Rational Design of Selective Adenine-Based Scaffolds for Inactivation of Bacterial Histidine Kinases

Article abstract of DOI:10.1021/acs.jmedchem.7b01066

Bacterial histidine kinases (HKs) are quintessential regulatory enzymes found ubiquitously in bacteria. Apart from their regulatory roles, they are also involved in the production of virulence factors and conferring resistance to various antibiotics in pathogenic microbes. We have previously reported compounds that inhibit multiple HKs by targeting the conserved catalytic and ATP-binding (CA) domain. Herein, we conduct a detailed structure-activity relationship assessment of adenine-based inhibitors using biochemical and docking methods. These studies have resulted in several observations. First, interaction of an inhibitor's amine group with the conserved active-site Asp is essential for activity and likely dictates its orientation in the binding pocket. Second, a N-NH-N triad in the inhibitor scaffold is highly preferred for binding to conserved Gly:Asp:Asn residues. Lastly, hydrophobic electron-withdrawing groups at several positions in the adenine core enhance potency. The selectivity of these inhibitors was tested against heat shock protein 90 (HSP90), which possesses a similar ATP-binding fold. We found that groups that target the ATP-lid portion of the catalytic domain, such as a six-membered ring, confer selectivity for HKs.

Full text of DOI:10.1021/acs.jmedchem.7b01066

Subscriber access provided by UNIVERSITY OF ADELAIDE LIBRARIES
Article
Rational Design Of Selective Adenine-Based Scaffolds
For Inactivation Of Bacterial Histidine Kinases
Manibarsha Goswami, Kaelyn E Wilke, and Erin E. Carlson
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.7b01066 • Publication Date (Web): 21 Sep 2017
Downloaded from http://pubs.acs.org on September 22, 2017
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 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Rational Design of Selective Adenine-Based Scaffolds for Inactivation of
Bacterial Histidine Kinases
1
2
1,2,3,4
Manibarsha Goswami, Kaelyn E. Wilke and Erin E. Carlson
*
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
Department of Chemistry, University of Minnesota, 207 Pleasant St. SE, Minneapolis, MN,
5454, United States
5
2
Department of Chemistry, Indiana University, 800 East Kirkwood Avenue, Bloomington,
Indiana 47405, United States
3
Department of Medicinal Chemistry, University of Minnesota
4
Department of Biochemistry, Molecular Biology and Biophysics, University of Minnesota
Abstract
Bacterial histidine kinases (HKs) are quintessential regulatory enzymes found ubiquitously in
bacteria. Apart from regulatory roles, they are also involved in the production of virulence
factors and conferring resistance to various antibiotics in pathogenic microbes. We have
previously reported compounds that inhibit multiple HKs by targeting the conserved catalytic
and ATPꢀbinding (CA) domain. Herein, we conduct a detailed structureꢀactivity relationship
assessment of adenineꢀbased inhibitors using biochemical and docking methods. These studies
have resulted to several observations. First, interaction of an inhibitor’s amine group with the
conserved, activeꢀsite Asp is essential for activity and likely dictates its orientation in the binding
pocket. Second, a NꢀNHꢀN triad in the inhibitor scaffold is highly preferred for binding to
conserved Gly:Asp:Asn residues. Lastly, hydrophobic, electronꢀwithdrawing groups at several
positions on the adenine core enhance potency. The selectivity of these inhibitors was tested
against Heat Shock Protein 90 (HSP90), which possesses a similar ATPꢀbinding fold. We found
0
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 2 of 52
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
that groups that target the ATPꢀlid portion of the catalytic domain, such as a sixꢀmembered ring,
confer selectivity for HKs.
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
The slow progress in the field of antibacterial drug discovery can be attributed to the lack
of novel targets in bacteria, especially those that will be less susceptible to the evolution of
antibiotic resistance. It is postulated that targeting bacterial virulence rather than viability could
1
be a powerful new strategy. Apart from preventing harmful bacteria from colonizing the host,
2ꢀ4
antivirulence therapies have several advantages that have been recently highlighted. Briefly,
antivirulence agents will not kill bacteria but rather deactivate pathogenic mechanisms, resulting
5
in less selective pressure for the evolution of resistant species. Since these compounds are not
bactericidal, they may preserve the natural host microbiome by fighting only pathogenic
2
, 6, 7
infections.
Significantly, this strategy increases the repository of antibacterial targets not yet
3
explored with clinical implications of superior antimicrobials with new mechanisms of action. It
has been hypothesized that antivirulence therapies will be effective when used alone, in
2
, 3, 7
combination therapy with traditional antibiotics, or as a prophylactic treatment.
Many virulence factors in both Gramꢀpositive and Gramꢀnegative bacteria are regulated
8
by bacterial signaling, primarily through twoꢀcomponent systems (TCSs). For example, in the
pathogenic bacterium Staphylococcus aureus, the SaeRS TCS affects the production of over 20
9
virulence factors. The PhoPR system controls polyketideꢀderived lipid biosynthesis in
Mycobacterium tuberculosis, and its absence causes dramatic changes to colonial and cording
1
0
morphology of this highly dangerous microbe. In Pseudomonas aeruginosa, many TCSs,
including PPrBA, are responsible for activating virulence genes, cell motility, and quorum
1
ACS Paragon Plus Environment

