Rapid discovery of a novel series of Abl kinase inhibitors by application of an integrated microfluidic synthesis and screening platform

Article abstract of DOI:10.1021/jm400099d

Drug discovery faces economic and scientific imperatives to deliver lead molecules rapidly and efficiently. Using traditional paradigms the molecular design, synthesis, and screening loops enforce a significant time delay leading to inefficient use of data in the iterative molecular design process. Here, we report the application of a flow technology platform integrating the key elements of structure-activity relationship (SAR) generation to the discovery of novel Abl kinase inhibitors. The platform utilizes flow chemistry for rapid in-line synthesis, automated purification, and analysis coupled with bioassay. The combination of activity prediction using Random-Forest regression with chemical space sampling algorithms allows the construction of an activity model that refines itself after every iteration of synthesis and biological result. Within just 21 compounds, the automated process identified a novel template and hinge binding motif with pIC₅₀ > 8 against Abl kinase - both wild type and clinically relevant mutants. Integrated microfluidic synthesis and screening coupled with machine learning design have the potential to greatly reduce the time and cost of drug discovery within the hit-to-lead and lead optimization phases.

Full text of DOI:10.1021/jm400099d

Article
pubs.acs.org/jmc
Rapid Discovery of a Novel Series of Abl Kinase Inhibitors by
Application of an Integrated Microfluidic Synthesis and Screening
Platform
Bimbisar Desai,† Karen Dixon,† Elizabeth Farrant,† Qixing Feng,† Karl R. Gibson,‡ Willem P. van Hoorn,§
James Mills,‡ Trevor Morgan,† David M. Parry,† Manoj K. Ramjee,† Christopher N. Selway,*,†
Gary J. Tarver,† Gavin Whitlock,‡ and Adrian G. Wright†
†Cyclofluidic Ltd, Biopark, Broadwater Road, Welwyn Garden City, AL7 3AX, U.K.
‡Sandexis LLP, Discovery Park, Sandwich, Kent CT13 9ND, U.K.
§Accelrys Ltd, 334 Cambridge Science Park, Cambridge, CB4 0WN, U.K.
S
* Supporting Information
ABSTRACT: Drug discovery faces economic and scientific
imperatives to deliver lead molecules rapidly and efficiently.
Using traditional paradigms the molecular design, synthesis, and
screening loops enforce a significant time delay leading to
inefficient use of data in the iterative molecular design process.
Here, we report the application of a flow technology platform
integrating the key elements of structure−activity relationship
(SAR) generation to the discovery of novel Abl kinase
inhibitors. The platform utilizes flow chemistry for rapid in-
line synthesis, automated purification, and analysis coupled with
bioassay. The combination of activity prediction using Random-
Forest regression with chemical space sampling algorithms
allows the construction of an activity model that refines itself
after every iteration of synthesis and biological result. Within just 21 compounds, the automated process identified a novel
template and hinge binding motif with pIC50 > 8 against Abl kinase  both wild type and clinically relevant mutants. Integrated
microfluidic synthesis and screening coupled with machine learning design have the potential to greatly reduce the time and cost
of drug discovery within the hit-to-lead and lead optimization phases.
otherwise of the program as well as the time and resources
required to launch.
1. GENERAL INTRODUCTION
The process of discovering a drug (Figure 1) is slow and
expensive, with the average cost of developing a new molecular
entity (NME) now estimated at $1.8 billion.1
Small molecule lead discovery involves an iterative process of
molecular design, chemical synthesis, biological assay, and data
analysis to feed into the next learning cycle. In a typical hit-to-
lead project, in vitro assays are used to measure the potency and
selectivity of the molecule at the target of interest as well as a
range of calculated and measured physical properties that help
to predict a lead molecule’s “drug-likeness”.4,6,7 Using conven-
tional approaches, each learning cycle in this process takes 1−8
weeks, depending on where compounds are made and tested.
These cycle times lead to slow and expensive hit-to-lead
exploration, limiting the number of lead series that can be
assessed. One of the key bottle-necks in this iterative process is
the time taken in synthesis and screening activities, which
lengthens the time between hypothesis generation/compound
design and the result (test data) of that design. This has two
potential detrimental effects to a drug discovery program,
Recent analysis1 has shown that for every NME launched,
19.4 hit-to-lead programs are required at a cost per launch of
$166 million, due to the high attrition rates in the drug
discovery process. Although much of this attrition (27%) is
attributed to lack of efficacy in humans in phase II, the majority
is reported to be related to problems of safety, toxicology,
formulation, pharmacokinetics, and cost of goods (53%).2
These characteristics are defined by the chemical structure of
the molecule itself.
Drug structures have been shown3−5 to be very closely
related to their lead compound series and these lead series to
the hit series from which they were optimized, indicating that
much of the success or failure of a potential drug molecule is
laid down at this very early point in the drug discovery process.
The decision regarding which lead series to pursue in the hit-to-
lead phase is therefore crucial and often dictates the success or
Received: January 21, 2013
© XXXX American Chemical Society
A
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
Figure 1. Drug discovery pipeline to produce one NME including cost per phase ($M).
either: (1) low program velocity  the number of design loops
undertaken in a given time frame may be limited, reducing the
progress that can be made; or (2) design “at risk”  designs
are made without full integration of the test results into the
hypothesis generation and design because the assay result of the
previous iteration is often not yet available when the next
design iteration occurs. Both of these factors, separately or in
combination, reduce program productivity.
To accelerate hit-to-lead processes, design hypotheses can be
explored in parallel. For example, if orthogonal stripes of
analogues are made from within a virtual matrix this allows
more rapid mapping of chemical space. Structure−activity
relationship (SAR) analysis can then be used to predict where
the most active compounds lie (Figure 2). This approach is
prediction model can select from and that could be synthesized
during an automated experiment  the set is defined by a
combination of the microfluidic chemical reactions and the
diversity of reactants available on the automated system. Some
chemistry development time is required to ensure that
analogues can be made in a microfluidic reactor, and the
efficient resolution of synthetic chemistry challenges is critical
to the success of this approach. To assist this there is an
expanding repertoire of flow chemistry reactions appearing in
the literature.14−20 The automated “in line” nature of
microfluidic synthesis, purification, and screening means that
a “continuous design process” is possible, with the objective of
accurately mapping the SAR within the virtual chemical space
using as few design−synthesis−screening loops as possible.
Initially, this approach mirrors that of a traditional hit-to-lead
program, namely, hit generation activities via, for example, high-
throughput screening (HTS), other screening approaches, or
prior art review. From this, the virtual chemical space of target
molecules is constructed that defines the boundaries of an SAR
heat map. An initial activity model is then built using data
available from a screening campaign or the literature against the
defined biological target. This model is used to decide which
analogue is made during each iteration of synthesis and testing,
and the model is updated after each individual compound assay
to incorporate the new data. Typically the coupled design,
synthesis, and assay times are 1−2 h per iteration.
Figure 2. Synthesis of two stripes of analogues (shown with an X)
within a matrix can be used to predict active compounds (e.g., P) if
SAR is additive.
