactions.
How could the toolbox be expanded?
Based on an understanding of the nature of
the limitations of the medicinal chemistry
toolbox and the underlying factors above,
the remainder of this Perspective highlights
strategies to expand the toolbox with the
aim of enabling more effective exploration
of chemical space. These strategies can be
divided into four broad types, which may
also be integrated: technologies that enable
accurate predictions of synthetic routes to
target compounds; technologies that enable
greater automation of compound synthesis
using a wider range of reaction conditions
and reaction types; application of novel
chemical transformations; and application of
broad synthetic strategies and platforms.
Predicting synthetic routes
How to best synthesize the designed
compounds is a key question in all medicinal
chemistry projects. Unless there is an obvious
route to the target compounds, the first
attempt to predict synthetic feasibility is
often to perform an exact structure search
for synthetic precedence, followed by a
substructure search if the exact compound
is not found. Contemporary tools such as
SciFinder and Reaxys provide easy access
to chemical databases with reaction data
extracted and curated from the scientific and
patent literature and are thus often used for
this purpose. In many cases, synthetic routes
to compounds that are sufficiently similar to
be useful are found.
In this context, synthesis planning can be
considered to be a pattern recognition process
in which substructures are identified and
subsequently associated with retrosynthetic
routes. Intuitive retrosynthetic disconnections
are influenced by past education, knowledge
and experience. To create improved ideas and
to diminish personal biases, a collaborative
approach to solve issues is for groups of
chemists to discuss synthetic routes. To make
this process even broader and more efficient,
computational tools developed for the design
and automated assessment of synthetic routes
provide an extension covering current organic
chemistry knowledge. Although the current
use of in silico synthesis prediction tools is
scattered, this is likely to change because this
field is undergoing rapid progress34.
Synthesis design tools can be grouped
into two major categories: retrosynthesis and
forward synthesis methods. In retrosynthesis,
the design process starts from the desired end
product and goes backwards, constructing the
synthesis tree of alternative pathways to the
product while also identifying feasible routes
to known intermediates or available building
blocks. By contrast, forward synthesis
methods predict chemical reactions between
available building blocks and the reaction
conditions needed.

NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 3

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

The performance of retrosynthesis
and forward synthesis tools are both
heavily dependent on the size and the
quality of the underlying chemical
reaction data source and their underlying
computational methods. A current
challenge in developing new improved
synthesis design tools is the lack of large
high-quality data sets. In general, the
extensive SciFinder or Reaxys data sets
are not available for developers. Hence,
in most cases, the reaction data used in
the development of in silico tools are
manually or automatically extracted from
the primary scientific and patent literature,
which has several limitations. Reactions
that do not work are rarely reported, and
negative data are important for developing
a predictive tool. Moreover, literature
protocols include a large number of
reactant–reaction combinations that are
neither reproducible nor reliable35, and
many synthetic transformations identified
this way may be challenging to implement
owing to issues such as high technical
expertise, expensive or custom catalysts,
high functional group sensitivity and toxic
reagents or intermediates.
Electronic laboratory notebooks
(ELNs), which are routinely used in
pharma companies, could be considered
as alternative sources of reaction data
that potentially give more confidence in
reliability, reproducibility and relevance,
and could also alleviate the issue of the lack
of negative synthesis data in the literature.
AstraZeneca and Roche have reported
extracting their ELN content to internal
reaction databases36, and NextMove14
and EBI37 provide solutions to create and
maintain reaction databases from ELN
systems. That said, the standardization and
curation of ELN data are major efforts (for
example, there are many reasons for a yield
of 0%, from the chemist not starting the
reaction to the reaction not working at all
for unknown reasons) and have historically
not been prioritized. A recently developed
data exchange format aims to facilitate the
precompetitive exchange of reaction data36.
In the future, data collected from automatic
synthesis robots will lead to more reliable
data owing to the systematic manner in
which they are generated and recorded38.
This in turn can be used to improve the
predictive power of synthesis design
tools39. These methods use expert-based
or automatically extracted reaction rules
(for example, REFS40,41) or machine-­learning
approaches (for example, REFS42,43). Although
rule-based approaches are relatively well
established, non-rule-based approaches
are emerging. Considering the increasing
amount of available reaction data, the
latter is expected to dominate future
developments.
Retrosynthesis tools are mainly
rule-based. They can consider the number
of steps, the availability of starting materials,
the types of building blocks and the ease
of incorporating them40. Although these
methods support the automatic evaluation
of possible synthetic paths, they cannot yet
fully evaluate synthetic feasibility objectively,
and their use has so far been limited.
This drawback has been compensated
by methodologies for quantitatively
predicting the synthetic accessibility by
the cheminformatics analysis of structural
descriptors. Software such as ChemPlanner
(Wiley), ICSynth (Infochem) and Chematica
(Merck KGaA) can assist medicinal chemists
with retrosynthetic analyses, as highlighted
in published studies. In a study comparing

Glossary
Chemical space
Chemical space is a nebulous term used in various ways,
but pertinent to drug discovery, it is the classification of
molecules in terms of their physicochemical make-up, such
as size, shape, lipophilicity, charge and hydrogen-bonding
potential, which together can be used to describe the
chemical space occupied.
Design–make–test–analyse (DMTA) cycle
The iterative central process in lead optimization, involving
a cycle of four steps: design (a hypothesis is constructed to
improve the profile of the lead molecule); make
(compounds exemplifying the design are synthesized); test
(synthesized compounds of confirmed structure and purity
are tested in one or more carefully constructed and
controlled assays); and analyse (the experimental data are
analysed, and the results are used to amend a design
hypothesis for the next cycle).
Drug-like compounds
Drug-likeness is another term that is used in various ways,
often to describe the possession of physicochemical
properties that are typical of orally absorbed
small-molecule drugs. Lipinski’s rule of five (Ro5) is one
common metric; if no more than one of the following
criteria is exceeded, then there should be a reasonable
chance of oral bioavailability: molecular mass <500 Da,
cLogP <5, number of hydrogen-bond donors <5 and
number of hydrogen-bond acceptors <10. More recently,
it has been demonstrated that oral activity is feasible
beyond Ro5, and these programmes follow specific
principles that contribute to the oral bioavailability.
Fragment
A simple, small and relatively polar molecule with ~8–17
heavy atoms, often screened using sensitive biophysical
techniques (such as X‑ray crystallography, NMR
spectroscopy and surface plasmon resonance) to identify
inherently weak binders that can be elaborated into lead
compounds.
Functional group tolerance
The range of organic functionalities that do not react with
or impede the reagents and/or catalysts involved in a
transformation. As drug molecules are predisposed to
contain charged or hydrogen-bonding motifs to, for
example, achieve potency and selectivity, this can often
cause issues and interfere with catalysts, ligands and
reactive partners.
Lead-like
Lead-likeness is a term that describes an aspirational
profile for a screening collection of molecules that have
physicochemical properties, together with predicted
safety, pharmacokinetic and pharmacodynamic data
and complexity, that bridge fragment space and
drug-like space, as well as appropriate chemical
functionalities that can be used in the optimization of
the molecules into candidate drugs.
Quality of drug candidates
Like chemical space, the notion of compound quality is
used in various ways, but physicochemical parameters
can predict the likely quality of a compound, in
conjunction with pharmacokinetic and
pharmacodynamic data, giving confidence in probable
exposure, efficacy and safety. This should not be
prescriptive, but more optimal properties indicate a
higher likelihood of success. Note that the actual set of
physicochemical parameters is dependent on the target,
the compartment where the target is engaged and the
route of administration.
Robust reactions
Reproducible chemical transformations applicable to
structurally diverse substrates, tolerating a range of
functionality and able to be realized on simple equipment
in a reasonable time period. Factors for robust reactions for
medicinal chemistry include the following:
• Provide structures relevant for drug discovery
• Technically straightforward (no special equipment needed)
• Moderately sensitive to reaction parameters
• Broad applicability (also with polar substrates)
• Broad availability of starting materials and reagents
• Broad functional group tolerance, including polar
functionalities
• Time for delivery of the target compounds is reasonably
short (<1 month ideally)
• Simple operational procedure (minimal training and
support needed)
• Low-risk reagents to comply with often onerous local
safety rules
A full-size poster depicting the set of most popular robust
reactions (available online for downloading; see
Supplementary Fig. 1) illustrates their impact on drug
discovery. Our hope is that displaying this poster in offices
and laboratories could highlight the importance of
expanding the medicinal chemistry synthetic toolbox and
stimulate debate.
Structurally diverse substrates
The breadth of diversity of a given reaction type is
dependent on the accessibility and intrinsic reactivity of
the substrates and/or building blocks involved in the
reactions. A reaction that can use a number of different
reactive groups can be advantageous to medicinal
chemists, as it will allow access to more analogues.

4 | ADVANCE ONLINE PUBLICATION www.nature.com/nrd

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

the routes predicted by ChemPlanner to
that developed by a chemist for a highly
complex triazole44, the software identified
the same route as the chemist and also found
more feasible alternative routes, one of
which shortened the synthesis of the target
compound. In a study at AstraZeneca on five
target compounds, including synthetically
challenging drug intermediates45, ICSynth
identified new, non-intuitive solutions to
synthetic problems compared with groups
of medicinal chemists under controlled
conditions, and novel disconnection
strategies and non-trivial reaction
sequences were subsequently used for
the development of new synthetic routes.
Finally, in a study in which Chematica was
used to design routes for the synthesis of
eight bioactive compounds46, the routes
enabled the efficient preparation of the
target compounds with improved yields and
provided new or more feasible alternatives to
known procedures.
Although machine learning is used less
frequently in retrosynthetic predictions,
Segler and colleagues reported34 a new
approach combining Monte Carlo tree
search and symbolic artificial intelligence.
An expansion policy neural network
was used to guide the search, and the
most promising solutions were selected
by a filter network to preselect the most
promising retrosynthetic steps. Training
the neural nets on the Reaxys database
enabled this approach to provide feasible
synthetic routes for more molecules much
faster than rule-based methodologies.
The results were comparable to reported
literature syntheses.
Forward synthesis tools use both
rule-based and machine-learning
algorithms. Rule-based approaches require
transformation of the available reaction
knowledge to reactivity, selectivity (transform
rules) and functional group tolerance rules.
This determines which reagents can generate
products and which product structures
can be obtained from the available list of
in silico reactions. Most disclosed methods
apply forward-­reaction-based technology
for only a single in silico step, but ICFRP
from Infochem and SynSpace by ChemPass
(the only commercialized software packages
available to date) can handle multistep
in silico forward synthesis. ICFRP is a
general literature-­trained tool that can be
used for compound design in the forward
direction and requires human intervention
to deal with the potential combinatorial
explosion of possible reaction products
arising in multistep sequences. SynSpace
from ChemPass is a fully expert-trained
system that considers long-range functional
group effects and has built‑in combinatorial
explosion management to assist medicinal
chemists in idea generation for scaffolds and
analogues.
The major disadvantage of these methods
is their limited scalability, as decoding the
entire organic chemistry reaction knowledge
into complete and reliable transform rules is
non-trivial. In forward reaction prediction,
one of the most challenging tasks is the
prediction of selectivity in a given reaction.
The selectivity problem can be broken down
into groups: chemoselectivity, regioselectiv-
ity and stereoselectivity. For chemoselectiv-
ity, the question is which of many functional
groups present will react most rapidly with
a given reagent. An example would be a
simple carbonyl reduction, in which one
type of reagent (for example, NaBH4) will
selectively reduce a ketone in the presence
of an amide, whereas another reagent such
as BH3 will display the opposite selectivity.
Maintaining a set of rules for each reagent
would be an alternative but can also become
cumbersome.
For regioselectivity, the challenge is to
accurately predict which of several possible
functional groups present will react. This
is probably most prominent in reactions
of C–H bonds, which are important in the
recently popularized late-stage function-
alization strategies (see further discussion
below). An example is electrophilic
aromatic substitution — for example,
aryl bromination — for which in silico
methods frequently suggest substitutions
that do not correspond to real observations,
especially for complex heterocyclic
systems. Mechanistic approaches, such as
the recently published RegioSQM47, have
begun to address these issues, although the
approaches are computationally demanding.
Finally, predicting the stereoselectivity
of reactions is also challenging. Such
selectivities are frequently determined by
through-space interactions with remote
substituents and are likely to require 3D
information. Although very accurate
predictions can be made48, these protocols
are time consuming and not general yet.
The development of new synthetic
methods (see discussion below) and
computational retrosynthesis or forward
synthesis tools in combination can be
especially powerful for increasing the
structural diversity of synthetically feasible
compounds. A number of recent initiatives
(such as EVOSpace by Evotec, ULTIMATE
by Mcule and REAL by Enamine)
demonstrate that combining synthetic
methods with the appropriate reactivity,
selectivity and functional group tolerance
information into a computational multistep
forward synthesis tool is a viable approach to
new and structurally diverse compound sets.
These approaches could benefit from recent
building block initiatives49,50,51 that increase
the number of accessible products, ‘design of
experiment’ approaches that are increasingly
employed in the optimization of reaction
conditions52,53 and large-scale automated
reaction testing 39 that helps to define the
parameter range of the reactions applied in
forward synthesis programmes.
Automated methods that harness the
power of artificial intelligence and machine
learning are expected to play a key role
in the faster adoption of new synthesis
methods and in facilitating the synthesis of
previously difficult-to-make compounds54.
The progress that has recently been made
with these approaches by various groups
using several different techniques strongly
indicates that synthetic chemistry can
be taught to machines using large and
high-quality data sets34,41,55,56,57,58,59. Thus far,
most of the publications have disclosed early
validation results, with 50–80% success rates
in predicting single-step reaction products,
with training sets machine-extracted from
large databases such as Reaxys or a patent
transform set 14. However, so far, no major
study has been published for real-life
multiple syntheses, preventing in-depth
validation by organic and medicinal
chemists. Nevertheless, this is an emerging
and rapidly evolving area (BOX 1), and these
techniques have the potential to cover a
significant fraction of relevant reactions
because they can be based on very large
unbiased databases of reactions (for example,
from company electronic laboratory
notebooks and automated synthesis
machines). In addition, we see the potential
to predict new synthetic methods41,58. The
high generality of the approach, however,
means that it could be particularly important
to incorporate practical considerations (see
BOX 2) in reaction planning, ranking and
selection schemes.
Overall, such tools will enable the more
effective generation of better results, thereby
aiding chemists, who struggle to find time
to exhaustively search the literature for
optimal routes outside of their previous
experience. In addition, the tools can also
provide value in evaluating safety, cost
and operational issues when prioritizing
potential synthetic routes. Although the
implementation of synthesis planning tools

NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 5

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

often requires substantial investment, some It should not be forgotten that synthetic Expanding reaction conditions
vendors have incorporated pay-for-service chemistry requires planning and creativity, The outcomes of chemical reactions
options and different collaboration models and this is something at which machines are heavily influenced by the reaction
that make the tools accessible for smaller still do not excel. Empowering chemists parameters. Enabling technologies such as
commercial organizations. More recently, with the appropriate in silico tools will be microwave-heated reactors, microreactor
non-commercial tools have also become more effective and efficient than trying to and microfluidic devices, flash reactors
available59. develop one-stop-shop machines, at least and high-pressure/high-temperature
However, there are many idiosyncrasies in the foreseeable future. Paraphrasing autoclaves widen the opportunities for
and poorly predictable practical hurdles Brynjolfsson and Mitchel60, we believe that changing conditions relative to conventional
(such as work‑up and product separation) “AI won’t replace chemists, but chemists who batch processes26,62.
that hinder fully automated processes. use AI will replace those that do not”61. Expanding the range of conditions
can lead to more efficient and/or quicker
synthetic routes and, more importantly,
Box 1 | Example of the impact of automated compound synthesis and testing can provide opportunities for the discovery
of new chemistries, enabling the synthesis
Automation can be particularly beneficial for repetitive and tedious tasks. A recent paper by Cernak of novel compounds (TABLE 1). Microwave
and colleagues highlights the potential impact of emerging technologies in automatically
heating is widely used in both discovery and
synthesizing and testing compounds in a drug discovery setting186. High-throughput nanomole-scale
synthesis (consuming less than 0.05 mg of substrate per reaction) was coupled with an
development phases to improve reaction
affinity-selection mass spectrometry (ASMS) binding affinity assay. Robust chemical reactions were rates and often results in higher yields or
used; 20 reactions were performed using amide formations, Suzuki couplings or Buchwald–Hartwig cleaner reaction profiles. For example,
carbon–nitrogen (C–N) couplings. A single reaction condition was enough for the amide couplings, the microwave-assisted multicomponent
whereas several conditions were surveyed for the other two reactions; although, it should be noted reaction of aldehydes, amines and alkynes
that a large number of reactions did not produce an observable product. The ASMS bioassay was (A3‑coupling) in water affords secondary
used to rank the affinity of the products by titrating the concentration of protein while maintaining a propargylamines as the precursor of
constant concentration of the compound. Synthesis (without the need for time-consuming nitrogen-containing heterocycles (such as
purification steps) and testing were performed rapidly with minimal consumption of starting pyrrolidines, pyrroles, oxazolidinones and
materials (120 mg), and potent inhibitors for three kinase targets (mitogen-activated protein kinase 1
aminoindolizines) and intermediates of
(MAPK1), MAP kinase-activated protein kinase 2 (MK2) and CHEK1) were identified (see figure).
This approach successfully demonstrated reductions in timelines and resource demands and is an
biologically active scaffolds (β‑lactams and
initial indication of what may be possible as efforts progress to include machine-learning algorithms peptidomimetics)63. Chemical reactions
and other artificial intelligence approaches to predict and select which compounds to make (for performed at high reaction temperature and
example, REF.187). Nevertheless, there are many hurdles to address, such as enabling the use of new high pressure might be used for the synthesis
chemical reactions and accurate multiparameter optimization, including functional assays. of novel scaffolds. Djuric and colleagues
Ki, inhibition constant; rt, room temperature.
showed that high-pressure and high-­
temperature chemistry makes substituted
fused pyrimidinone and quinolone
H O
H O R1 = H derivatives accessible with high yields64. The
N N
R1 HO technology also enabled the fast and efficient
N N R1
N
OH H R2 N synthesis and scale‑up of other challenging
R2 R2 =
HN HN heterocyclic scaffolds.
HATU, DIPEA, rt, 20 h
19 reactions Advantages of microfluidics–microreactor
flow chemistry include enhanced mixing,
49% yield
Cl MAPK1 Ki = 35 nM improved heat transfer and wide temperature
Cl
and pressure ranges. A novel synthesis
Br
R NH technology developed by the Ley group
N OH
N O is an elegant example of the reaction
B safety and automation benefits of the flow
R OH R=
technology. The flow synthesis of hazardous
Catalyst, base, rt, 20 h unstable aryl and vinyl diazo compounds
HN HN O
O 144 reactions from the corresponding hydrazones and
NH
89% yield their subsequent reaction with boronic
NH MK2 Ki = 59 nM acids resulted in a new, convenient and
metal-free sp2–sp3 cross-coupling strategy 65.
NH
NH The sequential addition of different diazo
OH R1 H
or N HN compounds to suitable boronic acids
N N
B N
R OH H R2 N provided a strategy for the synthesis of
NH Catalyst, base, rt, 20 h NH R1=H, R2 = structurally complex molecules66. Flow
170 reactions microreactors have also been used for
O
S
O
S flash chemistry, performing extremely fast
N N OMe reactions in a highly controlled way to obtain
Br
85% yield products that are challenging or practically
Nu
CHEK1 Ki = 23 nM
inaccessible in batch conditions67. A recent

Nature Reviews | Drug Discovery

6 | ADVANCE ONLINE PUBLICATION www.nature.com/nrd
©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

example reporting the chemoselective
functionalization of ortho-lithiated
aryl carbamates shows that even rapid
transformations, such as the anionic Fries
rearrangement, can be outpaced with
sub-millisecond microfluidic mixing 68.
Attempts are being made across industry
to automate and integrate many of the
routine aspects of the design–make–test–
analyse (DMTA) cycle69. Automation requires
robustness, and as the synthetic toolbox
advances, an increasing number of chemical
reactions may be suitable for automation
and robotic synthesis. The most ambitious
vision would integrate machine-­learning
algorithms coupled to an automated
synthesis and testing module. A continuous
loop of newly designed and synthesized test
molecules informed by real-time screening
data could be driven by computer and
robotics equipment.
Although this bold ambition has yet
to be fully realized, several groups have
made important in‑roads into this area.
Burke and colleagues have demonstrated70
an automated synthesis system that can
handle a variety of organic transformations
and purifications, built on the versatility
and scope of the recently discovered
N‑methyliminodiacetic acid (MIDA)
boronate reagents. Djuric and colleagues at
AbbVie reported an integrated automated
platform that incorporates synthesis,
purification and testing to complete the
DMTA cycle within hours (compared
with days using the traditional model),
using amide and Buchwald libraries as two
illustrative examples69,71. Eli Lilly has also
described a large-scale automated drug
discovery platform that has been reported to
be capable of performing >16,000 reactions
on the 100 mg scale per year on a diversity
of reaction types such as organometallic
cross-coupling reactions, alkylation
reactions, reductive aminations and
multicomponent synthesis reactions72.
The concept of using self-optimized
synthesis techniques is an exciting extension
that could be used on these platforms and
was recently described and demonstrated
by Buchwald and Jensen. In this model, a
design of experiment (DoE)-based algorithm
was coupled to online high-performance
liquid chromatography (HPLC) analysis to
optimize reaction yields in a Suzuki–Miyaura
reaction73. A comprehensive review of the
history and state of the art of automated
synthesis platforms was recently published by
Burke and Trobe and provides historical and
modern perspectives on the advances and
challenges of automated synthesis74.
Hartwig and Troshin also demonstrated75 handling and maintaining screening
the concept of reaction mining, which collections of 1 to 3 million compounds
couples automated synthesis and a from commercial sources. Automated77 and
deconvolution method based on mass integrated synthesis platforms with in‑line
spectral informatics to identify unknown purification are emerging for make–analyse–
reaction types as one solution to help expand format–bioassay systems78,79, but these often
the chemist’s toolbox in these types of have rudimentary purification steps and
automated platforms. Finally, it should be rely on prosaic chemical methodologies.
noted that many reported automation efforts Developments in affinity-based assay
typically involve the use of robust chemical systems may circumvent some impurity
reactions, such as the top five most popular issues, but, ultimately, highly pure
reaction types cited above, and readily compounds will be required to qualify hits
available reagents. Thus, expanding the and properly interpret data.
scope of the types of reaction used on these Methods to purify compounds are
platforms should also expand the uptake and typically variations on chromatographic
application of the technology 13. methods, frequently HPLC platforms,
A major issue in hit identification is which are often automated, based on
the provenance and purity of samples, mass-directed or UV‑triggered collection.
contributing to false results that waste A major drawback of HPLC purification is
considerable time and effort76. The reliable the need to liquid-load samples, which can
production of pure compounds remains an cause ‘surfing’ of products on this solvent
issue, especially given the cost of populating, front. Purification is fundamentally a
Box 2 | Example of the impact of continuous-flow technologies
Researchers at Eli Lilly recently published a pioneering example of the kilogram-scale manufacture
of an active pharmaceutical ingredient — the investigational anticancer drug prexasertib — in
a continuous-flow system188 (see figure). The flow process allows greater flexibility of mixing and
conditions, thus making the process more controllable and reproducible; additionally, reactor design
enables novel manipulations such as photochemistry, more controlled use of heterogeneous
catalysts or ozone generation. In this example, the continuous process enabled the production of
3 kg per day of current good manufacturing practice (CGMP) material in standard laboratory fume
hoods, thus avoiding the need for substantial capital investment in a plant while allaying many safety
issues owing to the small volume of hazardous mixing at any given time. Other processes remain
unpublished, and the potential for the technique is vast, not least because of rapid progress with the
scale‑up reactions, reproducibility and accessibility of non-standard conditions.
O O N
O O
DMF-DMA, alkylate,
H2NOH·HCl KOH, H2NNH2
O O
OH
N O
H
N
NH2 CN
HN
O N N
N O
N
Cl CN N
H N N
H
O O
O O
N O
H N O
H
N
CN
HN
N
Deprotection, O
N
salt manipulation
N
H
O
NH2 .mesylate or lactate
Prexasertib