Page 3 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
8
sensing signal production. With growing evidence that TCSs are intimately involved in bacterial
virulence, interest in the study of these ubiquitous systems has dramatically increased in recent
1
1ꢀ15
years.
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
TCSs are composed of homodimeric enzymes known as histidine kinases (HKs) and their
cognate response regulators (RRs). Through TCSs, bacteria respond to many environmental
stimuli, such as ions, temperature, pH, oxygen pressure, autoꢀinducers, and contact with host
1
6
cells. Once a signal is received, ATP binds to the catalytic and ATPꢀbinding (CA) domain of
HK and autophosphorylates at a catalytic histidine residue of the dimerization and histidine
17
phosphotransfer (DHp) domain. In a relay event, this phosphate group is transferred to a
conserved aspartate on the RR, which then often binds to DNA to yield a downstream genetic
response. In addition to virulence, TCSs and HKs have been associated with antibiotic resistance
18
in many pathogenic strains, such as the VraS of S. aureus that is linked to vancomycinꢀ and
1
9
daptomycinꢀresistance and MtrAB of M. tuberculosis, that participates in efflux pump
8
regulation of multidrug resistance (MDR) and survival inside macrophages. Consequently, there
has been great interest in discovering inhibitors of TCSs. Recently, Wells and Marina identified
phenyl ringꢀbased putative HK inhibitors using a combination of in silico and in vitro screens
that have broadꢀspectrum antibacterial effects against both Gramꢀpositive and Gramꢀnegative
2
0
pathogens. In a similar approach, Meffre and coworkers designed thiophene derivatives as HK
2
1
inhibitors and evaluated their potential as adjuvants for the treatment of resistant bacteria.
We have also reported the discovery of HK inhibitors by the development of a
2
2
fluorescence polarization (FP) assay to identify compounds that target the CA domain. Highly
conserved across HKs, it contains the Bergerat fold that is characteristic of the GHKL (Gyrase,
23
Heat shock proteins, histidine Kinase, MutL) family but is absent in mammalian kinases. We
2
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 4 of 52
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
envisioned that targeting this conserved domain would inactivate several HKs simultaneously for
a broadꢀspectrum type antibiotic therapy. A highꢀthroughput screen for inhibition of HK853
(Thermotoga maritima) resulted in 119 hits. Further analysis for doseꢀdependent activity against
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
HK853, as well as VicK (Streptococcus pneumoniae) and CheA (Escherichia coli) yielded nine
2
2
lead compounds capable of inhibiting multiple HKs.
Four of the nine leads (and 19 of the original 119 hits) contained an adenine core
scaffold, an expected result given that the target protein binds to adenineꢀnucleotides. The
2
4
adenine chemotype is considered a “privileged scaffold” as many natural productꢀderived or
synthetic FDAꢀapproved antiviral and anticancer drugs are purineꢀbased and hence are likely to
be more successful clinically. Interestingly, all of our adenineꢀbased leads [6ꢀN,Nꢀdipropyl
adenine (1), 6ꢀN,Nꢀmorpholine adenine (2), 6ꢀN-phenylethyladenine (3), 6ꢀN,Nꢀ(3,3,5ꢀ
trimethylcyclohexyl) adenine (4)] were functionalized at the 6ꢀposition on the ring with various
2
2
secondary or tertiary amines (Figure 1a). To better understand the structural requirements for
inhibition of these proteins, we conducted structureꢀactivity relationship (SAR) studies with
complementary adenine analogs. In addition, given the importance of other ATPꢀbinding
proteins in eukaryotic biology, we sought to identify compounds that selectively target HKs over
eukaryotic kinases (EKs) and other GHKL proteins (i.e., heat shock proteins).
Given the similarity of our leads to the natural substrate of the HKs, we sought to further
examine the binding of ATP/ADP to the CA domain (Figure 1b,c,d). Previously, we conducted
multiple sequence alignment of 150 HK homologs and generated a map of the highly conserved
2
5
residues in the ATPꢀbinding pocket. The CA domain of HKs consists of a Bergerat fold (Figure
1d shows the conserved Nꢀ, G1ꢀ, G2ꢀ, G3ꢀ, Fꢀ boxes), which is unique to the GHKL family. By
comparing three different HKs coꢀcrystallized with nucleotides – HK853 (T. maritima) with
3
ACS Paragon Plus Environment