Two design strategies are used to efficiently explore the
virtual chemical space. The initial activity model undergoes an
intensive learning process by systematically sampling unknown
areas of the space  “most active under sampled”. For 20
design−synthesis−screening runs, this could be completed
within 24 h. Depending on the preferences set the next step of
iterative design is guided by this updated activity model. For
example, further iterations can then be carried out to “chase
potency” and rapidly identify the most active compounds in the
virtual chemical space, while still only making a small
percentage of the molecules within that space. The aim is to
map out biological activity for the series by making a small
number of compounds, quickly identifying areas of potent and
weak activity without having to make large numbers of
compounds. These two strategies are complementary when
applied within a particular project, allowing both exploration of
new reagents and the optimization of activity.
most effective in those cases where SAR is additive; i.e., in the
case shown in Figure 2, the two axes represent independent
structural changes in which one structural change does not
interact or effect the activity of a structural change on the other
axis.8,9 However, the rate-limiting steps of synthesis and
screening still reduce the efficiency of this process, and the
matrix of active compounds will be poorly understood and
predicted in cases where SAR is nonadditive.10
In this paper, we detail a practical example of integrated hit-
to-lead optimization using a microfluidic synthesis and
screening approach11−13 (Figure 3), coupled with a “design
layer” activity mapping algorithm facilitating sampling within
multidimensional chemical space, thereby creating a bioactivity
prediction model that is automatically updated after every
single screening result. For the work described here, “virtual
chemical space” is defined as the set of molecules that the
B
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
Figure 3. (Left) Schematic of the integrated design, synthesis, and screening platform illustrating the fully automated processes implemented for
closed-loop drug discovery. Following initiation of the process the system completes multiple iterations of design, synthesis, and screening without
manual intervention. (Right) Schematic showing the continuous fluidic path taken by reagents and products on the platform. A full description of the
platform is provided in the Supporting Information.
Figure 4. First, second, and third generation BCR-Abl kinase inhibitors.
The potential of this technology is illustrated here through its
application to the discovery of novel breakpoint cluster region
Abelson tyrosine kinase (BCR-Abl) inhibitors.
and commercially available reagents31 would minimize assay
development time. All compounds prepared would be tested at
both Abl1 and Abl2 kinase subtypes, although activity models
and compound selection would be on the basis solely of Abl1
kinase activity.
The discovery of BCR-Abl inhibitors (Figure 4) has
revolutionized the treatment of chronic myeloid leukemia
(CML). The first BCR-Abl inhibitor to be launched, imatinib
(1), shows robust clinical efficacy and is now a front-line
therapy for the treatment of CML.21 However, a small number
of patients on imatinib therapy relapse and become imatinib-
resistant.22 Second generation inhibitors nilotinib (2)23 and
dasatinib (3)24 have shown efficacy against several clinically
relevant mutations in the Abl active site. However, imatinib,
nilotinib, and dasatinib all show poor activity against the T315I
“gatekeeper” mutant, which is responsible for around 15−20%
of all clinical mutants.25 The third generation BCR-Abl
inhibitor ponatinib (4)26 has shown impressive clinical efficacy
in patients expressing the T315I mutation27 and has recently
been approved by the FDA.28 The ability of a number of Abl
kinase mutations to confer resistance against the first and
second generation agents means that BCR-Abl remains an
attractive drug target that warrants continued investigation.29
BCR-Abl thus provides a clinically relevant drug target to
demonstrate all aspects of the integrated microfluidic syn-
thesis−screening approach (virtual chemical space enumera-
tion, design layer algorithms, automated synthesis-screening-
design loops). The significant existing knowledge around the
target also enables an assessment of whether this approach
could quickly find novel inhibitors in an area with significant
prior art and demonstrate whether the methodology can rapidly
deliver successful hit-to-lead optimization against soluble
protein targets such as kinases. Finally, validated bioassays30
2. PROJECT SETUP
2.1. Defining the Virtual Chemical Space. A thorough
review of the Abl inhibitor prior art was undertaken to assess
the templates from which to define a relevant virtual chemical
space. The focus was primarily on those templates that had
shown the potential to inhibit the T315I mutant enzyme. The
alkyne series exemplified by ponatinib combines wild type
(WT) Abl inhibition along with potent activity at key mutants
such as T315I.26 An X-ray structure of ponatinib bound to Abl
kinase has been published and the binding mode is under-
stood.32 The series contains an imidazopyridazine hinge
binding group linked to a structural motif that binds to the
“DFG out” conformation of Abl and terminates in a pendant
basic amine. The alkyne series also offer some synthetic
advantages, giving the potential to build a matrix of hinge-
binding heterocycles and a series of alternative DFG-out
binding units, while performing the key carbon−carbon bond
forming reaction under flow chemistry conditions (Figure 5).
A key criterion for operating in this template was delivering a
new series of Abl inhibitors, and we reasoned that novelty could
come both from previously unexplored hinge-binding hetero-
cycles and potential DFG-binding motifs. The search for such
motifs focused on identifying structures that would occupy the
DFG out conformation and also present an acetylene linker at
the correct vector to deliver the hinge binding group.
C
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
R2 (RMS < 0.2 Å). The moieties showing some overlap with
the imidazopyridazine group of ponatinib were considered as
potential Abl R2 groups if they were able to mimic the
hydrogen-bonding functionality of the ponatinib imidazopyr-
idazine (i.e., the hydrogen bonds to E316O, M318N, and
M318O). The most commonly observed R2 substructure to
meet these criteria was a heterocycle with a nitrogen acceptor
atom in the 3-position relative to the core attachment point
(Figure 9). Therefore, a diverse selection of 27 such R2 groups
was made from the list of bromoheterocycles, including a
control selection of building blocks (R2 = substituted phenyl)
that were expected not to bind but could serve to test the ability
of the model to discriminate inactives.
Figure 5. Design of the virtual chemical space for acetylene-linked Abl
inhibitors.
A protein data bank (PDB)34 mining approach was used to
identify potential alternative core groups that express vectors
similar to those present in ponatinib in its binding
conformation, as observed in PDB entry 3OXZ (Figure 6).
An in-house database of all ligands observed in the PDB was
mined to identify all pairs of acyclic single bonds that match
these two vectors to within a 0.5 Å threshold. These results
were filtered by overlapping the resultant linker moieties into
the 3OXZ crystal structure and removing those that clearly gave
steric clashes. The remaining examples were assessed by eye,
prioritizing those that could retain the hydrogen-bonding
features present in the original ligand, or modifying them
slightly to accommodate these interactions. Of the groups that
emerged from this analysis three were selected and are shown
in Figure 7. The amide template A, to which ponatinib belongs,
was included to determine if novel heterocycle hinge binding
motifs could quickly be found in this experiment. Whereas
there was some precedence for the reversed amide template
B,35 the pyrazole urea C was an unprecedented replacement in
Abl, and would represent a clearly novel inhibitor series if found
to be active.
As R2 was the main area targeted for diversity in the virtual
chemical space, only a small range of R1 basic groups was
chosen, some with precedent from the Abl literature, and others
that had not been exemplified before, such as 4-hydroxypiper-
idine (Figure 10).
2.2. Chemistry. Template Synthesis. A total of 10 DFG
binding templates (Table 1) were synthesized using the
synthetic schemes outlined below, which are separated into
three chemical series.
Ponatinib templates A to generate direct analogues of
ponatinib were synthesized according to Scheme 1. An SN2
substitution of 2-trifluoromethyl-4-nitro benzyl bromide 540
was carried out with amines 9−12. Reduction of the arylnitro
intermediates 6 to the corresponding anilines 7 was achieved
using Raney Nickel (CatCart) catalyzed hydrogenation in a H-
Cube hydrogenation reactor.41 Finally, the four iodobenzamide
templates 8-1 to 8-4 were prepared by reaction of anilines 7
with 3-iodo-4-methylbenzoyl chloride (prepared by refluxing 3-
iodo-4-methylbenzoic acid in thionyl chloride). In addition, the
un-substituted aniline derivative 8-5 was prepared by reaction
of 3-iodo-4-methylbenzoyl chloride with aniline.
The reversed amide templates B were prepared according to
Scheme 2. Bromination of the acid 13 was achieved using
sodium bromate and sodium hydrogen sulfite42 and cleanly
gave the intermediate bromoacid 14 which was coupled with 3-
iodo-4-methyl aniline using 1-chloro-N,N,2-trimethyl-1-prope-
nylamine to give intermediate 15. Subsequent displacement
reactions with amines 9−12 and standard purification methods
gave the four required templates 16-1 to 16-4.
