Structurally Diverse Mitochondrial Branched Chain Aminotransferase (BCATm) Leads with Varying Binding Modes Identified by Fragment Screening

Article abstract of DOI:10.1021/acs.jmedchem.5b01607

Inhibitors of mitochondrial branched chain aminotransferase (BCATm), identified using fragment screening, are described. This was carried out using a combination of STD-NMR, thermal melt (T_m), and biochemical assays to identify compounds that b

Full text of DOI:10.1021/acs.jmedchem.5b01607

Subscriber access provided by The University of British Columbia Library
Article
Structurally Diverse Mitochondrial Branched Chain Aminotransferase
(BCATm) Leads With Varying Binding Modes Identified by Fragment Screening
Jennifer A Borthwick, Nicolas Ancellin, Sophie Marie Bertrand, Ryan P Bingham,
Paul S Carter, Chun-Wa Chung, Ian Churcher, Nerina Dodic, Charlene Fournier,
Peter L Francis, Andrew Hobbs, Craig Jamieson, Stephen D. Pickett, Sarah E
Smith, Donald O. Somers, Claus Spitzfaden, Colin J. Suckling, and Robert J Young
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.5b01607 • Publication Date (Web): 03 Mar 2016
Downloaded from http://pubs.acs.org on March 10, 2016
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Journal of Medicinal Chemistry is published by the American Chemical Society. 1155
Sixteenth Street N.W., Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.

Page 1 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Structurally Diverse Mitochondrial Branched Chain Aminotransferase
(BCATm) Leads With Varying Binding Modes Identified by Fragment
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Screening
*1,3
2
1,3
1
Jennifer A Borthwick , Nicolas Ancellin , Sophie M Bertrand , Ryan P Bingham , Paul S
1
1
1
2
1
1
Carter , Chunꢀwa Chung , Ian Churcher , Nerina Dodic , Charlène Fournier , Peter L Francis ,
1
3
1
1
1
Andrew Hobbs , Craig Jamieson , Stephen D Pickett , Sarah E Smith , Donald O’N Somers ,
1
3
1
Claus Spitzfaden , Colin J Suckling , Robert J Young .
1
GlaxoSmithKline R&D, Medicines Research Centre, Gunnels Wood Road, Stevenage,
Hertfordshire, SG1 2NY, UK.
2
GlaxoSmithKline R&D, Les Ulis, Centre de Recherche, 25,27 Avenue du Québec, 91140
Villebon sur Yvette, France.
3
Department of Pure and Applied Chemistry, University of Strathclyde, 295 Cathedral Street,
Glasgow, G1 1XL.
Abstract
Inhibitors of mitochondrial branched chain aminotransferase (BCATm), identified using
fragment screening, are described. This was carried out using a combination of STDꢀNMR,
thermal melt (T ) and biochemical assays to identify compounds that bound to BCATm, which
m
were subsequently progressed to Xꢀray crystallography, where a number of exemplars showed
significant diversity in their binding modes. The hits identified were supplemented by searching
ACS Paragon Plus Environment
1

Journal of Medicinal Chemistry
Page 2 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
and screening of additional analogues, which enabled the gathering of further Xꢀray data where
the original hits had not produced liganded structures. The fragment hits were optimised using
structureꢀbased design, with some transfer of information between series, which enabled the
identification of ligand efficient lead molecules with micromolar levels of inhibition, cellular
activity, and good solubility.
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Introduction
Mitochondrial branched chain aminotransferase (BCATm) is a metabolic enzyme which
converts the branched chain amino acids (BCAAs), leucine, isoleucine and valine into their
corresponding αꢀketo acids. Over the past few years, evidence has appeared linking this enzyme
to obesity in animals and humans; accordingly inhibition of this enzyme has potential as a
treatment for obesity and related diseases. The BCATm isoform is expressed ubiquitously, with
very high levels in the stomach, pancreas and salivary glands, consistent with its role in protein
1
metabolism. Interestingly, BCATm knockꢀout mice are resistant to highꢀfat diet induced obesity
2
and diabetes; these effects are believed to be caused by the establishment of an energy
consuming cycle of protein synthesis and degradation. This phenotype is similar to that observed
in humans who consume high protein diets or BCAA supplements and experience increased
2
energy expenditure, lower appetite and ultimately, lower body weights. The role of BCATm and
BCAAs in the aetiology of obesity is complex and not fully understood; knockꢀout and dietary
studies discussed above, where high levels of BCAA appear to be associated with reduced risk of
becoming obese, are complicated by studies of obese animals. Genetically mutated obese rodents
and dietꢀinduced obese mice and humans have all been shown to have reduced BCATm
3
expression, resulting in higher BCAA levels than lean animals.
ACS Paragon Plus Environment
2