Page 5 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
17
26
ADPβN (PDB:3DGE, Figure 1c), WalK (Bacillus subtilis) with ATP (PDB:3SL2) and PhoQ
2
7
(
Escherichia coli) with AMPꢀPNP (PDB:1ID0) (Figure S1) – we observed many similarities
(
Figure 1b, S2). We also compared PDB:2C2A (Figure 1b), which is also the coꢀcrystal of
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
HK853 with ADP (in absence of its cognate partner RR468) along with the above structures as it
has crystallographic information of the water molecules within the protein unlike PDB:3DGE.
The direct ligand–protein interaction is a hydrogen bond (Hꢀbond) between a deeply seated,
conserved Asp (3DGE/2C2A D411, 3SL2 D533, 1ID0 D415) of the G1 box with the exocyclic
amine of the adenine ring. The carbonyl group from the same Asp residue interacts with N1
through a H O molecule, which also binds to a conserved Gly (G1 box; 3DGE/2C2A G415,
2
3SL2 G537, 1ID0 G419). The highly conserved Asn from the Nꢀbox (3DGE/2C2A N380, 3SL2
N503, 1ID0 N389) acts a bridging residue that participates in an H Oꢀmediated Hꢀbond with Nꢀ7
2
2+
on the adenine and ionic interactions with the βꢀphosphate and the coꢀfactor Mg ion (Figure 1b,
S2). Apart from these hydrophilic interactions, the adenine ring sits between the hydrophobic
residues Tyr (2C2A Y384, 3SL2 Y507, 1ID0 Y393), forming πꢀπ stacking, Ile/Val (3DGE/2C2A
I424, 3SL2 V546, 1ID0 I428), and Leu (3DGE/2C2A L446, 3SL2 L568, 1ID0 L446), which also
make contacts to the αꢀphosphate. Finally, additional bonds that anchor the nucleotide are the
2
+
coordination of the phosphates to the octahedral Mg ion and to basic residues: Lys from the Nꢀ
box (3DGE/2C2A K383, 3SL2 K506, 1ID0 K392) and Arg from the Fꢀbox (3DGE/2C2A R430,
3SL2 R552, 1ID0 R434). As the βꢀ and γꢀphosphates interact with the Fꢀbox on the ATPꢀlid, this
flexible loop is responsible for presenting a favorable conformation for the phosphate transfer to
1
7, 26ꢀ28
the DHp domain. The interactions observed here are corroborated by similar analyses.
In
summary, several conserved residues interact in unison to tightly place ADP in the binding
pocket, valuable insight that provides an opportunity to rationally design a potent and
4
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 6 of 52
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
competitive inhibitor to disrupt these interactions.
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. (a) Adenine leads with 6ꢀC substitution from original screen. IC50 values are for
2
9
HK853 inhibition, (b) Interactions of ligand obtained using MOE program of HK853 coꢀ
crystallized with ADP (PDB:2C2A). The legend of interactions is also shown here, (c) Ribbon
diagram of HK853 (PDB:3DGE) showing CA domain (white ribbon) and DHp domain (yellow
ribbon), (d) Expansion of the CA domain showing features of the Bergerat fold and important
residues binding to the coꢀcrystallized ligand, ADP.
Results and Discussion
5
ACS Paragon Plus Environment