On the basis of precedented synthetic methods,36 we
rationalized that Sonogashira coupling of the DFG template
halides with a set of aryl/heteroaryl alkynes (Figure 8) would
allow the hinge binding heterocycle R2 to be incorporated in
the integrated synthesis−screening flow apparatus.
Relatively few heteroaryl alkynes are commercially available,
but around 1000 of their precursor bromo- or iodo-heterocycles
are available.37 This large number of potential R2 groups was
triaged by examining the nature of other hinge binding motifs
as observed in other kinase PDB crystal structures. To achieve
this, a BLAST search of the in-house PDB ligand database was
conducted yielding 705 entries with a ligand bound to a protein
with high sequence similarity to Abl. All of these PDB binding
sites were overlaid onto that of 3OXZ using a quaternion
method,38 implemented using the Tinker molecular modeling
software suite,39 to overlap the binding site backbone atoms.
The resultant overlaps were then filtered to those in which
there is a bond that overlays with the bond joining the core to
The pyrazole urea based template C was prepared according
to Scheme 3. Piperazine 17 formed from reaction of compound
5 with 9 and subsequent H-Cube reduction was treated with
carbonyl diimidazole in DMSO for 30 min and subsequently
treated with aminopyrazole 18 for 1 h at room temperature.
Standard purification methods furnished the bromopyrazole 19.
Figure 6. Superposition of two molecules that mimic the linker bonds of ponatinib (green) as observed in its Abl-binding conformation (PDB entry
3OXZ). The reverse amide (orange) is found in a ligand bound to ABL2 (3GVU) and the pyrazole (cyan) in a p38-binding ligand (3HV6).33
D
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
Figure 7. Selected DFG binding templates for virtual chemical space construction and their synthetic precursors used in the experimental work.
Figure 8. Example of Sonogashira coupling chemistry used in the automated synthesis of Abl inhibitors.
Figure 9. Examples of heterocycles that mimic the R2 moiety of ponatinib in terms of the attachment vector and the interactions with E316 (top left)
and M318 (top right). The commonly observed hinge-binding donor−acceptor−donor functionality can be achieved from the same ring (PDB entry
1YDR, green) or by alternative ring systems (3SKC, yellow).
Sonogashira Flow Chemistry Methods. Compounds
selected by the design algorithm were synthesized on the
platform by employing Sonogashira reactions in flow using a
Vaportec R4 synthesizer (Figure 11) remotely controlled from
a central computer and using a linked modified Gilson 215
sample handler to automatically load injection loops. A 1 mm
ID copper coil heated at 150 °C was used as a flow reactor, and
the two selected solutions in DMF containing either template
or alkyne and palladium catalyst, both at 120 mM
concentration of reactant, were mixed and passed through the
coil at a total flow rate of 100 μL/min to give a total reaction
time of 20 min. The outflow was filtered through 250 mg of
silica contained in a 5 mL 0.6 cm ID Omnifit glass tube to
remove metal catalysts and then passed to a 10 μL injection
loop switched automatically at the point of maximum product
concentration and thence to the integrated LC/MS purification
and bioassay system.
2.3. Algorithm Driven Design Strategies. A key feature
of the technology is the employment of computational activity
prediction methodology coupled with automated synthesis of
compounds (from the virtual chemical space) according to a
chosen design strategy. This utilizes Random Forest activity
prediction employing one or both of two design strategies 
“chase potency” and “most active under sampled”.
Figure 10. Selected R1 substituents for the DFG binding templates.
Synthesis of Novel R2 Alkynes. Noncommercial aromatic
alkynes were synthesized from their precursor aromatic
bromide or iodide using standard methodology as shown in
Scheme 4.
The halo-aromatic or halo-heteroaromatic substrate was
treated with ethynyltrimethylsilane 20 under palladium and
copper-catalyzed Sonogashira conditions to yield the trime-
thylsilyl protected alkynes 21. The TMS group was easily
removed using potassium carbonate to yield the free alkynes
22. The commercial and synthesized alkynes used on the
platform as hinge binding motifs are shown in Table 2.
The combination of the 10 DFG binding templates and 27
hinge binding motifs defines the set of 270 compounds
available for immediate synthesis on the platform.
E
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
Table 1. Synthesised DFG Binding Templates of Types A, B, and C
a
b
c
Type A, Synthesis Scheme 1. Type B, Synthesis Scheme 2. Type C, Synthesis Scheme 3.
a
Scheme 1. Synthesis of Ponatinib Templates A
a
Reagents and conditions: (a) R = piperazine, 4-hydroxyethylpiperazine, imidazole, 4-hydroxypiperidine, CH2Cl2, 1 h, (b) H-Cube Hydrogenation,
40 °C, 10% Ra−Ni CatCart (THS01112), MeOH, (c) 3-iodo-4-methylbenzoyl chloride, cat. DMAP, (i-Pr)2NEt, THF, RT.
a
Scheme 2. Synthesis of Reversed Amide Templates B
a
Reagents and conditions: (a) NaBrO3, NaSO3H, EtOAc/H2O, 50 °C 18 h. (b) 1-Chloro-N,N,2-trimethyl-1-propenylamine, Et3N, (c) 9, 10, 11, or
12, CH2Cl2, 1 h.
Random Forest Activity Prediction. Regression models are
built using the Random Forest method implemented in R
accessed via the Pipeline Pilot ‘Learn R Forest Model’
component.43 The default settings of this component are
used except for the molecular descriptors: ALogP, Molecular_-
Weight, Num_H_Donors, Num_H_Acceptors, Num_Rotata-
bleBonds, Molecular_SurfaceArea, Molecular_PolarSurfaceAr-
ea, which are used together with ECFP_6 fingerprints. The
fingerprints are not folded and are used as an array of counts.
The Random Forest algorithm was chosen because it can
F
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
a
Prior Knowledge Compounds. Construction of the activity
prediction model requires knowledge of activity of compounds
from within the virtual chemical space or in similar external
chemical space. This may be available from HTS data, the
literature, or by off-line synthesis and screening of analogues.
Fortunately, the Abl1 IC50 activity of numerous ponatinib
analogues has been published,26 so a total of 36 literature
compounds were used to seed the design model in this work.
The full prior knowledge data set is provided in Supporting
Information, Table S1. The heat map of predicted activity of
the set of 270 compounds available for synthesis on the
platform, based on seeding the model with this prior knowledge
set, is shown in Figure 12.
Scheme 3. Synthesis of Pyrazole Urea Template C
a
Reagents and conditions: (a) 17, CDI, DMSO, rt, 30 min, 18 1 h.
Scheme 4. Synthesis of Aromatic Alkynes from Aryl Halides
2.4. Platform Validation. Prior to commencing SAR
generation on the microfluidic platform validation tests were
run to ensure platform IC50 data were consistent, both with a
manual assay and literature data. Imatinib was chosen as a
suitable Abl1 inhibitor and was synthesized in flow50 and
screened against Abl1 both on the platform and manually off-
line. The Abl1 activity values derived from the platform are
consistent with the manual assay and literature values (Table
3). Experimental details are provided in Supporting Informa-
tion.
2.5. Experimental Strategy. To fully explore SAR using
the design−synthesis−screen methodology, the experimenta-
tion on the microfluidic platform was split into three parts. In
part one the “most active under sampled” design strategy was
used to maximize coverage of the design space and rapidly
identify areas of higher activity; in part two the “chase potency”
strategy was applied to further map the high activity areas; and
finally in part three the “combined” strategy was used to
identify and optimize additional potential activity “hot spots”.
perform nonlinear regression, is resistant to overfitting, and has
very few tuning parameters.44 In all cases the dependent
variable is pIC50. Activity values for which only an upper or
lower bound is known are treated as follows: lower bound
values like ‘IC50 < value’ are considered the same as ‘IC50
=
value’. Upper bound values like ‘IC50 > value’ are interpreted as
‘inactive’. The user can supply a detection limit for IC50 values
which will substitute ‘inactive’ values. If this value is not
supplied, compounds with ‘inactive’ values are ignored. Since
the system is designed to run unsupervised, the training set is
not split to derive model performance metrics like the R2
correlation coefficient.