NATURE REVIEWS | DRUG DISCOVERY Nature Reviews

ADVANCE ONLINE| Drug Discovery| 7
PUBLICATION
©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

partitioning process based on lipophilicity,
which presents issues for compounds with
drug-like properties80. Practical tricks,
such as dry loading of medium pressure,
readily scalable ‘flash’ systems and new
techniques such as hydrophilic interaction
liquid chromatography (HILIC)81 and
supercritical fluid methods82 can provide
improved performance in relevant
lipophilicity ranges. Supercritical fluid
chromatography with chiral columns
has made separating and accessing chiral
compounds more routine. However, the
current practice often involves testing
racemates first as a practical measure, and
then if activity is found, a chiral separation
follows. As the technology expands and
becomes even more routine, this strategy
will change and enable all compounds to be
separated into chiral stereoisomers in the
absence of initial biological data on
a racemate.
Novel reaction technologies
Improvements in reaction technologies
have the potential to enable more
effective exploration of novel compounds
in medicinal chemistry programmes.
Here, we focus in particular on rapidly
emerging new technologies that could
enable reactions that are likely to be
valuable in medicinal chemistry but are
under-represented in current practice,
such as biocatalytic methods, photoredox
chemistry, electrochemical methods
and C–H bond activation. All these
technologies have potential applications
in improving access to novel and diverse
reaction intermediates or in late-stage
functionalization strategies. The impact
of such small changes can be profound in
optimization programmes83,84.
Biocatalytic methods. Biocatalysts (and
particularly their directed evolution) hold
promise for the synthesis of structurally
diverse molecules under ambient and
environmentally benign conditions. The
concept of mutating specific enzymes
to do unnatural reactions more typical
of a traditional synthesis is becoming
more general and practical. For example,
engineered enzymes can perform
carbon–carbon (C–C) bond-forming
reactions of carbenes derived from diazo
compounds, carbon–nitrogen (C–N)
bond-forming reactions, carbon–silicon
(C–Si) bond-forming reactions and carbon–
boron (C–B) bond-forming reactions85,86,87
(TABLE 2, entries 1–4).
Enzymes can also be valuable in
late-stage functionalization, for example,
by cytochrome P450 (CYP) preparations85,
microsomes87 or microbial broths. These
have found application in the generation
of metabolite standards88 (FIG. 2a,b) but can
also deliver more hydrophilic compounds
(FIG. 2c) through directed hydroxylation89.
Given the impact that methylation90 and
fluorination91,92 can have on bioactivity,
enzymes that could enable such
transformations at a late stage would be
valuable, and some progress has been made
with tandem CYP-mediated hydroxylation–
fluorination93 (FIG. 2d).
Photoredox chemistry. Photoredox
chemistry is a relatively old field94 but
has been limited in practice owing to a
lack of specific catalysts that can be used
across a diversity of reaction conditions
and chemical functional groups. Notable
recent discoveries have identified
multiple catalysts, ligands and reaction
conditions that enable important chemical
transformations that previously proved
difficult (for example, often requiring
multiple steps and/or protecting group
manipulation).
The basis of photoredox chemistry is
the activation of a photoredox catalyst
by light, which can then potentiate single
electron transfer between two partner
molecules, causing them to react94. As one
specific example, photoredox chemistry has
been used to make C–N bonds95 (TABLE 2,
entry 5), a key transformation for medicinal
chemists13. One advantage of the photoredox
method for C–N bond formation is that
it is mechanistically different from the
Buchwald–Hartwig reaction, which is one
of the few new reaction methodologies
among the top 20 most used reactions in
the medicinal chemists’ toolbox identified
in the Brown and Boström analysis13. Thus,
the photoredox method can expand the
scope of this important transformation
by providing alternative procedures when

Table 1 | Examples of the impact of enabling technologies

Entry Example Technology Impact Refs

1 R1 Microwave Fast and versatile 63

NH reaction
10 mol% CuCl /10 mol% CuCl2
R1 NH2 + R2 + R3 CHO R
H2O, microwave, 110 °C (100 W), 3
25 min R2
21 examples, 41–89%

2 Y N
High New heterocycles 64
H 390°C, 100 bar temperature
N R
CO2Et THF N and pressure
R Y
N
O
Y = CO2Et, CN, COMe 9 examples, 77–96%

3 N2 R1 N2 R1
Flow New synthesis 65,66

technology
B(OH)2 B(OH)2
R1 H B(OH)2 R2 H
R2

4 I E Flash Chemoselective 68

O R
functionalization
O R 1. PhLi
8 examples, 71–98%
O 2. E
+
O

8 | ADVANCE ONLINE PUBLICATION www.nature.com/nrd

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

Table 2 | Examples of emerging synthetic technologies applied to fragments and lead-like molecules
Entry Example Technology Bond formation Refs
1 R1 Biocatalysis C–C 86
O Bacillus subtilis truncated globin
F F
R1 + Y25L T45A Q49A EtO2C
OEt
N2 F R1 = H F

2 O O
Biocatalysis C–Si 86

R1 Si Rma cyt c variant

OEt + H Si
EtO
N2
R1
R1 = Me

3 Biocatalysis C–B 85

_
N N
BH
3
- + BH2
O +
OEt
R1 Escherichia coli expressing R1
OEt
Rma cyt c O
N2
R1= Me

4 NHTs Biocatalysis C–N (also an 87

TsN3 example of C–H
Me
Me activation)
Cytochrome P411CHA
MeO Whole cells MeO

5 5 mol% NiCl2 ·glyme Photocatalysis C–N 95

Br NBoc
+ NBoc
N
HN 0.02 mol% photocatalyst
F3C DABCO, DMA, rt
Blue LED F3C

6 1 mol% photocatalyst Photocatalysis C–C 96

O
+ 10 mol% NiCl2 ·glyme
N
N CO2H 15 mol% dtbbpy, Cs2CO3
Boc O
Boc Br DMF, 23°C
Blue LEDs

7 NBoc Photocatalysis C–C 97

HO Br 1. Oxalyl chloride
+ 2. 5 mol% Ni catalyst
NBoc N
1 mol% photocatalyst N
F 5:1 dioxane:THP
F
DMSO, 70°C
8 OCF3 Photocatalysis C–N and C–C 96
Photoredox N OCF3
couplings O
Cl NH
N
N
N
Q203, anti-TB clinical candidate
NH2C

9 Ar Ar Photocatalysis C–C, enhanced 99

366 nm, sp3 character

Ph2CO, 2+2 H H

O N
n n = 1,2 O N n

Ph Ph
10 Electrochemistry C–N 101
10 mol% NiCl2 .glyme
+ CF3
10 mol% di-tBupy, rt CF3

LiBr (4 equiv), DMA (0.1 M)

N Br N
H H (+)RVC/(–)Ni
Undivided cell H
I = 4 mA, 4.5 h

NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 9

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

Table 2 (cont.) | Examples of emerging synthetic technologies applied to fragments and lead-like molecules
Entry Example Technology Bond formation Refs
11 N Electrochemistry C–N 104

O
S
O 1. Anodic oxidation N
+ N
1.0 M LiClO4, CH3CN
N
2. Piperidine

12 F Electrochemistry C–N 105

NH F RVC(+)-Pt(–), N
I = 5 mA
N
N
Et4PF6, MeOH, THF

13 O Electrochemistry C–O 106

Me
Anodic oxidation CO2H
O
N AcOH, NaOAc N
N
H
O OBn O OBn

14 Electrochemistry C–N (also an 107

example of C–H
BDD anode
OH activation)
HFIP
+ Me

OH Undivided cell
OH
OMe OH
OMe

BDD, boron-doped diamond electrode; DABCO, 1,4‑diazobicyclo[2.2.2]octane; I, current; LED, light-emitting diode; MW, microwave heating; Rma cyt
c, Rhodothermus marinus cytochrome c; rt, room temperature; RVC, reticulated vitreous carbon; TB, tuberculosis.

traditional methods fail and thereby provide
more opportunities to explore diverse areas
of chemical space.
Photoredox techniques have recently
been applied to a drug discovery platform
with microscale chemistry to produce
‘informer libraries’ across a broad range of
substrates (78% reaction success rate), thus
illustrating the potential robustness of this
reaction type38. The concept of a microscale
informer library is that one can arrange a
large matrix of reaction conditions on the
desired medicinal scaffold of interest on a
small scale (for example, <1 mg). Subsequent
analysis can then inform the scientist on
not only one set of conditions but a range
of conditions that can be immediately
employed to produce large libraries of target
compounds. This concept of informer
libraries is not only applicable to the
identification of optimal reaction conditions
but also highly relevant to the previous
discussion on automated platform synthesis.
Innovations in photoredox chemistry
have also recently been applied to
the generation of C–C bonds, where
N‑protected amino acids can react to replace
the carboxylic acid by a functionalized
aromatic ring (sp3–sp2 cross coupling), a
process that previously would have required
multistep synthetic procedures. MacMillan
and colleagues showed that protected amino
acids can be reductively decarboxylated at
the α‑carbon and that a new bond can be
formed to an aromatic ring in one step96
(TABLE 2, entry 6). Another variant of this
chemistry (TABLE 2, entry 7) utilizes alcohols
as substrates for photoredox catalysis instead
of carboxylic acids, thus expanding the scope
of this transformation97. This technology
has been applied to the efficient synthesis of
Q203, a drug candidate in clinical trials for
tuberculosis96 (TABLE 2, entry 8).
Another recent example of photoredox
catalysis from Suero and co-workers98
involved the generation of carbynes by
photocatalysis and their application to
create chiral centres from three different
fragments by assembly-point functionali-
zation. The usefulness of the methodology
was demonstrated by the chiral func-
tionalization of known drugs and also by
illustrating a procedure whereby a chiral
centre can be directly attached to an
aromatic ring in a one-step reaction. This
type of chemistry should, in principle,
allow medicinal chemists to access a
broader and more diverse part of chemical
space. The photochemical synthesis of
3‑azabicyclo[3.2.0]heptanes represents a
good example of new building blocks with
enhanced sp3 character 99 (TABLE 2, entry 9).
Finally, photoredox chemistry has also
been used to facilitate the late-stage func-
tionalization of heterocycles. For example,
the direct addition of simple alkyl groups
such as methyl, ethyl and cyclopropyl to
drugs (such as loratidine) and drug-like
scaffolds has been reported100. Such
late-stage functionalization is likely to allow
for more versatility in synthetic planning
and access to potential test molecules that
previously would have each required their
own de novo synthesis.
Electrochemical methods. The
implementation of electrochemical reaction
protocols is gaining broader utility in
organic chemistry and the production of
medicinal-chemistry-relevant scaffolds101,102.
The increased adoption of this technology
is largely due to improvements in practical
methods of electrochemical synthesis. For
example, previously, an ‘electroauxillary’
functional group, such as arylthio, α‑silyl
and organostannane, often needed to be
installed on one reaction partner to control
the electron transfer event with the other
reaction partner 103. However, these types
of groups are not common reagents to a
medicinal chemist, and building them
into the reaction partner is a disincentive
because it requires substantial scoping and
optimization for each target compound.
By contrast, the recent examples reported
for C–N bond synthesis shown in TABLE 2
(examples 10, 11 and 12) illustrate very
common building blocks that require no
special electroauxillary to be attached

10 | ADVANCE ONLINE PUBLICATION www.nature.com/nrd

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

to the reagent 104,105. C–O bonds can new ways to form C–C bonds using setting for reaction and library scoping
also be formed under electrochemical electrochemical protocols have also been but can also be scaled up for chemical
reaction conditions with applicability in reported107 (TABLE 2, example 14). process applications108. Advances in the
medicinal chemistry, such as the novel Electrochemical methods are adaptable commercial accessibility and ease of use of
γ‑aminobutyric acid uptake inhibitors106 to flow chemistry, which has applications electrochemical reactors hold promise for
shown in TABLE 2 (example 13). Finally, in automated synthesis in the research this to become a routine methodology 109.
Commercial equipment suppliers play
a HO
a key role in the widespread use of new
OH approaches, especially when the equipment
O
O OH HO2C can simplify the set‑up and execution of
H2N S N OH
N UGTIA9 O normally complex reactions; flow reactors,
H NH NH O
O O which can be used from discovery through to
H2N S N production, are a good example.
N N N
O Br H NH NH