Page 3 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
The search for inhibitors of BCATm at GSK was pursued using three complementary methods;
fragment screening, high throughput screening (HTS) and encoded library technology. This
paper describes the diversity of chemical matter, displaying distinct binding modes discovered
by fragment screening. Hybridization of one of these hits with an HTS hit led to a potent in vivo
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
4
5
active series and a further series was identified by encoded library technology. Together, this
package of work has shed much light on the variable binding modes observed with BCATm and
is beginning to shed some light on the effects of specific pharmacological inhibition of BCATm
by small molecules and the role of this enzyme in disease.
6
7
In a commentary on our Fragment Assisted Drug Design paper, the difficulty of the target
was highlighted, as was the importance of having diverse chemical matter with appropriate in
vivo activities to interrogate a target; this paper illustrates structural information describing
diverse binding modes observed. In fragment screening, the lack of overlap between screening
8
methods has been acknowledged ; this was observed in the progression of hits identified by the
various methods in this fragment screen, perhaps compounded by the diversity of binding modes.
The fragment hits were clustered according to their distinct binding modes. Importantly, this
study provides new structural understanding of this enzyme, with several binding pockets
4,5
identified by the fragment screen, which were not occupied by either of the other series .
Additionally, further optimisation efforts on two of the fragment hit series are described, which
represent useful starting points and give important structural pointers to enable the identification
of structurally novel tools for further pharmacological studies.
Results and Discussion
Fragment Screening
ACS Paragon Plus Environment
3

Journal of Medicinal Chemistry
Page 4 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
FBDD techniques have become an integral part of the drug discovery process during the last
7
,9,10
decade.
Standard components of the FBDD methodology include the use of biophysical
screening and Xꢀray crystallography to enable the identification, optimisation and designed
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
11
growth of inherently weakly binding ligands into potent lead molecules. Commonly, these
methods have used a single form of the protein, for example, the apo or a cofactor bound state.
However, during enzyme turnover in solution, the dynamic changes in protein conformation
and biochemical state present a number of potential binding opportunities. The use of a single
form for screening in a binding assay may restrict the type of compounds that can be detected
and characterised. In the extreme case, this arbitrary choice may preclude sampling the most
tractable form of the enzyme. During the catalytic cycle (Scheme 1) PLP and PMP cofactor
bound forms of BCATm are sampled. Pragmatically we therefore used both PLP and PMP forms
in our studies where possible, as these can both be stably isolated.
At the start of the cycle, the cofactor pyridoxal phosphate (PLP) is covalently bound to the
1
2
enzyme via a lysine residue (Lysꢀ202) (Scheme 1). This lysine is then displaced by the αꢀ
amine of an incoming BCAA substrate. The imine formed is hydrolysed to the corresponding αꢀ
keto acid. In the course of the reaction, the coꢀfactor is released from its covalent bond to the
enzyme as pyridoxamine phosphate (PMP). A complementary halfꢀreaction occurs to complete
the cycle, whereby αꢀketoglutarate is transaminated to glutamate and the PLP form of the
enzyme is regenerated. It was possible to characterise both the PLP and PMP forms of the
enzyme (e.g. spectroscopically or by mass spectrometry) and to interconvert between these using
an excess of either leucine or αꢀketo glutarate as appropriate.
ACS Paragon Plus Environment
4

Page 5 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Scheme 1: Reaction cycle catalysed by BCATm
A functional biochemical assay was designed to detect inhibitors of this pathway, making use
of a coupled sequence of enzymes (Figure 1), whereby the glutamate product of BCATm activity
was reꢀoxidised to αꢀketo glutarate along with hydrogen peroxide by the enzyme glutamate
oxidase. The peroxide thus generated was detected by reaction with 1ꢀ(3,7ꢀdihydroxyꢀ10Hꢀ
phenoxazinꢀ10ꢀyl)ethanone to form resorufin, catalysed by the enzyme horseradish peroxidase.
Decoupled parallel control wells, without the BCATm initiated step were added to detect any
inhibitors of the coupled processes. A potential risk of carrying out fragment screening using
such a coupled enzyme system was that genuine inhibitors of BCATm could be discarded as
nonꢀspecific in nature due the higher chances of fragments interacting with more than one
ACS Paragon Plus Environment
5