Design Strategies. (a) Chase Potency. The activity is
predicted for all members of the pool of eligible virtual
compounds. The compounds are sorted high to low by
predicted activity, and the top scoring compound(s) are
selected for synthesis.
3. RESULTS AND CONCLUSIONS
Platform SAR Generation - Part One. Once the Abl1 and
Abl2 biochemical assays, synthetic protocols, and reactants
were in place on the platform, the first round of design−
synthesis−screening loops was initiated using the “most active
under sampled” methodology in order to accelerate the
learning of the activity algorithm and quickly provide an SAR
landscape across the virtual chemical space. The first fully
automated experiment consisted of 29 loops yielding SAR for
22 new compounds, with 6 compounds failing synthesis and
one failing assay quality control (QC) (76% success rate). Each
design−synthesis−screening loop took 90 min to complete
with a total instrument time of approximately 30 h. It quickly
delivered an SAR heat map (Figure 13) that identified “hot-
spots” of activity, along with areas of weak potency.
In particular, the pyrazole urea template (19) was rapidly
identified as having exciting levels of Abl1 and Abl2 enzyme
inhibition, with the 7th, 10th and 16th analogues to be made
containing this motif. The ponatinib template 8-1 showed
excellent levels of Abl1 and Abl2 potency and gave the most
potent compound (synthesized in loop 26) in combination
with a novel hinge-binding heterocycle (from alkyne 22-5).
Novel templates from the ponatinib series A, e.g., 8-2 used in
loop 27 also showed reasonable Abl1 potency (IC50 = 60 nM).
As stated above, a number of phenyl motifs were purposely
included that would not be expected to form interactions at the
hinge and therefore would be weak Abl inhibitors. This was
rapidly confirmed within this part of the experiment, for
example, the 4-fluorophenyl motif 22−20 combined with the
reverse amide template 16-1 in loop 17 had an Abl1 IC50 of
(b) Most Active Under Sampled. This is a variant on the
Latin Hypercube Design.45 For each reactant, the number of
times it has been incorporated into a previously selected
product is counted. Only attempted syntheses carried out
during the current experiment are included; i.e., the number of
times a reactant has been used to synthesize a prior knowledge
compound does not count. For each eligible virtual product, the
total is calculated of the number of times the constituting
reactants have been used previously. The compounds with the
lowest total previous usage are considered to be under sampled,
and synthesizing compounds from this pool ensures all
reactants are sampled as equally as possible. This is assumed
to accelerate learning and coverage of the chemical space
although diversity of the products is not explicitly taken into
account, and more advanced reagent selection algorithms46−49
may be needed if increasing and optimal sampling of product
diversity is considered paramount. Typically more compounds
are present in the group of under sampled compounds than are
desired for the next round of synthesis; therefore this group is
further ranked by the “chase potency” strategy to pick the most
promising compound(s).
Combined Strategy. It is also possible to operate the
platform using a combined strategy that automatically switches
between “chase potency” and “most active under sampled”
methods, sequentially running multiple loops of each type as
defined by the operator.
G
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
Table 2. Synthesized and Commercial Hinge Binding Aromatic Alkynes Used on the Platform
a
*
Compound stored with the TMS group present to assist stability. This was removed during platform synthesis. Commercially available
compounds.
Platform SAR Generation - Part Two. A second
experiment was carried out applying the “chase potency”
paradigm, looking to optimize potency within the active areas
identified in part one. The experiment consisted of 20 design-
loops, yielding 14 additional compounds (6 compounds failed
in synthesis, 70% success rate) and delivered a refined heat map
(Figure 14). The novel hinge-binding heterocycle (from alkyne
22-5) gave potent activity in the pyrazole urea (loop 30), amide
(loop 33), and reverse amide (loop 36) templates, confirming
the activity first seen in loop 26. As well as identifying these
areas of high activity, the model also predicted activity in the
areas associated with, for example, motifs 22-14, 22-25, and 22-
26.
Figure 11. Schematic of the flow apparatus used in the Sonogashira
flow synthesis.
Platform SAR Generation - Part Three. A final
experiment was carried out using the “Combined” design
strategy, by repeatedly applying six loops of “most active under
sampled” followed by six loops of “chase potency”. The
only 1.5 μM. The compounds synthesized in loops 19, 23, and
24 also showed this trend, enabling the model to correctly
predict weak activity for all the phenyl motifs.
H
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
Figure 12. Initial heat map of predicted Abl1 activities of the 270 potential inhibitors based on the prior knowledge data. Prior knowledge
compounds are labeled ‘0’, the most active compound found so far by ‘!’ and the virtual compound selected for synthesis by ‘P1’ (the ‘most active
under sampled’ strategy was applied precluding selection from column 22-6).
Table 3. Platform Validation Data
a
b
Average of two independent experiments. Reference 51.
Figure 13. Heat map of predicted Abl1 activities after 29 design−synthesis−screening loops using the “most active under sampled” strategy.
Numbers within the heat map indicate order of synthesis with prior knowledge compounds assigned ‘0’. The most active compound found so far is
labeled with ‘!’ and ‘?’ indicates a synthesis, QC, or bioassay failure.
experiment consisted of 41 design-loops, yielding 28 additional
compounds (13 compounds failed in synthesis, 68% success
rate) and delivered the final activity heat map (Figure 15). The
chemical space explored more thoroughly in this part included
the areas predicted to have activity at the end of part 2, in
particular, the compounds derived from hinge binders 22−25
and 22−26.
In total, 90 design−synthesis−screening loops provided 64
new compounds with a measured IC50 against Abl1 and Abl2.
Overall the flow chemistry, purification, and bioassay proceeded
with a success rate of 71%. It is worthy of note that the
microfluidic platform is able to screen and identify active
compounds even when the chemistry is low yielding; for
example, in loop 44 the chemical yield was only 8%, but a 3 nM
I
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
Figure 14. Heat map of predicted Abl1 activities after 49 design−synthesis−screening loops. Numbers within the heat map indicate order of
synthesis with prior knowledge compounds assigned ‘0’. The most active compound found so far is labeled with ‘!’ and ‘?’ indicates a synthesis, QC,
or bioassay failure.
Figure 15. Heat map of predicted Abl1 activities after 90 design−synthesis−screening loops. Numbers within the heat map indicate order of
synthesis with prior knowledge compounds assigned ‘0’. The most active compound found so far is labeled with ‘!’ and ‘?’ indicates a synthesis, QC,
or bioassay failure.
Abl1 inhibitor was nominally identified. This attribute is
associated with the excellent resolution of byproducts from the
desired compound during HPLC purification, combined with
microfluidic integration and minimized manipulation of
material between synthesis and screening. Full SAR and
analytical data from the 90 design-synthesis-screening loops
may be found in the Supporting Information, Table S2.
confirmed that the novel pyrazole urea Abl template retained
potent activity against all clinically relevant Abl1 mutants. In
addition, P38α/MAPK14 data were generated, as the pyrazole
unit was identified from vectors observed in pyrazole ureas
from P38 inhibitors. Gratifyingly, pyrazole ureas 23, 25, and 26
and the novel hinge-binder 24 were all selective for Abl1 over
P38α.53
SAR Follow-up. To cross-compare kinase inhibition data
generated on the microfluidic platform with that of a solid
sample, a number of key compounds were resynthesized (Table
4). These included compounds 23, 25, and 26 from the novel
pyrazole-urea series, along with compound 24 from the amide
series containing a novel hinge-binding heterocycle. This was
combined with a group of moderately potent (20−50 nM) and
weaker examples (>100 nM). In general, repeat data generated
on solid samples were in good agreement with data generated
via the microfluidic platform. Two exceptions were the
pyrimidine hinge-binder 27 that appeared to be Abl1 selective
on the platform but was confirmed as a weak and nonselective
compound on retest. Another exception was the pyrazole urea
28, which was predicted to be moderately active but was weak
when synthesized and screened on the platform. Retest of a
fresh sample indicated this was a relatively potent Abl inhibitor.