Page 7 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Assessment of ATP Binding Mode Through its Structural Elements
Most adenineꢀbased drugs differ in the substituents on the scaffold and/or modifications
that are intended to mimic the ribose and/or phosphates of ATP. Hence, we explored the SAR of
adenine ring functionalization, as well as the key components of ATP to facilitate design of
potent and selective HK inhibitors (Figure 2a). To test the importance of the αꢀ, βꢀ, and γꢀ
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
phosphates, we measured the affinity of HK853 for ATP (5), ADP (6), and AMP (7) using the
3
0
activityꢀbased fluorescence gel assay previously developed by our group (Figure 2b). 5 and 6
had similar IC values (27 ꢁM, 2.4 ꢁM), whereas 7 had marginal activity (1030 ꢁM), suggesting
5
0
2+
that one phosphate group is not enough for octahedral Mg coordination or interactions with
2
6, 27
Arg on the ATP lid and Lys of the Nꢀbox for optimal binding.
Indeed, when these residues
are mutated in PhoQ (E. coli, K392A, R434A), a 50ꢀfold loss in ATPꢀbinding and catalysis was
27
observed. Next, we investigated the role of ribose in nucleotide–protein binding. In general, the
solventꢀexposed 22, 32ꢀhydroxyls of the ribose unit interact with Arg and Val through a common
H O molecule. We measured the ability of adenosine (8) to inhibit HK853 and found it to be
2
moderately better than AMP with IC : 407 ꢁM (Figure 2b, Table 1). This was surprising as the
5
0
potency of the simple adenine base (9) (IC : 445 ꢁM) is comparable to adenosine, which
5
0
indicates that addition of the sugar does not increase affinity.
As mentioned, we were intrigued that the exocyclic N6 of most of the leads is not a
primary amine as it is in the native substrate, but was instead decorated with hydrophobic
substituents (Figure 1a). While the absence of 6ꢀCꢀNH could preclude Hꢀbonding with the
conserved Asp or Gly, these N6ꢀmodified leads bound with greater affinity. These results
prompted us to test the nucleoside/nucleotide analogs of some of the leads to decipher their
impact on interactions with the active site. First, we examined HK853 inhibition by nucleoside
6
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 8 of 52
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
derivativeof lead 3, 6ꢀN-phenylethyladenosine (10) and observed that it had negligible activity
despite the presence of the ribose moiety (Figure 2b, Table 1). Surprisingly, the related
diphosphate nucleotide, 6ꢀNꢀmethylphenyl ADP (11) was only marginally better than 10 (Note:
compound 11 has R = PhCH was used instead of PhCH CH due to commercial availability),
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
2
2
2
indicating that presence of phosphates does not increase activity. To further evaluate the role of
ribose, we tested the activity of 6ꢀN,Nꢀdipropyl adenosine (12), the nucleoside of lead 1 and
similar to 10. This analog failed to show any inhibition. Based on these observations and the
likely disturbance of the critical and direct 6ꢀNH bond to the conserved Asp, we speculate that
compounds 1–4 bind to the CA domain in an orientation different from that of ATP/ADP in
order to adopt other favorable interactions, exhibited by the lower IC values of these leads.
50
7
ACS Paragon Plus Environment