Journal of Medicinal Chemistry
Page 6 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
13
enzyme. Theoretical reasoning based on complexity, on which much of FBDD methodology is
predicated, suggest that there is a higher probability of smaller molecules binding to any given
protein than for larger molecules. Therefore, it is more likely that fragments binding to BCATm
may also bind to the coupled enzyme glutamate oxidase than would be the case if this assay was
used for a traditional HTS campaign with larger, more complex molecules. However, using this
functional biochemical assay had the advantage that all biochemical forms of the protein
necessary for reaction could be sampled, permitting the detection of inhibitors that inhibited only
one of these forms.
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 1: Biochemical assay format
As is common in fragment screening campaigns, complementary biophysical screening was
also carried out to detect compounds which bound to BCATm; in this case NMR and thermal
1
4
shift techniques were used. The saturation transfer difference (STD) NMR assay used a mixture
of the PLP and PMP forms of the enzyme. Hits identified may interact with either, or potentially
ACS Paragon Plus Environment
6

Page 7 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
both, the PLP and PMP forms of the protein. However, as both forms are catalytically relevant
the precise binding profile was not initially determined as all hits had the potential to yield
functional inhibitors. The STDꢀNMR approach was complemented by a thermal shift screen
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
15
(
T ), where both forms of the enzyme were independently assayed. This was necessary
m
because of the very different denaturation temperatures of the PLP(E) and PMP(F) bound
proteins, 73 °C and 54 °C, respectively.
The GSK core fragment library of 1056 compounds was screened in parallel through all three
assays and hit rates are shown in Figure 2. The crystallography was carried out by soaking high
concentration solutions (40 ꢀ 60 mM) of the ligand of interest into preformed crystals of both
PMP and PLP forms of the protein for one to fourteen days, sometimes yielding liganded crystals
that gave datasets sufficient to unambiguously determine binding modes (resolutions in the range
1
.8 ꢀ 2.4 Å). Although crystals of both the PLP and PMP forms of the BCATm were tried, the
initial hits only yielded complexes with the more stable PLP protein. We therefore took the
pragmatic decision to prioritise inhibitors for which crystal structures could be elucidated,
thereby concentrating efforts on inhibitors of the PLP form of BCATm.
ACS Paragon Plus Environment
7

Journal of Medicinal Chemistry
Page 8 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 2: Flow and Venn diagrams illustrating the output of the fragment screening
campaign, showing number of hits identified by each screening technique and the logic
ACS Paragon Plus Environment
8

Page 9 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
employed for X-ray prioritisation, plus the distribution of those which yielded X-ray
crystal structures.
The output of the fragment screening is summarised in Figure 2. The different screening
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
8
techniques produced different sets of hits, which is a common observation in FBDD campaigns ,
as screening is carried out close to the limits of detection for the assays and different
experimental conditions may favour a particular target conformation. All of the hits from either
of the biophysical assays were progressed to Xꢀray crystallography. The STDꢀNMR experiment
gave the better success rate in identifying apparent binders that led to liganded Xꢀray structures,
with the orthogonal thermal shift experiment showing a relatively poor prediction of success.
The biochemical assay had yielded a high number of specific hits, which could not all be
progressed to crystallography, so the STDꢀNMR data was used to triage these hits, using a lower
cutꢀoff (2 × signal to noise, versus 4 × signal to noise) than was used in the initial selection of
NMR hits.
Additional activities were carried out in an effort to increase the number of liganded Xꢀray
structures as potential starting points for chemical optimisation. Five compounds identified were
hits in both T and NMR assays but failed to produce crystal structures. Given reasonable
m
confidence that these compounds were genuinely binding to BCATm, structural analogues were
sourced from the GSK compound collection and screened in the STDꢀNMR and biochemical
assays. Analogue searching was performed by focussed substructure searching complemented by
a GSKꢀdeveloped workflow, FindAnalogues, specifically designed for fragment followꢀup.
FindAnalogues combines substructure and similarity searching methods with different
descriptors and constraints on compound properties to identify suitable fragment like analogues
from both inꢀhouse and external vendors (see Experimental for details). 80 compounds were
ACS Paragon Plus Environment
9