Compounds 23, 24, 25, and 26 were profiled against a
number of clinically relevant Abl1 kinase mutants, including
H396P, M351T, Q252H, T315I, and Y253F (Table 5). This
Compounds 23−26 were also screened in human liver
microsomal and cell permeability assays, to assess the broader
ADME properties of this series (Table 6).54
While membrane permeability for the ureas 23, 25, and 26 is
reduced when compared with amide 24, this is not entirely
unexpected due to the polar and solvated nature of ureas.
Compound 26 possessed the best membrane permeability of
the ureas, and this was also combined with the lowest in vitro
clearance in human liver microsomes. Further design and
optimization loops could subsequently look to address the
reduced permeability of the pyrazole ureas.
In summary, this program details a novel paradigm in hit-to-
lead and lead optimization. After the initial design of a virtual
chemical space and brief optimization of synthesis and
screening, a series of microfluidic synthesis-screening loops
were carried out in ∼90 min for each iteration. Incorporation of
each data point into the choice of subsequent analogue
synthesis and deliberately sampling across the virtual chemical
space resulted in a heat map visually remarkably similar to the
J
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
Table 4. Key Compound IC50 Data
K
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
Table 4. continued
a
b
c
d
Prediction after 90 synthesis to screening loops. Replicates n = 1. Replicates n = 4. Previously reported.52
Table 5. Activity of Resynthesized Compounds vs. WT and
Mutant Abl Kinases and P38α
4. EXPERIMENTAL SECTION
Chemistry. General Introductory Section. Identification, quanti-
tation (assay sample concentration and chemical yield are determined
using a calibrated evaporative light scattering detector (ELSD)), and
selection of compounds synthesized on the platform were performed
by analytical LC-MS-UV-ELSD analysis. This was conducted using the
following instrument and conditions. Waters Acquity BSD pump,
PolymerLabs 2100-ICE 385 ELS detector and Waters Acquity PDA
set at 254 nm with the MS detection performed on a Waters SQ mass
spectrometer with atmospheric pressure chemical ionization (APCI);
Phenomenex Luna 5 μm C18(2) 100 Å 150 × 4.6 mm column;
solvent A, water−0.1% formic acid; solvent B, acetonitrile−0.1%
formic acid; flow rate 1.5 mL/min; start 10% B with 0.2 min initial
hold, final 99% B in 8.8 min, linear gradient. Purity determination for
selected samples was performed using analytical high performance
liquid chromatography (HPLC) with the following instrument and
conditions. Gilson 322 pump, 155 detector, 819 injector, and Agilent
385 ELS detector; Phenomenex Gemini 5 μm C18 110 Å 100 × 4.6
mm column; solvent A, 20 mM ammonium acetate in water pH 7;
solvent B, acetonitrile; flow rate 1.5 mL/min; start 2% B, final 98% B
in 8 min, linear gradient. Compounds purified by reverse-phase high
performance liquid chromatography (RP-HPLC) used the following
instrument and conditions. Gilson 322 pump, 155 detector, 819
injector, and FC204 fraction collector; Phenomenex Luna 5 μm
C18(2) 100 Å, AXIA Packed, 150 × 21.2 mm column; solvent A,
water−0.5% formic acid; solvent B, acetonitrile−0.5% formic acid; flow
rate18 mL/min; start 10 to 40% B, final 50 to 98% B in 7 min, linear
gradient. Starting and final percentage of solvent B were on a per
compound basis. All microwave reactions were carried out in the
Biotage Initiator Sixty microwave reactor. NMR measurements were
performed on an Oxford Instruments 400 MHz NMR instrument
using CDCl3, CD3OD or DMSO-d6 as solvent. Chemical shifts are
reported in ppm and coupling constants (J) in Hz. Chemical shifts are
reported using solvent as internal standard. Final compounds for
biological assay were >95% purity as judged by HPLC and LC-MS
analysis as described above.
a
IC50 (nM)
23
24
25
26
Abl1 WT
H396P
1.6
0.7
0.9
1.6
1.9
0.9
0.6
0.3
0.3
0.5
0.4
0.3
1.0
0.3
0.5
1.0
0.8
0.4
1.0
0.2
0.4
0.7
0.4
0.4
M351T
Q252H
T315I
Y253F
P38α/MAPK14
35
43
35
24
a
Replicates n = 1.
Table 6. Data from Human Liver Microsomal and Cell
Permeability Assays
compound
a
23
24
25
26
HLM CLint (μL/min/mg protein)
76
90
26
51
29
2
b
PAMPA Papp (10−6 cm−1
)
0.1
0.2
a
b
Replicates n = 5. Replicates: 23 and 24, n = 3; 25, n = 2; 26, n = 4.
one at the end of the experiment for 270 compounds with only
22 compounds synthesized and quickly identified the pyrazole-
urea DFG-out motif as an unprecedented replacement for the
benzamide present in ponatinib. Further loops of design−
synthesis−screening refined the heat map and led to the
discovery of compounds 23, 24, 25, and 26, as novel and
potent Abl1 and Abl2 inhibitors. Further off-line screening
demonstrated that 23, 24, 25, and 26 possessed potent Abl
mutant activity, including the key T315I gatekeeper mutant.
Microfluidic synthesis, purification, and screening show great
potential for reducing the time required for the design−
synthesis−test loop at the core of preclinical drug discovery.
When this technology is combined with an activity prediction
algorithm and appropriate automated design strategies, the
technology platform can deliver extremely rapid optimization of
compound series and an accurate SAR heat map for the series.
Synthetic methods and analytical data for noncommercial
compounds in Tables 1, 2, and 3 are included in the Supporting
Information.
Experimental Section. General Experimental Procedure for
Sonogashira Couplings in Batch. To a 10 mL microwave reactor
tube fitted with a stirrer bar was added iodide (0.116 mmol), N-ethyl-
N-isopropylpropan-2-amine (0.081 mL, 0.464 mmol), alkyne (0.116
mmol), and DMF (4 mL). Nitrogen was bubbled through the mixture
for 2 min and tetrakis(triphenylphosphine)palladium(0) (3.35 mg,
L
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
2.90 μmol) and copper(I) iodide (0.828 mg, 4.35 μmol) were added.
The tube was sealed and heated to 100 °C in microwave for 45 min.
The resultant mixture was partitioned between dichloromethane (50
mL) and water (50 mL), the organics were dried (MgSO4) and
filtered, and the solvent was removed by evaporation to give a dark
brown residue. The residue was purified by preparative HPLC.
Analytical Data for Compounds in Table 4. 1-(5-(Imidazo[1,2-
a]pyrazin-5-ylethynyl)-1-methyl-1H-pyrazol-3-yl)-3-(4-((4-methylpi-
perazin-1-yl)methyl)-3-(trifluoromethyl)phenyl)urea (23) (5%). 1H
NMR (DMSO-d6): δ 9.4 (d, J = 12, 2H), 9.36 (s, 1H), 8.7 (d, J = 4,
1H), 8.1 (d, J = 4, 1H), 7.96 (s, 1H), 7.6 (m, 1H), 6.7 (s, 1H), 3.8 (s,
3H), 3.5 (s, 2H), 2.53 (m, 8H), 2.13 (s, 3H). 13C NMR (DMSO-d6):
δ 28.54, 37.57, 46.15, 53.62, 55.17, 55.83, 82.47, 88.44, 100.72, 100.08,
119.80, 122.03, 124.37, 130.55, 131.28, 131.80, 139.16, 140.61, 147.18,
152.29.