F C–H bond activation and other novel

N N
Epacadostat O Br reaction technologies. There is currently
substantial investment in both academia
F
Gut All metabolites were also isolated from and industry to develop late-stage func-
microbiota Hypha biotransformation panel tionalization methods, spurred by major
O recent discoveries. Late-stage functional-
O HN
H2N S HN ization through targeting specific reactive
N H2N NH
H NH NH CYP3A4/2C19/1A2 C–H bonds for substitution reactions is
N N
particularly desirable in medicinal chemistry,
Br
N N
Br
O as it can eliminate the requirements for
O
F
costly and, at times, cumbersome functional
F group manipulation early in the synthesis
b scheme. An excellent example of this type
Kd 53 nM OH
Kd 2.6 nM
LogD7.4 2.6 of limitation is the popular Suzuki–Miyaura
LogD7.4 3.9
LLE 3.4 LLE 6.0 reaction illustrated in FIG. 1, which typically
N Hypha microbial panel N
relies on boronic acids and halogenated
N N
coupling partners, each with the designed
R
NH
R
NH substitution patterns.
Cernak and colleagues comprehensively
c N N reviewed110 and classified late-stage func-
N Liver microsomes N
tionalization strategies, which can be broadly
N N divided into innate reactions (that is, those
N N not requiring a directing group111 (FIG. 3a))
N CF3 N CF3
HN HN and guided reactions (for example, using
O PDE2 IC50 0.6 nM O OH
PDE2 IC50 0.4 nM directing groups, steric effects or molecular
LipMetE 1.9 LipMetE 2.4 recognition95 (FIG. 3b)). Innate reactions such
LipE 5.9 LipE 7.4
as those illustrated in FIG. 3a can install groups
d O such as CF3 in the last step of the synthesis
O O of highly functionalized molecules95,111. Such
O N
O N H O N techniques can be leveraged in medicinal
H H
HO chemistry to potentially improve potency
CYP2C9 DAST F
but also potentially block reactive/metabolic
Ramelteon 2-(S)-hydroxyramelteon 2-(R)-fluororamelteon sites in these compounds without having to
design new synthetic schemes from scratch112.
Figure 2 | Examples of bioprocessing and natural product chemistry applied to bioactive or drug A compelling example of the directing group
molecules. a | Microbial biotransformations can be used to easily generate metabolic markers, such technology is shown in FIG. 3b (REF.113), in
as those from epacadostat, in good yield, avoiding often complex or capricious
Nature conventional
Reviews synthe-
| Drug Discovery which a celecoxib analogue can be readily
sis88. b | Microbial biotransformation can be used to generate useful structure–activity relationships converted to a diverse array of substituted
and property improvements from a lipophilic lead molecule; the kinase inhibitor shown that was pro- analogues, in this case driven by the
duced by using Hypha’s proprietary technology had reduced lipophilicity (logD7.4) and improved activ- coordination of a Pd catalyst to the pendant
ity and lipophilic efficiency88. c | Late-stage functionalization using liver microsomes can give similar sulfonamide and resultant insertion to a
outcomes to microbial oxidations in facilitating hydroxylations at sites difficult to access syntheti-
proximal ortho-C–H bond. The large-scale
cally89. d | Late-stage functionalization of the drug ramelteon using a combination of cytochrome
preparation and deoxyfluorination enables hydroxylation at particular sites and the introduction of synthesis of clinical candidates is also an area
fluorine, which can modulate properties and add conformational restraint93. CYP, cytochrome P450; of impact, as shown in FIG. 3c (REF.114).
DAST, diethylaminosulfur trifluoride; IC50, half-maximal inhibitory concentration; Kd, dissociation con- The practical uptake of C–H activation
stant; LipE, lipophilic efficiency; LipMetE, lipophilic metabolism efficiency; LLE, ligand lipophilicity chemistry by medicinal chemists will be
efficiency; PDE2, phosphodiesterase 2; UGT1A9, UDP-glucuronosyltransferase 1–9. enhanced when computational models

NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 11

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

a
Pd(OAc)2 (10 mol%)
O O Cl H Cl CF3
AgOAc (3 equiv)
H HFIP or CHCl3 CO2Et NaSO2CF3 (3 equiv)
CO2Et N N
+
3,5-bis(trifluoromethyl)- N N t
BuOOH (5 equiv) N N
O O H H
pyridin-2-ol (30 mol%) DCM:H2O (2.5:1)

b O O O O
C6F5 S C6F5 S
N N
H Pd(OAc)2-catalysed methods H
N N
H N R N
CF3 CF3
R = CO2H, Me,
Ph, I, others

MeO MeO

c N
N N N N H
X 1. BnBr Enantioselective
hydrogenation and
N N
elaboration to final
N 2. Pd(dppf)Cl2, Bn candidate
X = I or Br DBU, DMF O
11-β-HSD1 clinical candidate
(multikilogram synthesis)

d
Ph i
Pr O O O Ph i
Pr O O
NH NH
NiCl-catalysed methods R
N OH N
O O

From atorvastatin R = N, O and C bond formation

F F

e f Ts
Ts N I N
O Me
Me Fe(acac)3 (20 mol%) O
+ N
20% NiBr2·diglyme
PhSiH3 (1.5 equiv)
30% 4,4ʹ-di-tert-butyl-2,2ʹ-bipyridine
DCE:(CH2OH)2 (5:1) OTMS
2 equiv DEMS
60°C N N
2 equiv Na2CO3
OTMS
N

g h
H 1. iPrMgCl Br OMe
OMe Suzuki–Miyaura
N N
2. hydrogenation
Cl Cl HO
N B 55 examples, enhanced
Ts sp3 character
Cl Cl OH
Stock solution
Sertraline in Et2O ‘Propellerized’ sertraline N

Ts

Figure 3 | Examples of emerging synthetic methodologies applied to chemistry114. d | Late-stage functionalization provides
Nature novel
Reviews analogues
| Drug and
Discovery
fragments and lead-like scaffolds. a | Illustrations of innate carbon– diversification of approved drug scaffolds115. e | Reductive olefin coupling
hydrogen (C–H) bond activation, where drug-like scaffolds can be substi- as a new and diverse method to make carbon–carbon (C–C) bonds116.
tuted with various functional groups in the last step of a synthetic f | Reductive olefin hydrocarbonation of known drug scaffolds leads to
sequence95,111,113. b | An example of guided C–H bond activation where a novel and diverse analogues117. g | Strain-releasing agents react with
functional group is attached by direction of a neighbouring group113. nucleophiles under mild conditions to introduce complex ring systems118.
c | C–H arylation as used in the synthesis of a clinical candidate, thus h | Cross coupling/hydrogenation to impart enhanced sp3 character com-
demonstrating the suitability of this large-scale and high-yielding pared with typical Suzuki–Miyaura reactions119,120.

12 | ADVANCE ONLINE PUBLICATION www.nature.com/nrd

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

can generate enhanced regiochemical
predictions. In the conventional
methodology, a chemist installs a functional
group knowing precisely where the
chemistry will occur, but with C–H bond
activation, the site of reactivity may be
less certain. Cernak and colleagues110 also
described high-throughput experimental
methods to help identify suitable late-stage
functionalization strategies and outlined
decision-making strategies for determining
the likelihood of success.
Beyond C–H activation, a recent
reaction development relevant to late-stage
functionalization — and also with
strong potential to expand the medicinal
chemistry toolbox in general — allows
the direct conversion of carboxylic acids
to boronic acids. Baran and colleagues
demonstrated that this novel reaction
can be used to perform late-stage func-
tionalization of approved drugs such as
atorvastatin (FIG. 3d) and vancomycin and
to enable improvements in synthesizing
drugs (for example, bortezomib), as well as
in the synthesis of a library of compounds
that was used to identify potent inhibitors
of human neutrophil elastase115.
Furthermore, access to boron-substituted
compounds is important for the widely
used Suzuki–Miyaura reaction mentioned
previously.
Another example of potentially trans-
formational chemistry is the iron-catalysed
radical reaction discovered by Baran and
colleagues116 to form sp3–sp3 C–C bonds by
reductive coupling of two olefins (FIG. 3e).
In this procedure, C–C bonds can be
formed from relatively common reagents
and mild conditions to provide complex
scaffolds. A related new reaction, reductive
olefin hydrocarbonation using Ni catalysis,
also has strong potential to affect drug
design because it can establish sp3–sp3 C–C
bonds from halogenated alkyl groups and
a corresponding alkene117 (FIG. 3f). Such
reactions are interesting for medicinal
chemists, given the ability to construct
novel and diverse molecules via C–C bond
formation using common building blocks
and mild conditions.
As mentioned previously, the uptake of
new reactions that can be used to install
atypical groups or create novel scaffolds
in drug discovery can be hindered by
the concern of the potentially excessive
time and resources spent on optimizing a
procedure for a particular compound. One
major advancement that can help address
this issue has also come from Baran and
colleagues, who introduced the concept
of strain-released heteroatom functional-
ization118 (FIG. 3 g). Strain-releasing agents
can be thought of as high-energy complex
species that are spring-loaded to react with
amines under controlled reaction conditions
to install small, complex ring systems. In
essence, the complexity of the synthesis has
already been addressed in the strain-release
agents, which are becoming commercially
available (for example, FIG. 3 g).
Finally, in addition to entirely new
reactions, innovations can also improve the
most frequently used reactions. For example,
even for the widely adopted Suzuki–Miyaura
cross-coupling reactions, the recent report of
the formation of highly desirable quaternary
C–C bonds119,120 could substantially expand
their impact (FIG. 3 h).
Although the impact remains to be
seen for these highlighted emerging
methodologies, we believe that they will
enable access to more sp3-like chemical
space, which can address perceived
limitations of polyaromatic compounds121
and provide other potential advantages
(see discussion below). Furthermore,
companies that adopt these newer
methodologies could gain the advantage
of structural differentiation of their
drug candidates from those of the many
companies using established chemistry
and commercial building blocks,
and this differentiation can provide
important intellectual property benefits in
crowded areas, such as kinase targets.
Synthetic strategies and platforms
Finally, we briefly highlight some broad
synthetic strategies and platforms developed
over the past two decades that are already
valuable tools for medicinal chemists and
chemical biologists and may also have
considerable further potential. The breadth
of these strategies is far too wide to allow
anything more than a brief overview of
them in a single article, but they are
important to note in the context of the
discussion here and so are highlighted by
providing at least one illustrative example
of each approach, with references to
comprehensive reviews on each.
Click chemistry. Click chemistry (FIG. 4a)
— a concept proposed by Sharpless
and colleagues in 2001 that involves a
high-yielding mild reaction — has been
extensively used to create probes and tools
that operate under bioorthogonal conditions
for chemoproteomic interrogations122,123.
Although medicinal chemists may not often
use click chemistry in standard compound
preparation13, they increasingly also need
to make chemogenomic tool compounds in
order to investigate target biology, and click
chemistry can be valuable here. As a recent
example, click chemistry has been used to
investigate the gene regulatory function
of bromodomain inhibitors by studying
the localization of ‘click-able’ inhibitors
both within the cell and within tissue
compartments124.
Multicomponent reactions.
Multicomponent reactions (for example,
Ugi reactions (FIG. 4b)) can generate many
thousands of compounds through a single
reaction scheme125,126,127. Recently, Zarganes-
Tzitzikas and Dömling 128 reviewed the
application of this technology as applied
to the synthesis of telaprevir, a complex
molecule for which the number of synthetic
steps could be halved by using two
multicomponent reactions. Although these
techniques have been around for many
years, with the potential increased use of
automated synthesis platforms as well as the
use of DNA-encoded libraries (discussed
below), multicomponent reactions will
probably gain in utility.
DNA-encoded libraries. DNA-encoded
libraries enable the rapid affinity-based
screening of billions to trillions of
compounds, based on the construction
of split-and-mix combinatorial libraries
tagged with specific DNA sequences
that allow identification of hits by
PCR sequencing 129. These libraries
pose chemistry challenges, given the
requirement for conditions that ensure the
stability and integrity of the encoding DNA
element of the molecule. Thus, near neutral
aqueous environments are necessary, for
which an ever-growing toolbox of reactions
is being exploited130,131. Surprisingly, given
the aqueous environment and that a DNA
polymerase is used to encode the library,
enzyme catalysis has been exemplified
synthetically only in the construction of
on‑DNA glycans132.
Several success stories with this
technology have recently been published133,
exemplified in BOX 3 by the discovery
of the potent RIPK1 clinical candidate
by GlaxoSmithKline (GSK)134,135, interesting
hits for tuberculosis target InhA136 and
potential covalent kinase inhibitors137. As the
methodology matures, further applications
are emerging, such as selections against
multiple targets in parallel as a means
of both identifying hits and assessing
ligandability 138.

NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 13

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

a b
OBn HO CHO CO2H
Click HO O N N OH
MeOH O
N +
OBn HO N
OH OH NH2 EtO2C N
H
OBn EtO2C NC O
OH

c O OH
[Si]O
NHNs O
O MeO
Natural-product-like library, H
HO by macrocyclization N N
O N O

Ar O OMe
O N O
OAc H

NsHN O
Mycobacterium bovis MIC 11 μM

d Androgen receptor agonists

O
H O
Ar
N Ar O
N O
H
OH EC50 860 nM
EC50 7.3 nM O

O F3C N O
H
Ar
N O O
NC N2
H [Rh2(esp)2]
EC50 4.7 nM (1 mol%), EtOAc

O O
O
H
Ar Ar O
N NC N
H N

EC50 3.8 nM EC50 730 nM

e
Li Li Me Me
O O
H Me N N
TIBO TIBO Li Cl Li Li
Me H
Cl H
Si H Si Cl
[Si] [Si]
O Iterative stereocontrolled homologation R1 Bpin
Br
R1 O

O
N
N OH OH OH
1. Bpin R2
2. Oxidation
S HO2C
(+)-Kalkitoxin Post-FGI O
H H
OH

C15H31 OH OH
HO2C
R1 R2
OH
(+)-Hydroxyphthioceranic acid (+)-Zincophorin

f OMe
Ir/chiral phosphine OMe O
Chiral amine OMe
R,R + other 3 variants H
O H
C5H11 OMe H
H C5H11 OMe
+ C5H11 O

Stereodivergent use of (–) Δ9-trans-tetrahydrocannabinol

enantiomeric pairs of catalysts (and 3 isomers from chiral intermediates)

14 | ADVANCE ONLINE PUBLICATION

Nature Reviews | Drug Discovery
www.nature.com/nrd
©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

◀ Figure 4 | Broad synthetic strategies and platforms applied to lead and drug-like scaffolds. bioactive compounds emerge from rounds
a | Click chemistry has become a routine reaction, as exemplified by the synthesis of aldose reductase of synthesis, deliberately chosen to have
inhibitors123. b | Multicomponent reactions can generate diverse scaffolds in large numbers under mild multiple potential outcomes, as exemplified
conditions126. c | Natural-product-inspired macrocycle synthesis leads to compounds with anti-­ by the identification of androgen receptor
tuberculosis activity164. d | Activity-directed synthesis results in the identification of androgen receptor
agonists by synthetic expansion from a
agonists166. e | Generic scheme outlining the iterative ‘machine’ synthesis of polyketide linkages pio-
neered by the Aggarwal group, showing the building blocks and examples of synthetic targets known fragment 166 (FIG. 4d).
accessed pursuing the methodology170. f | Stereodivergent total synthesis of all four trans/cis diaster- Approaches inspired by natural
eomers of Δ9-tetrahydrocannabinol from common reagents and catalysts171. EC50, effective concen- product biosynthesis, such as automated
tration for half-maximum response; FGI, functional-group interconversion; MIC, minimal inhibitory assembly-line synthesis167 of polypropion-
concentration. ate-containing molecules168 or biomimetic
cascades169 (FIG. 4e), also have exciting
potential170. For example, this approach
Fragment-based drug discovery. Various active site or engineered cysteine residues is illustrated by recent work by Carriera’s
fragment-based strategies, which were have been used to aid in initial fragment group171 to make all four diastereomers
pioneered more than two decades ago, tethering 148. More recently, however, Cravatt of Δ9-tetrahydrocannabinols from
are now well established and are often and colleagues equipped fragments with common reagents and catalysts (FIG. 4f).
particularly attractive for challenging targets. photoactivatable moieties (for example, Complementary to wholly synthetic
The first drugs identified by fragment-based diazerene-to-carbene transformation) and strategies, opportunities are now arising for
strategies — vemurafenib139 and venetoclax 140 established a more generic chemical biology harnessing ‘directed evolution’. The notion of
— have been approved, erdafitinib has approach to target discovery 149 based on exploiting natural sources as feedstocks for
breakthrough designation141, and other covalent fragments for isolated proteins and chemical manipulation of drug molecules is
compounds are progressing in the clinic18. even for whole-cell screening 150. attractive as a greener and sustainable source
Such strategies screen libraries of fragments of complex molecules. The idea of targeted
— compounds with roughly 10–17 heavy Natural products as inspiration. Given evolution of source plants or microbial
atoms (~130–250 Da) — using various the long and successful history of natural agents as factories for drug production could
biophysical techniques (particularly X‑ray products in drug discovery, various be a viable future strategy, and opportunities
crystallography and NMR spectroscopy) groups have proposed that populating with directed evolution strategies have
to detect weak interactions with biological screening libraries with molecules that recently been reviewed172.
targets. Hits are then expanded and optimized have natural-product-like characteristics
preferably using structural knowledge of their (such as higher content of sp3-hydridized Outlook
interaction with the target. carbons151,152) could be valuable, particularly Discovering new drugs is a formidable
Here, we briefly highlight some points for novel targets153. Various strategies, such challenge. Biology is amazingly complex,
related to the synthesis of the types of as diversity-oriented synthesis154, have been and ultimately, it is the target biology
compounds included in fragment libraries, developed to achieve these goals by using that will largely determine whether a
which require their own particular reaction pathways designed to efficiently drug discovery project succeeds or fails.
characteristics to improve the likelihood synthesize libraries of such compounds155. Nevertheless, synthetic chemistry is often
of finding hits. The ideal fragment Progress in the development of new a rate-limiting step in the process of
should be soluble (given necessarily synthetic methodologies and applications identifying molecules that have the potential
high concentrations of stock solutions towards molecules with lead-like and/or to become safe and efficacious drugs.
for biophysical screening of their likely less planar character has been achieved Importantly, this can particularly be the
low-affinity binding) and relatively through the funding of European academic case when the biological validation for the
uncomplex 142,143. A case has been made groups in partnership with industry in target is compelling but it falls into a class
for increased three‑­dimensionality in the establishment of a European Lead of targets that have proved challenging to
fragments144 and also for populating Factory 156. This consortium, with close modulate, such as KRAS or c‑Myc in cancer.
screening collections with the motifs of consultation with industry, has focused on Although some novel targets may eventually
fragmented natural products145. However, molecular shape and property guidelines157 prove routinely tractable with platforms
analysis of Astex data indicated that and has been proactive in questioning such as antisense technologies173 or protein
less complex and flatter fragments have the status quo in terms of contemporary degradation technologies174, expanding the
higher hit rates, although effective growth practice in academic synthesis and drug medicinal chemistry toolbox as reviewed
often requires more functionalized discovery 158,159. As their work has progressed, here is also an important strategy.
and/or asymmetric chemistry to yield the consortium has charted progress in When considering how to achieve this, an
candidate-quality structures146. The achieving chemical space expansion160,161, important factor is that the role of medicinal
structure-based design inherent to realizing the goals of lead-oriented chemistry has changed considerably since
fragment-based drug discovery leads synthesis162 and assessing potential its inception. There are many examples from
to optimal binding and ligand-efficient applicability to central nervous system the past when key biological discoveries
molecules, but discipline with properties147 targets163. Interesting screening data are and therapeutic advances were made
is still necessary to ensure the best emerging, such as natural-product-inspired through synthetic chemists exploring
developability characteristics25. macrocycles with anti-tuberculosis activity 164 various chemical routes and SARs in
Another interesting development involves (FIG. 4c). An intriguing concept developed drug classes such as benzodiazepines175,
covalently binding fragments. Electrophilic by the Nelson group in this consortium penicillins, tetracyclines176 and taxanes177,178.
fragments that react with an enzyme is activity-directed synthesis165, whereby Unfortunately, synthetic chemistry is now

NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 15

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

often considered a bottleneck in DMTA entities (SMEs) and big pharma together
cycles, which can put considerable pressure with academic researchers — as exemplified
on medicinal chemists to deliver molecules by the European Innovative Medicines
in a specific time frame for the planned Initiative (IMI) and the Joint European
screening cascades. Reward systems and Compound Library (JECL)180 at the
pressure on delivery have pushed medicinal European Lead Factory — may be one
chemists to meet metric-based goals within a way to help address this barrier. Initiatives
progression-seeking R&D culture179. In these aiming to synthesize novel building
circumstances, a numbers-game attitude and blocks and develop relevant synthetic
a focus on easy‑to‑make compounds can transformations are also important 49,50
often divert medicinal chemists from design and provide an excellent opportunity for
principles, and the potential impact of extending industry–academia partnerships29.
more complex and/or challenging synthetic Collaborations and incubators could also
strategies is likely to be lost. In addition, help smaller biotech firms and university
the way CROs have been used for synthetic start-ups capitalize on their novel chemistry
chemistry has affected what has been made, in drug discovery settings.
undeniably tilting towards robust chemistry. Focusing on the technological issues,
With this in mind, we see two academic groups have reported notable
complementary approaches to enabling achievements in automated syntheses62,74.
medicinal chemists to make fuller use of the Given the diversity of reaction types
tools discussed above. The first approach is and conditions used in drug discovery
to reduce some of the technical barriers to programmes, however, optimization of all
using such tools. Precompetitive consortia the useful reactions may not be possible. An
consisting of small and medium-sized alternative approach is to limit these efforts
Box 3 | Example of the impact of new synthesis strategies: DNA-encoded libraries
Rapid affinity-based screening of billions to trillions of compounds can be achieved with
DNA-encoded libraries, for which hits can be identified by PCR of specific DNA sequences that are
linked to each library member. The figure shows an overview of the process from a DNA-encoded
library134 to GSK772, a clinical candidate that targets receptor-interacting protein kinase 1 (RIPK1)135.
Affinity screening of a ~7.7 billion-membered three‑cycle capped amino acid library, encoded by
DNA, identified a key feature through PCR of bound compounds, leading to the off-DNA synthesis of
the 1.3 nM hit compound. This was optimized to the clinical candidate with minimal but important
structural changes. Replacement of the isoxazole heterocycle improved the developability profile of
this series. The 3‑benzyl‑1,2,4‑triazole isomer showed the optimal combination of in vitro potency,
lipophilicity, kinetic aqueous solubility and rat oral exposure. GSK772 is currently in phase IIa clinical
studies for psoriasis, rheumatoid arthritis and ulcerative colitis.
pIC50, negative logarithmic value of the half-maximal inhibitory concentration.
R1 O R1 O
O
H
N
H
N
H
N N
N
H
R3 N
H NH
O R2 O
R3
O
O can be identified at both the management
Component of library, Selection, PCR and analysis
~7.7×109 compounds identified this structure
O strategies is needed. Projects could benefit
O
Off-DNA synthesis O from multiple values, such as a more
NH N
N chemotype, that require more time and
O
RIPK1 pIC50 1.3 nM
O N
O
Lead optimization NH
NH N
N
O new technologies. Industry–academia
GSK772
to the most versatile and robust reactions
(highlighted in BOX 1). Recent contributions
from the Burke group70 realized this concept
and provided a viable synthesis option for
more than ten compound families. Further
work in this direction could result in robust
reactions for automated synthesis that cover
a considerable percentage of the relevant
bioactive chemotypes. Recent advances
in testing the functional group tolerances
of reactions and the stability of specific
chemical motifs under reaction conditions181
— together with a larger set of available
building blocks — may help to extend the
scope of robust reactions.
The predictive power of computational
tools for both designing synthetic routes
and identifying promising molecules to
synthesize depends on the amount and
quality of underlying experimental data.
The future will bring more data with better
quality, more data-processing power and
hopefully better computational methods,
including those using deep-learning
technologies and artificial intelligence182,183.
This could provide greater confidence in the
value of synthetically challenging molecules
and help move away from ‘more shots on
goal’ approaches using the easiest chemistry
available. Furthermore, the technologies
discussed above may also enable the
synthesis of more lead-like molecules.
For example, biocatalytic methods
work in aqueous media and favour the
incorporation of hydrophilic functionalities,
and late-stage functionalization methods
can address metabolic ‘soft spots’ with the
introduction of F or CF3 groups116. Another
important feature of some of these emerging
technologies is the ability to introduce
higher sp3 content in the lead molecules.
The second approach is to address
the cultural barriers that have limited the
uptake of newer approaches. These barriers
and bench chemistry levels. From a
management perspective, replacement of
the numbers-game attitude with value-based
feasible synthesis, a new catalyst or a new
effort than raw synthesis schemes with
robust reactions. These values should be
acknowledged in improved reward schemes.
Establishing new types of partnerships
and collaborations with academia could
further facilitate the implementation of
relationships should move beyond CRO-type
contributions that are not matched with the

Nature Reviews | Drug Discovery

16 | ADVANCE ONLINE PUBLICATION www.nature.com/nrd
©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.