Journal of Medicinal Chemistry
Page 10 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
selected from this exercise and screened. 32 compounds were identified with activity in one of
the two assays, from which three yielded liganded structures where the original fragment hit had
not (12, 13, 15). An additional 120 similarly selected analogues were subsequently screened and
yielded valuable SAR for compounds 4 and 5 as described below.
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Concurrent with this fragment screen, a high throughput screen of the GSK HTS compound
collection (1.7 million compounds) had been conducted. The singleꢀshot percentage inhibition
data and subsequent XC data were interrogated for compounds that were identified by the
5
0
FindAnalogues search. Four analogues of 1 were identified with micromolar activity in the HTS
and were submitted to crystallography, yielding additional liganded structures 9 and 10.
ACS Paragon Plus Environment
10

Page 11 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Fragment Hits
Compound
Heavy
Structure
pIC50
clogP
LE
LLEAT
Source
Binding Mode
Number
Atoms
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
PLP lipophilic pocket,
Hydrogen bond acceptor
to Alaꢀ314
1
2
3
<
<
<
2.9
2.9
2.9
a
0.78
13
STDꢀNMR
PLP lipophilic pocket,
Hydrogen bond acceptor
to Alaꢀ314
T and
m
1.25
1.57
15
15
STDꢀNMR
PLP lipophilic pocket,
Hydrogen bond acceptor
to Alaꢀ314
STDꢀNMR
STDꢀNMR
and
πꢀ stacking to Pheꢀ30,
4
4
.2
0.62
15
0.38
0.27
0.43
0.26
Hydrogen bond acceptor
biochemical to Lysꢀ79
STDꢀNMR
5
6
πꢀ stacking to Pheꢀ30,
b
3.0
1.31
1.96
15
11
and
PLP lipophilic pocket
biochemical
Interface between
protein dimers (no
functional effect)
<2.9
STDꢀNMR
ACS Paragon Plus Environment
11

Journal of Medicinal Chemistry
Page 12 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
PLP lipophilic pocket,
STDꢀNMR
and
Hydrogen bond acceptor
to Alaꢀ314,
7
3
.1
ꢀ0.2
13
0.33
0.45
biochemical Displacement of water
adjacent to Cysꢀ315
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
STDꢀNMR
8
πꢀ stacking to Pheꢀ30,
and
PLP lipophilic pocket
biochemical
3.9
1.6
14
0.38
0.33
HTS hit with
9
similarity to
fragment
hits
Hydrogen bond acceptor
to Alaꢀ314
6
.6
3.8
18
0.48
0.30
HTS hit with
similarity to
fragment
hits
1
0
5
.1
2.69
18
0.39
0.29
πꢀ stacking to Pheꢀ30
Tm, STDꢀ
1
1
1
2
No crystal structure
obtained
c
3
.3
0.21
16
15
0.28
0.40
0.41
0.50
NMR and
biochemical
PLP lipophilic pocket,
Displacement of water
adjacent to Cysꢀ315
<3.2
Analogue of
0
.14
d
1
1
ACS Paragon Plus Environment
12

Page 13 of 60
Journal of Medicinal Chemistry
1
2
3
4
PLP lipophilic pocket,
Displacement of water
adjacent to Cysꢀ315
1
3
Analogue of
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
4
.6
0.77
19
0.44
0.45
1
1
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Tm, STDꢀ
1
1
4
5
No crystal structure
obtained
3
3
.8
.8
1.9
14
15
0.37
0.35
0.29
0.26
NMR and
biochemical
PLP lipophilic pocket,
Hydrogen bond acceptor
to Alaꢀ314
Analogue of
2.08
1
4
a
Table 1 Summary of fragment hits identified. compound reported inactive (pIC < 4.2) on 3
5
0
b
out of 35 test occasions; compound reported inactive (pIC < 2.9) on 1 out of 2 test occasions;
5
0
c
d
compound reported inactive (pIC < 2.9) on 1 out of 3 test occasions; compound reported
50
inactive (pIC < 3.2) on 2 out of 4 test occasions.
5
0
Table 1 summarises the fragment hits identified which yielded liganded crystal structures, along
1
6
with their calculated ligand efficiency (LE) and modified lipophilic ligand efficiency (LLE )
AT
17
as defined by researchers at Astex. The crystal structures obtained for the hits, showed
occupation of a range of positions in the BCATm binding site, and that significant movement in
the protein was induced by the compounds. Some of the most important features for binding are
summarised in the Xꢀray crystal structures shown in Figure 3.
ACS Paragon Plus Environment
13