4-((4-(2-Hydroxyethyl)piperazin-1-yl)methyl)-N-(3-(imidazo[1,2-
a]pyridin-3-ylethynyl)-4-methylphenyl)-3-(trifluoromethyl)-
benzamide (32) (25%). 1H NMR (CDCl3): δ 8.35 (d, J = 8, 1H), 8.24
(s, 1H), 8.15 (s, 2H), 8.04 (d, J = 8, 1H), 7.91 (m, 2H), 7.90 (s, 1H),
7.65 (d, J = 12, 1H), 7.53 (dd, J = 4, 8, 1H), 7.29 (m, 2H), 6.96 (t, J =
4, 1H), 3.74 (s, 1H), 3.67 (t, J = 4, 2H), 2.65 (br m, 8H), 2.60 (br m,
3H), 2.54 (s, 2H).
MS: m/z 562 [M + H]+
4-((1H-Imidazol-1-yl)methyl)-N-(4-methyl-3-(pyrimidin-5-
ylethynyl)phenyl)-3-(trifluoromethyl)benzamide (33) (8.49%). 1H
NMR (DMSO-d6): δ 10.5 (s, 1H), 9.18 (s, 1H), 9.08 (s, 2H), 8.31 (s,
1H), 8.18 (d, J = 8, 1H), 8.02 (d, J = 4, 1H), 7.78 (s, 1H), 7.67 (dd, J =
8, 8, 1H), 7.33 (d, J = 8, 1H), 7.19 (s, 1H), 7.08 (d, J = 8, 1H), 6.98 (s,
1H), 3.15 (s, 2H), 2.45 (s, 3H)
MS: m/z 462 [M + H]+
(a) Biology Platform Kinase Assays: The Omnia kinase activity
assay technology was employed to monitor real-time kinase activity on
the platform. Full details are provided in Supporting Information.
MS: m/z 538.5 [M + H]+
3-(Imidazo[1,2-a]pyrazin-5-ylethynyl)-4-methyl-N-(4-((4-methyl-
piperazin-1-yl)methyl)-3-(trifluoromethyl)phenyl)benzamide (24)
1
(31%). H NMR (DMSO-d6): δ 10.53 (s, 1H), 9.19 (s, 1H), 8.65
ASSOCIATED CONTENT
* Supporting Information
(d, J = 4, 1H), 8.27 (m, 1H), 8.25 (s, 1H), 8.19 (m, 1H), 8.12 (d, J = 4,
1H), 8.03−8.06 (m, 1H), 7.93−7.95 (m, 1H), 7.68 (d, J = 8, 1H), 7.53
(d, J = 8, 1H), 3.15 (s, 2H), 2.60 (s, 3H), 2.37 (br, 8H), 2.16 (s, 3H).
13C NMR (DMSO-d6): δ 165.09, 143.88, 143.70, 138.58, 131.28,
129.09, 123.94, 121.86, 119.46, 119.45, 117.65, 98.39, 80.25, 57.87,
57.85, 55.11, 53.03, 46.05, 22.95, 21.02.
■
S
Detailed synthetic chemistry data, detailed biological methods,
prior knowledge data table, full SAR derived using the
automated platform. This material is available free of charge
via the Internet at http://pubs.acs.org.
MS: m/z 533 [M + H]+
1-(5-(Imidazo[1,2-a]pyrazin-3-ylethynyl)-1-methyl-1H-pyrazol-3-
AUTHOR INFORMATION
Corresponding Author
*E-mail: chris.selway@cyclofluidic.co.uk. E-mail: +44 (0)1707
358674.
yl)-3-(4-((4-methylpiperazin-1-yl)methyl)-3-(trifluoromethyl)-
■
1
phenyl)urea·formate salt (25) (5.2%). H NMR (DMSO-d6): δ 9.4
(s, 1H), 9.32 (s, 1H), 9.2 (s, 1H), 8.7 (d, 1H), 8.28 (d, 1H), 8.1(s,
1H), 7.6 (t, 1H), 6.71 (s, 1H), 3.88 (s, 3H), 2.39 (m, 8H), 2.21 (s,
3H). 13C NMR (DMSO-d6): δ 37.57, 45.61, 52.62, 54.26, 54.86,
57.10, 84.63, 87.92, 100.72, 109.98, 115.41, 122.03, 124.39, 126.16,
127.86, 130.39, 131.29, 131.86, 139.20, 140.63, 143.69, 147.10, 152.36.
MS: m/z 538.5 [M + H]+
Author Contributions
The manuscript was written through contributions of all
authors.
1-(5-(Imidazo[1,2-b]pyridazin-3-ylethynyl)-1-methyl-1H-pyrazol-
3-yl)-3-(4-((4-methylpiperazin-1-yl)methyl)-3-(trifluoromethyl)-
phenyl)urea·formate salt (26) (20.6%). 1H NMR (DMSO-d6): δ 9.37
(d, J = 4, 2H), 8.7 (d, J = 4, 1H), 8.25 (s, 2H), 7.93 (s, 1H), 7.6 (s,
2H), 7.4 (dd, J = 8, 4, 1H), 6.66 (s, 1H), 3.86 (s, 3H), 3.5 (s, 2H), 2.3
(m, 8H), 2.2 (s, 3H). 13C NMR (DMSO-d6): δ 37.32, 45.08, 52.14,
54.56, 57.58, 83.79, 86.49, 100.07, 111.37, 115.4, 119.85, 123.41,
126.14, 127.88, 130.26, 139.25, 140.33, 145.59, 152.27, 152.36, 164.24.
MS: m/z 538 [M + H]+
Notes
The authors declare the following competing financial
interest(s): Sandexis LLP act as medicinal and computational
chemistry consultants to Cyclofluidic Ltd; and Accelrys Ltd
provide consulting services to Cyclofluidic Ltd.
ACKNOWLEDGMENTS
■
We would like to thank Pfizer, UCB and Accelrys for their
technical support and advice, and Heptares Therapeutics is duly
acknowledged for the generous loan of their H-Cube
Hydrogenation Reactor. Funding is by Pfizer Ltd., Ramsgate
Road, Sandwich, Kent, CT13 9NJ, U.K. UCB Pharma, 208 Bath
Road, Slough, Berkshire SL1 3WE, UK. UK Technology
Strategy Board. The UK Technology Strategy Board is an
executive nondepartmental public body (NDPB), established
by the Government in 2007 and sponsored by the Department
for Business, Innovation and Skills (BIS).
4-((1H-Imidazol-1-yl)methyl)-N-(4-methyl-3-((5-(phenylamino)-
pyrimidin-2-yl)ethynyl)phenyl)-3-(trifluoromethyl)benzamide (27)
1
(6.45%). H NMR (CDCl3): δ 8.50 (s, 2H), 8.40 (m, 1H), 8.15 (m,
1H), 8.0 (s, 1H), 7.85 (m, 1H), 7.65 (m, 3H), 7.55 (m, 1H), 7.25−
7.35 (m, 6H), 7.1 5(s, 1H), 7.0 (m, 1H), 5.6 (s, 2H), 2.4 (s, 3H)
MS: m/z 553 [M + H]+
1-(5-(Imidazo[1,2-a]pyridin-3-ylethynyl)-1-methyl-1H-pyrazol-3-
yl)-3-(4-((4-methylpiperazin-1-yl)methyl)-3-(trifluoromethyl)-
phenyl)urea.formate salt (28) (16%). 1H NMR (DMSO-d6): δ 9.34
(d, J = 8, 1H), 8.6 (d, J = 8, 1H), 8.1 (s, 1H), 8.06 (s, 1H), 7.95 (s,
1H), 7.74 (d, 1H), 7.58 (m, 2H), 7.45 (m, 1H), 7.14 (m, 1H), 6.6 (s,
1H), 3.8 (s, 3H), 3.5 (s, 2H), 2.48 (m, 8H), 2.30 (s, 3H)
MS: m/z 537.5 [M + H]+
ABBREVIATIONS USED
BLAST, Basic Local Alignment Search Tool
■
1-(1-Methyl-5-(pyridin-3-ylethynyl)-1H-pyrazol-3-yl)-3-(4-((4-
methylpiperazin-1-yl)methyl)-3-(trifluoromethyl)phenyl)-
1
ureaformate salt (29) (8%). H NMR (DMSO-d6): δ 9.21 (d, J = 8,
REFERENCES
1H), 9.14 (d, J = 8, 1H), 8.8 (s, 1H), 8.6 (d, J = 4, 1H), 8.15 (s, 1H),
8.05 (d, J = 7.5, 1H), 7.58 (m, 1H), 7.49 (dd, J = 8, 4, 1H), 6.62 (s,
1H), 3.85 (s, 3H), 3.5 (s, 2H), 2.3 (m, 8H), 2.15 (s, 3H)
MS: m/z 498.5 [M + H]+
■
(1) Paul, S. M.; Mytelka, D. S.; Dunwiddie, C. T.; Persinger, C. C.;
Munos, B. H.; Lindborg, S. R.; Schacht, A, L. How to improve R&D
productivity: the pharmaceutical industry’s grand challenge. Nat. Rev.