academic goals of discovering and publishing
new science. First, they should help academic
scientists to understand the true challenges
and needs of medicinal chemistry. Second,
the new chemistries and technologies that
medicinal chemists in industry have access to
could be further improved by incorporating
technology from academic researchers
and SMEs. Third, training schemes with
industry participation (such as GSK with the
University of Nottingham)184, new types of
PhD courses developed through initiatives
such as the Marie Sklodowska Curie Action
(MSCA) Innovative Training Networks,
internships and postdoctoral programmes
at industry partners may also improve the
background knowledge and creativity of the
next generation of medicinal chemists.
Finally, it is noteworthy that
small-molecule drugs currently constitute
a smaller proportion of the top echelon of
drug revenues by sales than ever before185.
This could be a catalyst for the practitioners
of small-molecule drug discovery to
better define the key problems that need
to be addressed and the areas in which
small-molecule drug discovery has the
greatest potential to provide major advances
compared with alternative platforms. We
remain optimistic about the future of
small-molecule drug discovery, as despite
the challenges, expanding the medicinal
chemistry toolbox with the arsenal of
methodologies highlighted above is set
to free up more time to be creative and
implement new chemistries, efficient
synthetic routes and improved reaction
technologies. Exciting times lie ahead.
Jonas Boström1, Dean G. Brown2, Robert J. Young3 and
György M. Keserü4*
Medicinal Chemistry, Cardiovascular, Renal and
1
Metabolism, IMED Biotech Unit, AstraZeneca,
Gothenburg, Sweden.
2
Hit Discovery, Discovery Sciences, IMED Biotech Unit,
AstraZeneca, Waltham, MA, USA.
3
NCE-Medicinal Chemistry, GlaxoSmithKline R&D,
Stevenage, Hertfordshire, UK.
Medicinal Chemistry Research Group, Research
4 overlooked strategy to improve compound quality.
Center for Natural Sciences, Hungarian Academy of
Sciences, Budapest, Hungary.
*e-mail: keseru.gyorgy@ttk.mta.hu
doi:10.1038/nrd.2018.116
Published online 24 Aug 2018
1. Bohacek, R. S., McMartin, C. & Guida, W. C. The art
and practice of structure-based drug design: a
molecular modeling perspective. Med. Res. Rev. 16,
3–50 (1996).
2. Mullard, A. The drug maker’s guide to the galaxy.
Nature 549, 445–447 (2017).
3. Ertl, P. J. Cheminformatics analysis of organic
substituents: identification of the most common
substituents, calculation of substituent properties &
automatic identification of drug-like bioisosteric groups.
J. Chem. Inf. Comput. Sci. 43, 374–380 (2003).
NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 17
4. Virshup, A. M., Contreras-García, J., Wipf, P., Yang, W.
& Beratan, D. N. Stochastic voyages into uncharted
chemical space produce a representative library of all
possible drug-like compounds. J. Am. Chem. Soc. 135,
7296–7303 (2013).
5. Bemis, G. W. & Murcko, M. A. The properties of known
drugs. 1. Molecular frameworks. J. Med. Chem. 39,
2887–2893 (1996).
6. Bemis, G. W. & Murcko, M. A. Properties of known
drugs. 2. Side chains. J. Med. Chem. 42, 5095–5099
(1999).
7. Wang, J. & Hou, T. Drug and drug candidate building
block analysis. J. Chem. Inf. Model. 50, 55–67 (2010).
8. Taylor, R. D., MacCoss, M. & Lawson, A. D. Rings in
drugs. J. Med. Chem. 57, 5845–5859 (2014).
9. Taylor, R. D., MacCoss, M. & Lawson, A. D. Combining
molecular scaffolds from FDA approved drugs:
application to drug discovery. J. Med. Chem. 60,
1638–1647 (2017).
10. Pitt, W. R., Parry, D. M., Perry, B. G. & Groom, C. R.
Heteroaromatic rings of the future. J. Med. Chem. 52,
2952–2963 (2009).
 12548–12549 (2016).
11. Visini, R., Arús-Pous, J., Awale, M. & Reymond, J. L.
Virtual exploration of the ring systems chemical
universe. J. Chem. Inf. Model. 57, 2707–2718 (2017).
12. Roughley, S. D. & Jordan, A. M. The medicinal
chemist’s toolbox: an analysis of reactions used in the
pursuit of drug candidates. J. Med. Chem. 54,
3451–3479 (2011).
13. Brown, D. G. & Boström, J. Analysis of past and
present synthetic methodologies on medicinal
chemistry: where have all the new reactions gone?
J. Med. Chem. 59, 4443–4458 (2016).
14. Schneider, N., Lowe, D. M., Sayle, R. A., Tarselli, M. A.
& Landrum, G. A. Big data from pharmaceutical
patents: a computational analysis of medicinal
chemists’ bread and butter. J. Med. Chem. 59,
4385–4402 (2016).
15. Carey, J. S., Laffan, D., Thomson, C. & Williams, M. T.
Analysis of the reactions used for the preparation of
drug candidate molecules. Org. Biomol. Chem. 4,
2337–2347 (2006).
16. Dugger, R. W., Ragan, J. A. & Brown Ripin, D. H.
Survey of GMP bulk reactions run in a research facility
between 1985 and 2002. Org. Proc. Res. Dev. 9,
253–258 (2005).
17. Satyanarayanajois, S. D. & Hill, R. A. Medicinal
chemistry for 2020. Future Med. Chem. 14,
1765–1786 (2011).
18. Erlanson, D. A., Fesik, S. W., Hubbard, R. E.,
Jahnke, W. & Jhoti, H. Twenty years on: the impact of
fragments on drug discovery. Nat. Rev. Drug Discov.
15, 605–619 (2016).
19. Boström, J., Grant, J. A., Fjellström, O., Thelin, A. &
Gustafsson, D. Potent fibrinolysis inhibitor discovered
by shape and electrostatic complementarity to the
drug tranexamic acid. J. Med. Chem. 56, 3273–3280
(2013).
20. Lionta, E., Spyrou, G., Vassilatis, D. K. & Cournia, Z.
Structure-based virtual screening for drug discovery:
principles, applications and recent advances. Curr. Top.
Med. Chem. 14, 1923–1938 (2014).
21. Santos, R. et al. A comprehensive map of molecular
drug targets. Nat. Rev. Drug Discov. 16, 19–34
(2017).
22. Nadin, A., Hattotuwagama, C. & Churcher, I. Lead-
oriented synthesis: a new opportunity for synthetic
chemistry. Angew. Chem. Int. Ed. Engl. 51, 1114–1122
(2012).
23. Goldberg, F. W., Kettle, J. G., Kogej, T., Perry, M. W. &
Tomkinson, N. P. Designing novel building blocks is an
Drug Discov. Today 20, 11–17 (2015).
24. Keserü, G. M. & Makara, G. M. The influence of lead
discovery strategies on the properties of drug
candidates. Nat. Rev. Drug Discov. 8, 203–212
(2009).
25. Young, R. J. & Leeson, P. D. Mapping the efficiency and
physicochemical trajectories of successful
optimizations. J. Med. Chem. https://doi.org/10.1021/
acs.jmedchem.8b00180 (2018).
26. Keserü, G. M., Soós, T. & Kappe, C. O. Anthropogenic
reaction parameters — the missing link between
chemical intuition and the available chemical space.
Chem. Soc. Rev. 43, 5387–5399 (2014).
27. Boström, J. & Brown, D. G. Stuck in a rut with old
chemistry. Drug Discov. Today 21, 701–703 (2016).
28. Walters, W. P., Green, J., Weiss, J. R. & Murcko, M. A.
What do medicinal chemists actually make? A
50‑year retrospective. J. Med. Chem. 54,
6405–6416 (2011).
©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES
29. Blakemore, D. C. et al. Organic synthesis provides
opportunities to transform drug discovery. Nat. Chem.
10, 383–394 (2018).
30. Hann, M. M. & Keserü, G. M. Finding the sweet spot:
the role of nature and nurture in medicinal chemistry.
Nat. Rev. Drug Discov. 11, 355–365 (2012).
31. Rafferty, M. F. No denying it: medicinal chemistry
training is in big trouble. J. Med. Chem. 59,
10859–10864 (2016).
32. Campbell, I. B., Macdonald, S. J. F. & Procopiou, P. A.
Medicinal chemistry in drug discovery in big pharma:
past, present and future. Drug Discov. Today 23,
219–234 (2018).
33. Hartenfeller, M. et al. A collection of robust organic
synthesis reactions for in silico molecule design.
J. Chem. Inf. Model. 51, 3093–3098 (2011).
34. Segler, M. H. S., Preuss, M. & Waller, M. P. Planning
chemical syntheses with deep neural networks and
symbolic AI. Nature 555, 604–610 (2018).
35. Bergman, R. G. & Danheiser, R. L. Reproducibility in
chemical research. Angew. Chem. Int. Ed. Engl. 55,
36. Engkvist, O. et al. Computational prediction of
chemical reactions: current status and outlook.
Drug Discov. Today 23, 1203–1218 (2018).
37. Rahman, S. A. et al. Reaction decoder tool (RDT):
extracting features from chemical reactions.
Bioinformatics 32, 2065–2066 (2016).
38. Buitrago Santanilla, A. et al. Organic chemistry.
Nanomole-scale high-throughput chemistry for the
synthesis of complex molecules. Science 347, 49–53
(2015).
39. Perera, D. et al. A platform for automated nanomole-
scale reaction screening and micromole-scale
synthesis in flow. Science 359, 429–434 (2018).
40. Szymkuc´, S. et al. Computer-assisted synthetic
planning: The end of the beginning. Angew. Chem. Int.
Ed. Engl. 55, 5904–5937 (2016).
41. Segler, M. H. S. & Waller, M. P. Modelling chemical
reasoning to predict and invent reactions. Chem. Eur.
J. 23, 6118–6128 (2017).
42. Kayala, M. A. & Baldi, P. ReactionPredictor: prediction
of complex chemical reactions at the mechanistic level
using machine learning. J. Chem. Inf. Model. 52,
2526–2540 (2012).
43. Kayala, M. A., Azencott, C.‑A., Chen, J. H. & Baldi, P.
Learning to predict chemical reactions. J. Chem. Inf.
Model. 51, 2209–2222 (2011).
44. Stark, S. A., Neudert, R. & Threlfall, R. Wiley
ChemPlanner predicts experimentally verified
synthesis routes in medicinal chemistry. CHEManager
http://www.chemanager-online.com/en/whitepaper/
wiley-chemplanner-predicts-experimentally-verified-
synthesis-routes-medicinal-chemistry (2016).
45. Bøgevig, A. et al. Route design in the 21st century:
the ICSYNTH Software tool as an idea generator for
synthesis prediction. Org. Proc. Res. Dev. 19,
357–368 (2015).
46. Klucznik, T. et al. Efficient syntheses of diverse,
medicinally relevant targets planned by computer and
executed in the laboratory. Chem 4, 522–532 (2018).
47. Kroman, J. C. et al. Fast and accurate prediction of the
regioselectivity of electrophilic aromatic substitution
reactions. Chem. Sci. 9, 660–665 (2018).
48. Hansen, E. et al. Prediction of stereochemistry using
Q2MM. Acc. Chem. Res. 49, 996–1005 (2016).
49. Goldberg, F. W., Kettle, J. G., Kogej, T., Perry, M. W. &
Tomkinson, N. P. Designing novel building blocks is an
overlooked strategy to improve compound quality.
Drug Discov. Today 20, 11–17 (2015).
50. Carreira, E. M. & Fessard, T. C. Four-membered ring-
containing spirocycles: synthetic strategies and
opportunities. Chem. Rev. 114, 8257–8322 (2014).
51. Helal, C. J. et al. Increased building block access
through collaboration. Drug Discov. Today https://doi.
org/10.1016/j.drudis.2018.03.001 (2018).
52. Murray, P. M. et al. The application of design of
experiments (DoE) reaction optimisation and solvent
selection in the development of new synthetic
chemistry. Org. Biomol. Chem. 14, 2373–2384
(2016).
53. Carlson, R. & Carlson, J. E. in Design and
Optimization in Organic Synthesis Vol. 24 1–574
(Elsevier, 2005).
54. Cook, A. Computer-aided synthesis design: 40 years
on — WIREs. Comput. Mol. Sci. 2, 79–107 (2012).
55. Schwaller, P., Gaudin, T., Lanyi, D., Bekas, C. &
Laino, T. “Found in translation”: predicting outcomes
of complex organic chemistry reactions using neural
sequence-to‑sequence models. arXiv https://arxiv.org/
abs/1711.04810 (2017).
PERSPECTIVES
56. Baskin, I., Madzhidov, T. I., Antipin, I. S. &
Varnek, A. A. Artificial intelligence in synthetic
chemistry: achievements and prospects. Rus. Chem.
Rev. 86, 1127 (2017).
57. Wengong, J. Predicting organic reaction outcomes with
Weisfeiler-Lehman network. arXiv https://arxiv.org/
abs/1709.04555 (2017).
58. Fooshee, D. Deep learning for chemical reaction
prediction. Mol. Syst. Des. Eng. https://doi.
org/10.1039/C7ME00107J (2018).
59. Coley, C. W., Barzilay, R., Jaakkola, T. S., Green, W. H.
& Jensen, K. F. Prediction of organic reaction
outcomes using machine learning. ACS Cent. Sci 3,
434–443 (2017).
60. Brynjolfsson, E. & Mitchell, T. What can machine
learning do? Workforce implications: profound change
is coming, but roles for humans remain. Science 358,