Journal of Medicinal Chemistry
Page 14 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 3: X-ray crystal structures of compounds 1(cyan, PDB entry 5I5S), 4(magenta
4
PDB entry 5I5V), 5(pink PDB entry 5I5W) and 12(orange PDB entry 5BWT ) in the
ACS Paragon Plus Environment
14

Page 15 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
BCATm binding site in panels a, b, c and d respectively. Panel e shows an overlay of all
these compounds within the binding site. Panel f shows the same overlay but with Tyr-173
in green stick format, highlighting the mobility of the 170-180 loop region.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Benzo[d]isoxazole 1 (Figure 3a) occupied a binding pocket defined by a lipophilic region
adjacent to the PLP cofactor, which is confined by a closed ‘gate’ between residues Pheꢀ30 and
Tyrꢀ173. The carboxylic acid carbonyl group acted as a hydrogen bond acceptor to the backbone
NH of Alaꢀ314.
The crystal structure of thienopyrimidine 4 (Figure 3b) showed occupation of a distinct
binding site. The site adjacent to the cofactor was occupied only by ethylene glycol (present in
crystallography medium) and the Pheꢀ30 Tyrꢀ173 ‘gate’ had opened to allow the formation of a
πꢀstacking interaction between Pheꢀ30 and the heterocyclic core. The opening of the gate
involved significant conformational rearrangement, with Pheꢀ30 moving by around 2 Å, whilst
Tyrꢀ173 moved by approximately 6 Å. Additionally, the amide carbonyl acted as a hydrogen
bond acceptor to a lysine NH (Lysꢀ79).
The pyrimidine ring of the biaryl amide 5 (Figure 3c) formed a similar πꢀstacking interaction
with Pheꢀ30 to thienopyrimidine 4, whilst the phenyl ring occupied a similar lipophilic aromatic
pocket to benzo[d]isoxazole 1. The tyrosine residue, which moved to open the induced pocket in
the structure of 4 (Tyrꢀ173) was completely disordered in this structure. There are no direct
hydrogen bonding interactions between the heteroatoms of the fragment and the protein, which
may explain the poor ligand efficiency observed. However the amide carbonyl is well solvated
and there is a water mediated hydrogen bond to the hydroxyl of Tyrꢀ141. Biaryl amide 5 was an
ACS Paragon Plus Environment
15

Journal of Medicinal Chemistry
Page 16 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
effective bridge between the two binding sites identified, and demonstrated that it is possible to
have both sites occupied simultaneously.
4
Pyrazolopyrimidine 12 (Figure 3d PDB entry 5BWT ) displayed an additional distinct binding
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
mode. The pendant ethyl group was close to the lipophilic site occupied by aromatic groups for
compounds 1 and 5, and the ‘gate’ was closed. The heterocyclic core was positioned in a site
which was unoccupied in the other structures, with the nitrile substituent displacing a water
molecule, which formed a hydrogen bonding interaction with the NH of Cysꢀ315 in other
structures. The pyridone carbonyl forms hydrogen bonds with Tyrꢀ173 and the backbone of Valꢀ
155. The pyridone NH and pyrazole N are both involved in water networks.
When all of these liganded crystal structures of the fragment hits were considered together
(Figure 3e and f), it was clear that the BCATm protein has considerable flexibility to
accommodate a range of ligand binding modes. However, there were several binding ‘hotꢀspots’
identified, which appeared to act as anchors for the fragments. The most notable of these was the
lipophilic pocket adjacent to the PLP, hydrogen bonding with Alaꢀ314, πꢀstacking with Pheꢀ30
and displacement of the water molecule adjacent to Cysꢀ315. All of the hits identified satisfied at
least one of these interactions (see Table 1). This flexibility in the protein would thus have
caused difficulties in attempting to predict the binding modes of hits using docking methods
alone. Therefore, to enable structure based fragment growth, it was important to generate
multiple Xꢀray crystal structures (as is common practice), even for compounds which appeared
to be closely structurally related. The diversity of structural information obtained provided
1
8
opportunities to grow the fragment hits in various ways .
Hits to Leads Optimisation
ACS Paragon Plus Environment
16