Drug Discovery 2010, 9, 203−241.
1-(5-((6-Aminopyridin-3-yl)ethynyl)-1-methyl-1H-pyrazol-3-yl)-3-
(2) Kola, I.; Landis, J. Can the pharmaceutical industry reduce
attrition rates? Nat. Rev. Drug Discovery 2004, 3, 711−715.
(3) Rossi, T.; Braggio, S. Quality by Design in lead optimization: a
new strategy to address productivity in drug discovery. Curr. Opin.
Pharmacol. 2011, 11, 515−520.
(4-((4-methylpiperazin-1-yl)methyl)-3-(trifluoromethyl)phenyl)-
1
urea·formate salt (30) (5%). H NMR (DMSO-d6): δ: 9.1 (s, 1H),
8.15 (m, 1H), 7.91 (m, 1H), 7.55 (m, 2H), 6.44 (m, 2H), 3.80 (s, 3H),
3.5 (s, 2H), 2.3 (m, 8H), 2.2 (s, 3H) (5%)
MS: m/z: 513.5 [M + H]+
M
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
(4) Proudfoot, J. R. Drugs, leads, and drug-likeness: an analysis of
some recently launched drugs. Bio. Org. Med. Chem. Lett. 2002, 12
(12), 1647−1650.
Blackwood-Chirchir, M. A.; Iyer, V.; Chen, T.-T.; Huang, F.; Decillis,
A. P.; Sawyers, C. L. Dasatinib in imatinib-resistant Philadelphia
chromosome- positive leukemias. N. Engl. J. Med. 2006, 354, 2531−
2541.
(25) Tanaka, R.; Kimura, S. Abl Tyrosine Kinase Inhibitors for
Overriding Bcr-Abl/T3151: From the Second to Third Generation.
Expert Rev. Anticancer Ther. 2008, 8, 1387−1398.
(26) Huang, W. S.; Metcalf, C. A.; Sundaramoorthi, R.; Wang, Y. Z.;
Thomas, R. M.; Zhu, X.; Cai, L.; Wen, D.; Liu, S.; Romero, J.; Qi, J.;
Chen, I.; Banda, G.; Lentini, S. P.; Das, S.; Xu, Q.; Keats, J.; Wang, F.;
Wardwell, S.; Ning, Y.; Snodgrass, J. T.; Broudy, M. I.; Russian, K.;
Zhou, T.; Commodore, L.; Narasimhan, N. I.; Mohemmad, Q. K.;
Iuliucci, J.; Rivera, V. M.; Dalgarno, D. C.; Sawyer, T. K.; Clackson, T.;
Shakespeare, W. C. Discovery of 3-[2-(Imidazo[1,2-b]pyridazin-3-
yl)ethynyl]-4-methyl-N-{4-[(4-methylpiperazin-1-yl)methyl]-3-
(trifluoromethyl)phenyl}benzamide (AP24534), a Potent, Orally
Active Pan-Inhibitor of Breakpoint Cluster Region-Abelson (BCR-
ABL) Kinase Including the T315I Gatekeeper Mutant. J. Med. Chem.
2010, 53, 4701−4719.
(27) Carlson, R. H. Ponatinib effective in heavily pretreated CML
patients with T315I mutation. Oncology Times UK 2011, 8 (3), 17−18.
(28) http://www.fda.gov/Drugs/InformationOnDrugs/
ApprovedDrugs/ucm332368.htm
(29) Lu, X. Y.; Cai, Q.; Ding, K. Recent developments in the third
generation inhibitors of Bcr-Abl for overriding T315I mutation. Curr.
Med. Chem. 2011, 18, 2146−2157.
(30) Shults, M. D.; Imperiali, B. Versatile Fluorescence Probes of
Protein Kinase Activity. J. Am. Chem. Soc. 2003, 14248−14249.
(31) Kinase assay reagents were purchased from Life Technologies,
Paisley, Renfrewshire, UK
(5) Oprea, T. I.; Davis, A. M.; Teague, S. J.; Leeson, P. D. Is There a
Difference between Leads and Drugs? A Historical Perspective. J.
Chem. Inf. Comput. Sci. 2001, 41, 1308−1315.
(6) Clark, D. E.; Pickett, S. D. Computational methods for the
prediction of ‘drug-likeness. Drug Discovery Today 2000, 5 (2), 49−58.
(7) Leeson, P. D.; Springthorpe, B. The influence of drug-like
concepts on decision-making in medicinal chemistry. Nat. Rev. Drug
Discovery 2007, 6, 881−890.
(8) Free, S. M.; Wilson, J. W. A Mathematical Contribution to
Structure-Activity Studies. J. Med. Chem. 1964, 7 (4), 395−399.
(9) Craig, P. N. Structure-activity correlations of antimalarial
compounds 1. Free-Wilson analysis of 2-phenylquinoline-4-carbinols.
J. Med. Chem. 1972, 15 (2), 144−149.
(10) Schaper, K.-J. Free-Wilson-Type Analysis of Non-Additive
Substituent Effects on THPB Dopamine Receptor Affinity Using
Artificial Neural Networks. Quant. Struct.−Activity Relationships 1999,
18 (4), 354−360.
(11) Wong Hawkes, S. Y. F.; Chapela, M. J. V.; Montembault, M.
Leveraging the advantages offered by microfluidics to enhance the
drug discovery process. QSAR Comb. Sci. 2005, 24 (6), 712−721.
(12) Wong-Hawkes, S. Y. F.; Matteo, J. C.; Warrington, B. H.; White,
J. D. Microreactors as new tools for drug discovery and development.
Ernst Schering Foundation Symposium Proc. 2006, 3, 39−55.
(13) Guetzoyan, L.; Nikbin, N.; Baxendale, I. R.; Ley, S. V. Flow
chemistry synthesis of zolpidem, alpidem and other GABAA agonists
and their biological evaluation through the use of in-line frontal affinity
chromatography. Chem. Sci. 2013, 4, 764−769.
(14) Ley, S. V.; Baxendale, I. R. The Changing Face of Organic
Synthesis. CHIMIA Int. J. Chem. 2008, 62 (3), 162−168.
(15) Malet-Sanz, L.; Susanne, F. Continuous Flow Synthesis. A
Pharma Perspective. J. Med. Chem. 2012, 55 (9), 4062−4098.
(16) Watts, P. The Application of Microreactors in Combinatorial
Chemistry. QSAR Comb. Sci. 2005, 24, 701−711.