1530–1534 (2017).
61. Griffen, E. J. et al. Can we accelerate medicinal
chemistry by augmenting the chemist with Big Data
and artificial intelligence? Drug Discov. Today 23,
1373–1384 (2018).
62. Fitzpatrick, D. E., Battilocchio, C. & Ley, S. V. Enabling
technologies for the future of chemical synthesis. ACS
Cent. Sci. 2, 131–138 (2016).
63. Trang, J., T. T. T., Ermolat’ev, D. S. & Van der
Eycken, E. V. Facile and diverse microwave-assisted
synthesis of secondary propargylamines in water using
CuCl/CuCl2. RSC Adv. 5, 28921–28924 (2015).
64. Tsoung, J. et al. Synthesis of fused pyrimidinone and
quinolone derivatives in an automated high-
temperature and high-pressure flow reactor. J. Org.
Chem. 82, 1073–1084 (2017).
65. Tran, D. N., Battilocchio, C., Lou, S.‑B., Hawkins, J. M.
& Ley, S. V. Flow chemistry as a discovery tool to
access sp2–sp3 cross-coupling reactions via diazo
compounds. Chem. Sci. 6, 1120–1125 (2015).
66. Battilocchio, C. et al. Iterative reactions of transient
boronic acids enable sequential C–C bond formation.
Nat. Chem. 8, 360–367 (2016).
67. Yoshida, J., Takahashi, Y. & Nagaki, A. Flash
chemistry: flow chemistry that cannot be done in
batch. Chem. Commun. 49, 9896–9904 (2013).
68. Kim, H. et al. Submillisecond organic synthesis:
outpacing Fries rearrangement through microfluidic
rapid mixing. Science 352, 691–694 (2016).
69. Schneider, G. Automating drug discovery. Nat. Rev.
Drug Discov. 17, 97–113 (2018).
70. Li, J. et al. Synthesis of many different types of organic
small molecules using one automated process. Science
347, 1221–1226 (2015).
71. Baranczak, A. et al. Integrated platform for expedited
synthesis-pPurification-testing of small molecule
libraries. ACS Med. Chem. Lett. 8, 461–465 (2017).
72. Godfrey, A. G., Masquelin, T. & Hemmerle, H. A
remote-controlled adaptive medchem lab: an
innovative approach to enable drug discovery in the
21st Century. Drug Disc. Today 18, 795–802
(2013).
73. Reizman, B. J., Wang, Y.‑M., Buchwald, S. L. &
Jensen, K. F. Suzuki-Miyaura cross-coupling
optimization enabled by automated feedback. React.
Chem. Eng. 1, 658–666 (2016).
74. Trobe, M. & Burke, M. D. The molecular industrial
revolution: automated synthesis of small molecules.
Ang. Chem. Int. Ed. 57, 4192–4214 (2018).
75. Troshin, K. & Hartwig, J. F. Snap deconvolution: an
informatics approach to high-throughput discovery of
catalytic reactions. Science 357, 175–181 (2017).
76. Chakravorty, S. J. et al. Nuisance compounds, PAINS
filters, and dark chemical matter in the GSK HTS
collection. SLAS Discov. 23, 532–545 (2018).
77. Stalcup, A. M. Chiral separations. Annu. Rev. Anal.
Chem. 3, 341–363 (2010).
78. Desai, B. et al. Rapid discovery of a novel series of Abl
kinase inhibitors by application of an integrated
microfluidic synthesis and screening platform. J. Med.
Chem. 56, 3033–3047 (2013).
79. Czechtizky, W. et al. Integrated synthesis and testing
of substituted xanthine based DPP4 inhibitors:
application to drug discovery. ACS Med. Chem. Lett. 4,
768–772 (2013).
80. Hobbs, A. N. & Young, R. J. Practical purification of
hydrophilic fragments and lead/drug-like molecules by
reverse phase flash chromatography: tips, tricks and
contemporary developments. Drug Discov. Today 18,
148–154 (2013).
81. Buszewski, B. & Noga, S. Hydrophilic interaction liquid
chromatography (HILIC)—a powerful separation
technique. Anal. Bioanal. Chem. 402, 231–247
(2012).
18 | ADVANCE ONLINE PUBLICATION www.nature.com/nrd
82. Hettiarachchi, K., Kong, M., Yun, A., Jacobsen, J. R. &
Xue, Q. Development of an automated dual-mode
supercritical fluid chromatography & reversed-phase
liquid chromatography mass-directed purification
system for small-molecule drug discovery. J. Sep. Sci.
37, 775–781 (2014).
83. Tarcsay, A., Nyíri, K., Keserü, G. M. Impact of lipophilic
efficiency on compound quality. J. Med. Chem. 55,
1252–1260 (2012).
84. Johnson, T. W., Gallego, R. A., Edwards, M. P.
Lipophilic efficiency as an important metric in drug
design. J. Med. Chem. https://doi.org/10.1021/acs.
jmedchem.8b00077 (2018).
85. Kan, S. B. J., Huang, X., Gumulya, Y., Chen, K. &
Arnold, F. H. Genetically programmed chiral
organoborane synthesis. Nature 552, 132–136 (2017).
86. Arnold, F. H. Directed evolution: bringing new
chemistry to life. Angew. Chem. Int. Ed. Engl. 57,
4143–4148 (2018).
87. Prier, C. K., Zhang, R. K., Buller, A. R., Brinkmann-
Chen, S. & Arnold, F. H. Enantioselective,
intermolecular benzylic C‑H amination catalysed by an
engineered iron-haem enzyme. Nat. Chem. 9, 629–634
(2017).
88. Boer, J. et al. Roles of UGT, P450, and gut microbiota
in the metabolism of epacadostat in humans. Drug
Metab. Dispos. 44, 1668–1674 (2016).
89. Obach, R. S. et al. Lead diversification at the
nanomole scale using liver microsomes and
quantitative nuclear magnetic resonance
spectroscopy: application to phosphodiesterase 2
inhibitors. J. Med. Chem. 61, 3626–3640 (2018).
90. Schonherr, H. & Cernak, T. Profound methyl effects
in drug discovery and a call for new C‑H
methylation reactions. Angew. Chem. Int. Ed. Engl.
52, 12256–12267 (2013).
91. Gillis, E. P., Eastman, K. J., Hill, M. D., Donnelly, D. J.
& Meanwell, N. A. Applications of fluorine in medicinal
chemistry. J. Med. Chem. 58, 8315–8359 (2015).
92. Pettersson, M. et al. Quantitative assessment of the
impact of fluorine substitution on P‑glycoprotein (P‑gp)
mediated efflux, permeability, lipophilicity, and
metabolic stability. J. Med. Chem. 59, 5284–5296
(2016).
93. Obach, R. S., Walker, G. S. & Brodney, M. A.
Biosynthesis of fluorinated analogs of drugs using
human cytochrome P450 enzymes followed by
deoxyfluorination and quantitative nuclear magnetic
resonance spectroscopy to improve metabolic stability.
Drug Metab. Dispos. 44, 634–646 (2016).
94. Romero, N., A. & Nicewicz, D. A. Organic photoredox
catalysis. Chem. Rev. 116, 10075–10166 (2016).
95. Ji, Y. et al. Innate C‑H trifluormethylation of
heterocycles. Proc. Natl Aacad. Sci. USA. 108,
14411–14415 (2011).
96. Zuo, Z. et al. Merging photoredox with nickel catalysis:
coupling of α-carbonyl sp3‑carbons with aryl halides.
Science 345, 437–440 (2014).
97. Zhang, X. & MacMillan, D. W. Alcohols as latent
coupling fragments for metallophotoredox catalysis:
sp3‑sp2 cross coupling of oxalates with aryl halides.
J. Am. Chem. Soc 138, 13862–13865 (2016).
98. Wang, Z., Herraiz, A. G., del Hoyo, A. M. &
Suero, M. G. Generating carbyne equivalents with
photoredox catalysis. Nature 554, 86–91 (2018).
99. Denisenko, A. V. et al. Photochemical synthesis of
3‑azabicyclo[3.2.0]heptanes: advanced building blocks
for drug discovery. J. Org. Chem. 82, 9627–9636
(2017).
100. Dirocco, D. A. et al. Late-stage functionalization of
biologically active heterocycles through photoredox
catalysis. Angew. Chem. Int. Ed. Engl. 53, 4802–4806
(2014).
101. Yan, M., Kawamata, Y. & Baran, P. S. Synthetic organic
electrochemical methods since 2000: on the verge of a
renaissance. Chem. Rev. 117, 13230–13319 (2017).
102. Jiang, Y., Xu, K. & Zeng, C. Use of electrochemistry in
the synthesis of heterocyclic structures. Chem. Rev.
118, 4485–4540 (2017).
103. Yoshida, J., Kataoka, K., Horcajada, R. & Nagaki, A.
Modern strategies in electroorganic synthesis. Chem.
Rev. 108, 2265–2299 (2008).
104. Morofuji, T., Shimizu, A. & Yoshida, J. Direct C‑N
coupling of imidazoles with aromatic and benzylic
compounds via electrooxidative C‑H functionalization.
J. Am. Chem. Soc. 136, 4496–4499 (2014).
105. Zhao, H.‑B., Hou, Z.-W., Liu, Z.‑J., Zhou, Z.‑F., Song, J.
& Xu, H.-C. Amidinyl radical formation through anodic
N‑H bond cleavage and its application in aromatic C‑H
bond functionalization. Angew. Chem. Int. Ed. Engl.
56, 587–590 (2017).
©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
106. Faust, M. R., Höfner, G., Pabel, J. & Wanner, K. T.
Azetidine derivatives as novel γ -aminobutyric acid
uptake inhibitors: synthesis, biological evaluation, and
structure-activity relationships. Eur. J. Med. Chem. 45,
2453–2466 (2010).
107. Elsler, B., Schollmeyer, D., Dyballa, K. M., Franke, R.
& Waldvogel, S. R. Metal- and reagent-free highly
selective anodic cross-coupling reactions of phenols.
Angew. Chem. Int. Ed. Engl. 53, 5210–5213 (2014).
108. Green, R. A., Brown, R. C. D. & Pletcher, D. A
microflow electrolysis cell for laboratory synthesis on
a multigram scale. Org. Process Res. Dev. 19,
1424–1427 (2015).
109. Ajami, A. Converging trends brings organic
electrochemistry to the front line of drug discovery.
BiopharmaTrend.com www.biopharmatrend.com/
post/40-converging-trends-brings-organic-
electrochemistry-to-the-front-line-of-drug-discovery/
(2017).
110. Cernak, T., Dykstra, K. D., Tyagarajan, S., Vachal, P. &
Krska, S. W. The medicinal chemist’s toolbox for late
stage functionalization of drug-like molecules. Chem.
Soc. Rev. 45, 546–576 (2016).
111. Wang, P. et al. Ligand-accelerated non-directed C‑H
functionalization of arenes. Nature 551, 489–493
(2017).
112. Loh, Y. Y. et al. Photoredox-catalyzed deuteration and
tritiation of pharmaceutical compounds. Science 358,
1182–1187 (2017).
113. Dai, H.‑X., Stepan, A. F., Plummer, M. S., Zhang, Y.‑H.
& Yu, J.‑Q. Divergent C‑H functionalizations directed
by sulfonamide pharmacophores: late-stage
diversification as a tool for drug discovery. J. Am.
Chem. Soc. 133, 7222–7228 (2011).
114. Wei, X. et al. Arylation and enantioselective
hydrogenation enables ideal asymmetric entry to the
indenopiperidine core of an 11β‑HSD‑1 inhibitor.
J. Am. Chem. Soc. 138, 15473–15481 (2016).
115. Li, C. et al. Decarboxylative borylation. Science 356,
eaam7355 (2017).
116. Lo, J. C., Yabe, Y. & Baran, P. S. A practical and
catalytic reductive olefin coupling reaction. J. Am.
Chem. Soc. 136, 1304–1307 (2014).
117. Lu, X. et al. Practical carbon-carbon bond formation
from olefins through nickel-catalyzed reductive
olefin hydrocarbonation. Nat. Commun. 7, 11129
(2016).
118. Lopchuk, J. M. et al. Strain-release heteroatom
functionalization: development, scope, and
stereospecificity. J. Am. Chem. Soc. 139, 3209–3226
(2017).
119. Ariki, Z. T., Maekawa, Y., Nambo, M. & Crudden, C. M.
Preparation of quaternary centers via nickel-catalyzed
Suzuki–Miyaura cross-coupling of tertiary sulfones.
J. Am. Chem. Soc. 140, 78–81 (2018).
120. Campbell, P. S., Jamieson, C., Simpson, I. &
Watson, A. J. B. Practical synthesis of
pharmaceutically relevant molecules enriched in sp3
character. Chem. Commun. 54, 46–49 (2018).
121. Ritchie, T. J. & Macdonald, S. J. Physicochemical
descriptors of aromatic character and their use in
drug discovery. J. Med. Chem. 57, 7206–7215
(2014).
122. Kolb, H. C., Finn, M. G. & Sharpless, K. B. Click
chemistry: diverse chemical function from a few good
reactions. Angew. Chem. Int. Ed. Engl. 40,
2004–2021 (2001).
123. Li, L. et al. Design of an amide N-glycoside derivative
of β-glucogallin: a stable, potent, and specific inhibitor
of aldose reductase. J. Med. Chem. 57, 71–77
(2014).
124. Tyler, D. S. et al. Click chemistry enables preclinical
evaluation of targeted epigenetic therapies. Science
356, 1397–1401 (2017).
125. Dömling, A., Wang, W. & Wang, K. Chemistry and
biology of multicomponent reactions. Chem. Rev. 112,
3083–3135 (2012).
126. Osipova, A., Yufit, D. S. & De Meijere, A. Synthesis of
new cyclopropylisonitriles and their applications in Ugi
four-component reactions. Synthesis 1, 131–139
(2007).
127. Liddle, J. et al. The discovery of GSK221149A: a
potent and selective oxytocin antagonist. Bioorg.
Med. Chem. Lett. 18, 90–94 (2008).
128. Zarganes-Tzitzikas, T. & Dömling, A. Modern
multicomponent reactions for better drug syntheses.
Org. Chem. Front. 1, 834–837 (2014).
129. Goodnow, R. A. & Davie, C. P. DNA-encoded library
technology: a brief guide to its evolution and impact
on drug discovery. Annu. Rep. Med. Chem. 50, 1–15
(2017).
 PERSPECTIVES