Page 17 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
The fragment hits identified were grouped (“clustered”) according to their binding modes in
the BCATm active site. An example of this process is shown for compounds 1, 2 and 3, which
look dissimilar on the basis of their chemical structures, but which all make similar binding
interactions with the protein (Figure 4). One exemplar was chosen from each vectorial cluster,
either because it demonstrated superior efficiency of binding, or because it offered more flexible
and tractable chemistry than other structures within the same cluster. The fragment growth and
further optimisation of the pyrazolopyrimidinone series represented by compounds 11-13 is
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
4
described in the complementary paper; a noteworthy feature of this series was that the original
fragment hit (11) did not give a crystal structure, but analogue searching identified closely
related compounds that did yield liganded crystal structures. Herein, the initial lead optimisation
efforts on two other series to illustrate the strategies employed are described.
Figure 4: Example of vectorial clustering, showing an overlay of compounds 1 (cyan PDB
entry 5I5S), 2(green PDB entry 5I5T) and 3(beige PDB entry 5I5U) with similar binding
interactions.
ACS Paragon Plus Environment
17

Journal of Medicinal Chemistry
Page 18 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
Thienopyrimidine 4 had very favourable LE and LLE values and contained a tractable
AT
functional group (carboxylic acid), which could be used to grow towards the PLP pocket. The
crystal structure of biaryl amide 5 provided further confidence in this approach because it
showed that concomitant binding in both of these pockets could be achieved. With these aims in
mind, initial optimisation focussed on a small array of amides, mostly containing lipophilic
substituents, which were designed to fill the PLP pocket.
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
These compounds were synthesised using HATUꢀmediated coupling chemistry or via acylation
using the appropriate acid chloride (Table 2). Due to monomer availability, some of the
chemistry was carried out on the closely related desꢀmethyl series e.g. 16, which showed similar
activity to the initial hit 4. Analogues 17 and 18 confirmed that activity with either of these two
substituents was comparable. In general, aromatic amides were preferred to aliphatic
substituents, as demonstrated by the comparisons between compounds 17 and 18 plus 19 and 20.
Substitution on the pendant phenyl ring was only tolerated in the ortho position (compounds 21,
2
2, 23). Attempts to accommodate heteroaromatic rings in the lipophilic pocket were not
productive (e.g. compounds 24 and 25), with the exceptions of the lipophilic thiophene (26) and
thiadiazole (27), both of which also maintained the highly favourable LE and LLEAT values of
the hit.
ACS Paragon Plus Environment
18

Page 19 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 5: Overlay of compounds 27 (purple PDB entry 5I5X) and 29 (pale green PDB entry
5I5Y). Panel a. shows the common interactions with Lys-79, Phe-30 and Ala-314. Panel b.
illustrates the difference in interactions around PLP/PMP for the two compounds
The potency associated with compound 27 could not be rationalised by knowledge of the
crystal structure of compound 4. Soaking experiments failed to produce a crystal structure of
compound 27, however coꢀcrystallisation was successful, which yielded the first liganded
structure of BCATm in the PMP form (Figure 5). The PMP form shows a few significant
differences with respect to the PLP form. Firstly, the Lysꢀ202 sidechain, now no longer
covalently linked to cofactor, as in the PLP form, is free to shift slightly to optimally interact
with one of the nitrogen atoms in the thiadiazole ring. Secondly, the PMP cofactor pyridine
rotates by ~11° from the PLP pyridine plane and Leuꢀ266 adopts a lower energy rotamer
conformation due to the additional space created by the rotation of the PMP pyridine ring. Due
to its potent inhibition of BCATm, compound 27 was progressed to a cellular assay measuring
BCATm inhibition, and was found to have a cellular pIC value of 5.4.
5
0
ACS Paragon Plus Environment
19

Journal of Medicinal Chemistry
Page 20 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Compound
Number
R1
CH3
H
R2
pIC50
LE
LLEAT
0.44
4
a
OH
OH
4.2
4.3
5.1
5.1
0.38
16
0.44
0.33
0.36
0.51
1
1
1
2
7
8
9
0
CH3
0.34
H
0.37
CH3
< 4.2
< 3.2
CH3
N
2
1
b
H
4.3
0.28
0.30
ACS Paragon Plus Environment
20