(17) Glasnov, T. N.; Kappe, C. O. Continuous-flow syntheses of
heterocycles. J. Heterocyclic Chem. 2011, 48 (1), 11−30.
(18) Webb, D.; Jamison, T. F. Continuous flow multi-step organic
synthesis. Chem. Sci. 2010, 1, 675−680.
(19) Bogdan, A. R.; James, K. Efficient Access to New Chemical
Space Through FlowConstruction of Druglike Macrocycles
Through Copper-Surface-Catalyzed Azide−Alkyne Cycloaddition
Reactions. Chem. Eur. J. 2010, 16, 14506−14512.
(20) Lange, P. P.; James, K. Rapid Access to Compound Libraries
through Flow Technology: Fully Automated Synthesis of a 3-
Aminoindolizine Library via Orthogonal Diversification. ACS Comb.
Sci. 2012, 14, 570−578.
(21) Druker, B. J.; Alpaz, M. T.; Resta, D. J.; Peng, B.; Buchdunger,
E.; Ford, J. M.; Lydon, N. B.; Kantajian, H.; Capdeville, R.; Ohno-
Jones, S.; Sawyers, C. L. Activity of a Specific Inhibitor of the BCR-
ABL Tyrosine Kinase in the Blast Crisis of Chronic Myeloid Leukemia
and Acute Lymphoblastic Leukemia with the Philadelphia Chromo-
some. N. Engl. J. Med. 2001, 344, 1031−1037.
(22) Shah, N. P.; Nicoll, J. M.; Nagar, B.; Gorrel, M. E.; Paquette, R.
L.; Kuriyan, J.; Sawyers, C. L. Multiple BCR-ABL kinase domain
mutations confer polyclonal resistance to the tyrosine kinase inhibitor
imatinib (STI571) in chronic phase and blast crisis chronic myeloid
leukemia. Cancer Cell 2002, 2 (2), 117−125.
(32) Zhou, T.; Commodore, L.; Huang, W.-S.; Wang, Y.; Thomas,
M.; Keats, J.; Xu, Q.; Rivera, V. M.; Shakespeare, W. C.; Clackson, T.;
Dalgarno, D. C.; Zhu, X. Structural Mechanism of the Pan-BCR-ABL
Inhibitor Ponatinib (AP24534): Lessons for Overcoming Kinase
Inhibitor Resistance. Chem. Biol. Drug Des. 2011, 77 (1), 1−11.
(33) Kluter, S.; Grutter, C.; Naqvi, T.; Rabiller, M.; Simard, J. R.;
̈
̈
Pawar, V.; Getlik, M.; Rauh, D. Displacement assay for the detection of
stabilizers of inactive kinase conformations. J. Med. Chem. 2010, 53
(1), 357−367.
(34) Berman, H. M.; Westbrook, J.; Feng, Z.; Gilliland, G.; Bhat, T.
N.; Weissig, H.; Shindyalov, I. N.; Bourne, P. E. The Protein Data
Bank. Nucleic Acids Res. 2000, 28, 235−242, www.rcsb.org.
(35) Deng, X.; Lim, S. M.; Zhang, J.; Gray, N. S. Broad spectrum
alkynyl inhibitors of T315I Bcr-Abl. Bioorg. Med. Chem. Lett. 2010, 20,
4196−4200.
(36) Zhang, Y.; Jamison, T. F.; Patel, S.; Mainolfi, N. Continuous
Flow Coupling and Decarboxylation Reactions Promoted by Copper
Tubing. Org. Lett. 2011, 13 (2), 280−283.
(37) Source for bromoheterocycles search was www.emolecules.com.
(38) Kearsley, S. J. An algorithm for the simultaneous superposition
of a structural series. J. Comput. Chem. 1990, 11, 1187−1192.
(39) Ponder, J. W. Tinker Molecular Modelling Package; Washington
University, St. Louis, MO.
(40) Shakespeare W. C.; Haluska F. G. WO2012139027, 2012.
(41) www.thalesnano.com
(42) Kikuchi, D.; Sakaguchi, S.; Ishii, Y. An Alternative Method for
the Selective Bromination of Alkylbenzenes Using NaBrO3/NaHSO3
Reagent. J. Org. Chem. 1998, 63 (17), 6023.
(43) Liaw, A.; Wiener, M. Classification and Regression by Random
Forest. R News 2002, 2 (3), 18−22.
(23) Kantarjian, H. J.; Giles, F.; Gattermann, N.; Bhalla, K.; Alimena,
G.; Palandri, F.; Ossenkoppele, G. J.; Nicolini, F.-E.; O’Brien, S. G.;
Litzow, M.; Bhatia, R.; Cervantes, F.; Haque, A.; Shou, Y.; Resta, D. J.;
Weitzman, A.; Hochhaus, A.; le Coutre, P. Nilotinib (formerly
AMN107), a highly selective BCR-ABL tyrosine kinase inhibitor, is
effective in patients with Philadelphia chromosome−positive chronic
myelogenous leukemia in chronic phase following imatinib resistance
and intolerance. Blood 2007, 110, 3540−3546.
(44) Breiman, L. Random Forests. Machine Learning 2001, 45, 5−32.
(45) McKay, M. D.; Conover, W. J.; Beckman, J. R. Comparison of
Three Methods for Selecting Values of Input Variables in the Analysis
of Output from a Computer Code. Technometrics 1979, 21 (2), 239−
245.
(46) Agrafiotis, D. K. Stochastic Algorithms for Maximizing
Molecular Diversity. J. Chem. Inf. Comput. Sci. 1997, 37, 841−851.
(24) Talpaz, M.; Shah, N. P.; Kantarjian, H.; Donato, N.; Nicoll, J.;
Paquette, R.; Cortes, J.; O’Brien, S.; Nicaise, C.; Bleickardt, E.;
N
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry
Article
(47) Agrafiotis, D. K.; Rassokhin, D. N. Design and Prioritization of
Plates for High-Throughput Screening. J. Chem. Inf. Comput. Sci. 2001,
41, 798−805.
(48) Brown, R. D.; Martin, Y. C. Designing Combinatorial Library
Mixtures Using a Genetic Algorithm. J. Med. Chem. 1997, 40, 2304−
2313.
(49) Gillet, V. J.; Khatib, W.; Willett, P.; Fleming, P. J.; Green, D. V.
S. Combinatorial Library Design Using a Multiobjective Genetic
Algorithm. J. Chem. Inf. Comput. Sci. 2002, 42, 375−385.
(50) Hopkin, M. D.; Baxendale, I. R.; Ley, S. V. An expeditious
synthesis of imatinib and analogues utilising flow chemistry methods.
Org. Biomol. Chem. 2013, 11, 1822−1839.
(51) Capdeville, R.; Buchdunger, E.; Zimmermann, J.; Matter, A.
Glivec (STI571, imatinib), a rationally developed, targeted anticancer
drug. Nat. Rev. Drug Discovery 2002, 1, 493−502.
(52) Zou, D.; Huang, W. S.; Thomas, R. M.; Romero, J. A. C.; Qi, J.;
Wang, Y.; Zhu, X.; Shakespeare, W. C.; Sundaramoorthi, R.; Metcalf,
C. A., III; Dalgarno, D. C.; Sawyer, T. K. Bicyclic Heteroaryl
Compounds. US Patent Application 2006/048758, December 22,
2006.
(53) Inhibition assays with ABL1 (H396P), ABL1 (M351T), ABL1
(Q252H), ABL1 (T315I), ABL1 (Y253F), ABL2/ARG and P38α/
MAPK14 were carried out by Reaction Biology Corporation, Malvern,
PA, USA.
(54) In vitro human microsomal stability and parallel artificial
membrane permeability assay (PAMPA) assays were carried out by
Cyprotex Discovery Ltd., Macclesfield, Cheshire, U.K.
O
dx.doi.org/10.1021/jm400099d | J. Med. Chem. XXXX, XXX, XXX−XXX