130. Arico-Muendel, C. C. From haystack to needle: finding
value with DNA encoded library technology at GSK.
MedChemComm 7, 1898–1909 (2016).
131. Satz, A. L. et al. DNA Compatible multistep synthesis
and applications to DNA encoded libraries. Bioconj.
Chem. 26, 1623–1632 (2015).
132. Thomas, B. et al. Application of biocatalysis to on‑DNA
carbohydrate library synthesis. Chembiochem 18,
858–863 (2017).
133. Goodnow Jr, R. A., Dumelin, C. E. & Keefe, A. D. DNA-
encoded chemistry: enabling the deeper sampling of
chemical space. Nat. Rev. Drug Discov. 16, 131–147
(2017).
134. Harris, P. A. et al. DNA-encoded library screening
identifies benzo[b][1,4]oxazepin‑4‑ones as highly potent
and monoselective receptor interacting protein 1 kinase
inhibitors. J. Med. Chem. 59, 2163–2178 (2016).
135. Harris, P. A. et al. Discovery of a first‑in‑class receptor
interacting protein 1 (RIP1) kinase specific clinical
candidate (GSK2982772) for the treatment of
inflammatory diseases. J. Med. Chem. 60,
1247–1261 (2017).
136. Soutter, H. H. et al. Discovery of cofactor-specific,
bactericidal Mycobacterium tuberculosis InhA
inhibitors using DNA-encoded library technology. Proc.
Natl Acad. Sci. USA 113, E7880–E7889 (2016).
137. Chan, A. I. et al. Discovery of a covalent kinase
inhibitor from a DNA-encoded small-molecule
library × protein library selection. J. Am. Chem. Soc.
139, 10192–10195 (2017).
138. Machutta, C. A. et al. Prioritizing multiple therapeutic
targets in parallel using automated DNA-encoded
library screening. Nat. Comm. 8, 16081 (2017).
139. Bollag, G. et al. Vemurafenib: the first drug approved
for BRAF-mutant cancer. Nat. Rev. Drug Discov. 11,
873–886 (2012).
140. Souers, A. J. et al. ABT‑199, a potent and selective
BCL‑2 inhibitor, achieves antitumor activity while
sparing platelets. Nat. Med. 19, 202–208 (2013).
141. Perera, T. P. S. et al. Discovery and pharmacological
characterization of JNJ‑42756493 (Erdafitinib), a
functionally celective small-molecule FGFR family
inhibitor. Mol. Cancer Ther. 16, 1010–1020 (2017).
142. Keserü, G. M. et al. Design principles for fragment
libraries: maximizing the value of learnings from
pharma fragment-based drug discovery (FBDD)
programs for use in academia. J. Med. Chem. 59,
8189–8206 (2016).
143. Palmer, N., Peakman, T. M., Norton, D. & Rees, D. C.
Design and synthesis of dihydroisoquinolones for
fragment-based drug discovery (FBDD). Org. Biomol.
Chem. 14, 1599–1610 (2016).
144. Morley, A. D. et al. Fragment-based hit identification:
thinking in 3D. Drug Discov. Today 18, 1221–1227
(2013).
145. Rizzo, S., Wakchaure, V. & Waldmann, H. in Natural
Products in Medicinal Chemistry (ed. Hanessian, S.)
Vol. 60 43–80 (Wiley-VCH Verlag GmbH & Co. KGaA,
2014).
146. Hall, R. J., Mortenson, P. N. & Murray, C. W. Efficient
exploration of chemical space by fragment-based
screening. Progr. Biophys. Mol. Biol. 116, 82–91
(2014).
147. Ferenczy, G. G. & Keserü, G. M. How are fragments
optimized? A retrospective analysis of 145 fragment
optimizations. J. Med. Chem. 56, 2478–2486 (2013).
148. Kathman, S. G. & Statsyuk, A. V. Covalent tethering of
fragments for covalent probe discovery. Med. Chem.
Commun. 7, 576–585 (2016).
149. Backus, K. M. et al. Proteome-wide covalent ligand
discovery in native biological systems. Nature 534,
570–574 (2016).
150. Parker, C. G. et al. Ligand and target discovery by
fragment-based screening in human cells. Cell 168,
527–541 (2017).
151. Wetzel, S., Lachance, H. & Waldmann, H. in
Comprehensive in Natural Products II (eds Mander, L.
& Liu, H.-W.) 5–46 (Elsevier, 2010).
152. Zaid, H., Raiyn, J., Nasser, A., Saad, B. & Rayan, A.
Physicochemical properties of natural based products
versus synthetic chemicals. Open Nutraceuticals J. 3,
194–202 (2010).
153. Li, J. W.‑H. & Vederas, J. C. Drug discovery and natural
products: end of an era or an endless frontier? Science
325, 161–165 (2009).
154. Gerry, C. J. & Schreiber, S. L. Chemical probes and
drug leads from advances in synthetic planning and
methodology. Nat. Rev. Drug Discov. 17, 333–352
(2018).
155. Pascolutti, M. & Quinn, R. J. Natural products as lead
structures: chemical transformations to create lead-like
libraries. Drug Discov. Today 19, 215–221 (2014).
156. Karawajczyk, A. et al. Expansion of chemical space for
collaborative lead generation and drug discovery: the
European Lead Factory Perspective. Drug Discov.
Today 20, 1310–1316 (2015).
157. Colomer, I. et al. A divergent synthetic approach to
diverse molecular scaffolds: assessment of lead-
likeness using LLAMA, an open-access computational
tool. Chem. Commun. 52, 7209–7212 (2016).
158. Foley, D. J., Nelson, A. & Marsden, S. P. Evaluating
new chemistry to drive molecular discovery: fit for
purpose? Angew. Chem. Int. Ed. Engl. 55,
13650–13657 (2016).
159. Chow, S. Y. & Nelson, A. Embarking on a chemical
space odyssey. J. Med. Chem. 60, 3591–3593
(2017).
160. Dow, M., Fisher, M., James, T., Marchetti, F. &
Nelson, A. Towards the systematic exploration of
chemical space. Org. Biomol. Chem. 10, 17–28
(2012).
161. Morgentin, R. et al. Translation of innovative
chemistry into screening libraries: an exemplar
partnership from the European Lead Factory. Drug
Discov. Today https://doi.org/10.1016/j.
drudis.2018.05.007 (2018).
162. Doveston, R., Marsden, S. & Nelson, A. Towards the
realisation of lead-oriented synthesis. Drug Discov.
Today 19, 813–819 (2014).
163. Mayol-Llinàs, J., Nelson, A., Farnaby, W. &
Ayscough, A. Assessing molecular scaffolds for CNS
drug discovery. Drug Discov. Today 22, 965–969
(2017).
164. Dow, M. et al. Modular synthesis of diverse natural
product-like macrocycles: discovery of hits with
antimycobacterial activity. Chem. Eur. J. 23,
7207–7211 (2017).
165. Karageorgis, G., Warriner, S. & Nelson, A. Efficient
discovery of bioactive scaffolds by activity-directed
synthesis. Nat. Chem. 6, 872–876 (2014).
166. Karageorgis, G., Dow, M., Aimon, A., Warriner, S. &
Nelson, A. Activity-directed synthesis with
intermolecular reactions: Development of a fragment
into a range of androgen receptor agonists. Angew.
Chem. Int. Ed. Engl. 54, 13538–13544 (2015).
167. Bootwicha, T., Feilner, J. M., Myers, E. L. &
Aggarwal, V. K. Iterative assembly line synthesis of
polypropionates with full stereocontrol. Nat. Chem. 9,
896 (2017).
168. Balieu, S. et al. Toward ideality: the synthesis of
(+)-kalkitoxin and (+)-hydroxyphthioceranic acid by
assembly-line synthesis. J. Am. Chem. Soc. 137,
4398–4403 (2015).
169. Ardkhean, R. et al. Cascade polycyclizations in natural
product synthesis. Chem. Soc. Rev. 45, 1557–1569
(2016).
170. Dückert, H. et al. Natural product-inspired cascade
synthesis yields modulators of centrosome integrity.
Nat. Chem. Biol. 8, 179 (2011).
171. Schafroth, M. A., Zuccarello, G., Krautwald, S.,
Sarlah, D. & Carreira, E. M. Stereodivergent total
synthesis of Δ9‑tetrahydrocannabinols. Angew. Chem.
Int. Ed. Engl. 53, 13898–13901 (2014).
172. Davis, A. M., Plowright, A. T. & Valeur, E. Directing
evolution: the next revolution in drug discovery? Nat.
Rev. Drug Disc. 16, 681 (2017).
173. Shen, X., Corey, Chemistry, D. R. mechanism and
clinical status of antisense oligonucleotides and duplex
RNAs. Nucl. Ac. Res. 46, 1584–1600 (2018).
174. Lai, A. C. & Crews, C. M. Induced protein degradation:
an emerging drug discovery paradigm. Nat. Rev. Drug
Discov. 16, 101–114 (2017).
175. Sternbach, L. The benzodiazepine story. J. Med.
Chem. 22, 1–7 (1979).
176. Wright, P. M., Seiple, I. B. & Myers, A. G. The evolving
role of chemical synthesis in antibacterial drug
discovery. Angew. Chem. Int. Ed. Engl. 34,
8840–8869 (2014).
177. Flam, F. The race to synthesize taxol ends in a tie.
Science 263, 910–911 (1994).
178. Donehower, R. C. The clinical development of
paclitaxel: a successful collaboration of academia,
industry and the national cancer institute. Oncol. 1,
240–243 (1996).
179. Ringel, M., Tollman, P., Hersch, G. & Schulze, U. Does
size matter in R&D productivity? If not, what does?
Nat. Rev. Drug Disc. 12, 901–902 (2013).
180. Besnard, J., Jones, P. S., Hopkins, A. L. &
Pannifer, A. D. The Joint European Compound Library:
boosting precompetitive research. Drug Discov. Today
20, 181–186 (2015).
181. Collins, K. D. & Glorius, F. A robustness screen for the
rapid assessment of chemical reactions. Nat. Chem. 5,
597–601 (2013).
182. Chen, H., Engkvist, O., Wang, Y., Olivecrona, M. &
Blaschke, T. The rise of deep learning in drug
discovery. Drug Discov. Today. 23, 1241–1250
(2018).
183. Fleming, N. How artificial intelligence is changing drug
discovery. Nature 557, S55–S57 (2018).
184. Macdonald, S. J., Fray, M. J. & McInally, T. Passing on
the medicinal chemistry baton: training
undergraduates to be industry-ready through research
projects between the University of Nottingham and
GlaxoSmithKline. Drug Discov. Today 21, 880–887
(2016).
185. Urquhart, L. Market watch: top drugs and companies
by sales in 2017. Nat. Rev. Drug Discov. 17, 232
(2018).
186. Cernak, T. et al. Nanoscale synthesis and affinity
ranking Nature 557, 228–232 (2018).
187. Pant, S. M. et al. Design, synthesis and testing of
potent, selective hepsin inhibitors via application of an
automated closed-loop optimization platform. J. Med.
Chem. 61, 4335–4347 (2018).
188. Cole, K. P. et al. Kilogram-scale prexasertib
monolactate monohydrate synthesis under
continuous-flow CGMP conditions. Science 356,
1144–1150 (2017).
189. Cooper, T. W., Campbell, I. B. & Macdonald, S. J.
Factors determining the selection of organic reactions
by medicinal chemists and the use of these reactions
in arrays (small focused libraries). Angew. Chem. Int.
Ed. Engl. 49, 8082–8091 (2010).
Acknowledgements
G.M.K. is supported by the National Brain Research Program
(2017–1.2.1‑NKP‑2017‑00002) of the National Research,
Development and Innovation Office, Hungary.
Competing interests
The authors declare competing interests: see Web version for
details.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
Supplementary information
Supplementary information is available for this paper at
https://www.nature.com/articles/nrd.2018.116.
RELATED LINKS
Chemplanner: www.chemplanner.com
Enamine REAL: www.enamine.net/index.php?option=com_
content&task=view&id=254
EvoSpace: www.evotec.com
Hypha Discovery: www.hyphadiscovery.co.uk
ICFRP: www.infochem.de/products/software/icfrp.shtml
Mcule ULTIMATE: mcule.com/ultimate-project
Reaxys: www.reaxys.com
SciFinder: https://scifinder.cas.org/scifinder/
Synspace: www.chempassltd.com/synspace
ALL LINKS ARE ACTIVE IN THE ONLINE PDF

NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 19

©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.