Page 21 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
2
2
3
H
H
< 4.2
< 4.2
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
2
2
2
2
4
5
6
H
H
3.6
0.25
0.33
< 4.2
H
H
5.8
6.7
0.48
0.48
0.45
0.63
27
Table 2: Activity and efficiency data for amide analogues of compound 4.
a
b
compound reported inactive (pIC < 4.2) on 3 out of 35 test occasions; compound reported
5
0
inactive (pIC < 4.2) on 2 out of 4 test occasions.
5
0
The next logical expansion was to grow towards Alaꢀ314 to try to exploit a hydrogenꢀbonding
opportunity, which had been identified from the fragment screening as a binding hotꢀspot. A
ACS Paragon Plus Environment
21

Journal of Medicinal Chemistry
Page 22 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
range of heteroatomꢀcontaining functional groups were introduced in the orthoꢀposition of
compound 18 (Table ). From this effort, only the aryl acetate 29 showed any significant
improvement in BCATm inhibition resulting in micromolar activity, whilst maintaining good
efficiencies. Crystallisation of this compound in BCATm (Figure 5) confirmed that the acid
carbonyl moiety was acting as a hydrogen bond acceptor to the backbone NH of Alaꢀ314.
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Compound
R
H
pIC50
5.1
LE
LLEAT
0.37
Number
18
0.36
28
< 4.2
6.0
2
3
9
0
0.34
0.45
5
.0
.9
0.31
0.28
0.44
0.41
4
31
ACS Paragon Plus Environment
22

Page 23 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
32
< 4.2
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 3: Results of targeting Ala-314 through growth from the ortho-position of 18
By using the structural information obtained from different fragment hits to identify binding hot
spots it was thus possible to grow weakly active fragment 4 into more potent analogues, with
activity in a cellular assay.
Analysis of structures showed that biaryl amide 5 offered potential vectors towards the Alaꢀ
314 NH donor and the Lysꢀ79 NH donor, to which the pyrimidinone carbonyl of 4 was
2
hydrogen bonded. The biaryl amide substructure is an abundant motif in the GSK compound
collection, so the initial SAR was readily delineated (Table 4).
Compound Structure
pIC50
LE
LLEAT clogP
5
3
.0
0.27
0.26
1.31
2.65
33
<
3.2
ACS Paragon Plus Environment
23

Journal of Medicinal Chemistry
Page 24 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
3
4
5
3
.7
.5
0.27
0.25
0.26
0.25
1.66
1.58
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
3
a
3
36
5.6
0.39
0.36
0.31
0.26
2.86
3.00
37
38
5.3
5
.7
0.33
0.32
1.94
Table 4: Screening results from analogues of 5 available in the GSK collection.
a
compound reported inactive (pIC < 4.2) on 2 out of 4 test occasions.
5
0
ACS Paragon Plus Environment
24

Page 25 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
The pyrimidine ring was initially replaced with phenyl, as a greater number of analogues were
available, and the crystal structure of 5 suggested that the pyrimidine nitrogens were not
participating in any binding interactions. As with the optimisation of 4, it was found that adding
hydrogen bond acceptors, which could interact with Alaꢀ314, provided modest improvements in
binding (34 and 35), with maintained ligand efficiency. A more dramatic improvement in
enzyme inhibition was achieved when hydrogen bond acceptors, represented by carboxamide
and pyrazole (36 and 37), enabled engagement with Lysꢀ79. Carboxamide 36 also had additional
lipophilic chlorine groups that enhanced binding in the PLP pocket, contributing to micromolar
levels of activity, which represented a roughly 1000ꢀfold increase in activity from the fragment
hit. This compound also displayed a significant improvement in the efficiency measures relative
to the hit compound 5. Synthesis of compound 38, which combined the acceptors targeting both
Lysꢀ79 and Alaꢀ314 demonstrated that the effects on the activity could be additive, with good LE
and LLEAT values achieved. The intent of this design was confirmed by the Xray structure of 38
within BCATm shown in Figure 6.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 6: Overlay of compounds 38 (blue PDB entry 5I60) and 5 (pink PDB entry 5I5W)
confirms 38 makes additional interactions with Lys-79 and Ala-314.
ACS Paragon Plus Environment
25

Journal of Medicinal Chemistry
Page 26 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
Once a good understanding of the SAR within this series had been established through
screening of available analogues, attention was turned to improving the physicochemical profiles
1
9
of the compounds. Biaryl amides are notoriously insoluble, being highly planar and crystalline,
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
and indeed, the measured solubilities of these compounds were poor (Table , compounds 37 and
39). Two strategies were adopted to improve the solubility within the series. The amide NꢀH was
constrained in a tetrahydroisoquinoline ring to prevent intermolecular hydrogen bonding, which
could contribute to the high crystallinity (40 and 42). Additionally, the pyrimidine ring from the
initial hit was returned to the molecules in place of the phenyl ring to lower the lipophilicity
(
compound 41). The targeted analogues were synthesised as shown in Scheme 2. Both of these
strategies resulted in a modest improvement in enzyme inhibition and approximately 10ꢀfold
increased solubility. As in the optimisation of thienopyrimidine 4, the biaryl amide series had
been grown into a series of leadꢀlike molecules through focussed structureꢀbased design
combined with consideration of molecular properties.
Compound Structure
pIC50
LE
LLEAT clogP Sol(µg/ml)
3
4
9
0
a
4
.3
.8
0.33
0.33
0.36
1.46
1.14
13
4
0.33
132
ACS Paragon Plus Environment
26

Page 27 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
3
4
4
7
1
2
5.2
0.36
0.38
0.36
0.26
0.35
0.29
3.00
1.94
2.83
12
82
94
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
5
.5
.8
5
Table 5: Data from the biaryl amides series, illustrating structural modifications
a
designed to improve solubility. compound reported inactive (pIC < 4.2) on 1 out of 4 test
5
0
occasions
ACS Paragon Plus Environment
27

Journal of Medicinal Chemistry
Page 28 of 60
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Reagents and conditions: (a) HATU, DIPEA, DMF, 20 – 40 °C, 41ꢀ79 %; (b) 1Hꢀpyrazolꢀ3ꢀ
ylboronic acid, Pd(PPh ) , Na CO (sat. aq.), 130 °C, microwave 20 min. 41 – 72 %; (c) CuI,
3
4
2
3
K CO , DMF, 160 °C, 18 h, 57 – 66%; (d) 2M NaOH (aq.), 30 % H O (aq.), EtOH, 50 °C, 20 h,
2
3
2
2
1
9 %.
Scheme 2: Synthesis of analogues of compound 5
Conclusions
When carrying out any diversity screening campaign, but particularly when utilising fragmentꢀ
based approaches, it is important to realise that different biochemical and biophysical screening
techniques can, and indeed do, identify different subꢀsets of hits. Therefore, there could be an
ACS Paragon Plus Environment
28

Page 29 of 60
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
advantage to using multiple, complementary techniques for the initial screening phase. This is
particularly important in FBDD, where the binding of hits will likely to be near to the limit of
detection in some assays. This set of hits identified for BCATm, and the crystal structures
generated have illustrated many aspects of the binding of small molecules to this enzyme for the
first time. It was particularly interesting to observe such different conformations of the protein
between different fragment hits, whilst gratifying to be able to identify several binding ‘hot
spots’ in the protein structure and confirm that these could be accessed simultaneously and from
different structural types. The insights from multiple liganded crystal structures during both the
screening and chemical optimisation stages were invaluable in guiding progress, consistent with
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
1,20
contemporary practice
in fragment based design. Another important aspect of this work was
the use of analogue searching from an extensive collection. This was important both in the hit
finding phase, where it was used to generate additional hits, and in fragment optimisation and
growth, where significant improvements in hit activities were obtained without recourse to
synthesis. The chemical libraries are a major asset for large pharmaceutical companies, which
allow scientists to rapidly explore SAR around fragment hits. The “SAR by catalogue” allowed
expansion beyond the deliberately designed screening sets, whilst continuing to exercise
discipline with the molecular properties of compounds screened. Finally, micromolar inhibitors
of BCATm have been identified, with characterised novel binding modes and measurable
activities in a cellular assay. This represents a major step forward towards the goal of
establishing if pharmaceutical inhibition of BCATm using small molecules could provide a
4
useful treatment for metabolic diseases; the accompanying paper describes further progress
achieved in one series.
Experimental
ACS Paragon Plus Environment
29