Page 9 of 52
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 2. a) SAR approach to develop selective HK inhibitors. b) HK853 inhibitory activity of
ATP and its components. Docking of 8 (c) and 1 (d). The top images show the docking pose of
the compound in the receptor cavity. For clarity, only the cavity (green ribbons), and not the full
protein, is shown, including important residues (sticks) for ligandꢀprotein binding. The ligands
8
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 10 of 52
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
are shown as sticks with C = magenta, N = blue. These images were produced with PyMOL after
docking was performed in SYBYL SurflexꢀDock. The lower ligand–protein interactions were
generated with MOE program. Legend of the possible ligand interactions generated by MOE.
To evaluate this hypothesis, we utilized molecular modeling to examine possible poses
and interactions between the target protein and ligand. We used the SurflexꢀDock SybylꢀX
program to dock various ligands and MOE and PyMOL for computing interactions and
visualizing. First, we docked 8 into the ATPꢀbinding pocket of HK853 (PDB:3DGE) and found
its binding poses to be similar to ADPꢀβN, the coꢀcrystallized ligand (Figure 2c). This includes
conserved interactions with D411 and ribose hydroxyl group interactions with V431 and G443
from the G1ꢀbox. A similar docking pose was observed with 9 (not shown). On the other hand,
the docking of lead 1 yielded a binding pose different from that of 8 or 9 where its secondary
amine 9ꢀNH forms the crucial Hꢀbond to the conserved Asp (Figure 2d). The propyl groups
appear to access a nearby hydrophobic region consisting of conserved residues I424, L446 and
V431. Our hypothesis that availability of the 9ꢀNH for binding to D411 is essential for activity is
consistent with the low potency of 12 (and compounds 10, 11), where the Hꢀbond donor is no
longer available due to the addition of the ribose. Similarly, the docking of lead 3 suggests that
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
9ꢀNH Hꢀbonds to D411 while the phenylꢀethyl group occupies the previously mentioned
hydrophobic pocket (Figure S4a). In all of these compounds, another conserved hydrophobic
region consisting of I411, I416 and F472 is situated near the adenine ring. Lead 3’s riboside
derivative 10 exhibited >20ꢀfold decreased potency, further substantiating that 9ꢀNH is important
for lead binding, which is abolished when the ribose is present.
Contribution of the Conserved Aspartate Residue
As most of our hypotheses are based on the interactions of the conserved Asp with an Hꢀ
9
ACS Paragon Plus Environment

Page 11 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
bond donor –NH, we compared the potency of these inhibitors between wildꢀtype (WT) HK853
and the mutant protein HK853 D411N. Our previous competition assays were ineffective for
detecting mutant HK activity and binding (BODIPYꢀATPγS (Figure S3), ADPꢀBODIPY (Figure
S3), respectively), presumably because both probes also require the D411 for binding. Thus to
obtain data with the mutant protein, we utilized a thermal shift assay, differential scanning
fluorimetry (DSF). DSF analysis has become popular in the past few years as it is a fast, robust,
highꢀthroughput method and can be easily performed with a commercially available dye on an
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
3
1, 32
RTꢀPCR instrument.
DSF monitors the thermal unfolding of proteins in the presence of a
fluorescent dye (SYPRO orange). When a ligand binds to a protein favorably, a shift (ꢂT ) is
m
observed in the melting temperature of the protein (T ) due to increased stability of the protein in
m
the presence of the ligand. Using this strategy, we investigated if inhibitors bind to the HK in
absence of the conserved D411. Thermalꢀshift assays or T s have not been reported elsewhere
m
for HK853, a protein from the hyperthermophilic organism T. maritima and thus, expected to be
3
3
very thermally stable.
We first tested the T changes for the WT HK853 in the presence and absence of ADP 6.
m
We found that WT HK853 alone has a T = 70 °C, and in the presence of 6, the T increases to
m
m
8
0 °C (Figure S6). A similar T shift is seen with lead 1, which we expect binds to D411 through
m
9
ꢀNH. The ꢂT of 10 °C shows that favorable binding occurs between 6 or 1 and the enzyme.
m
The same experiment was executed with HK853 D411N (Figure S5). The D411N mutant protein
has a T of 67 °C, which is lower than the WT T (Figure S6). In the presence of 6 or 1, there
m
m
was no significant change in the T value (≤2 °C), which suggests that D411 is crucial for
m
interaction of 6 or the adenine inhibitors to the bacterial enzyme. Reduction of inhibitor affinity
2
5
was seen by our group previously for guanidineꢀbased compounds in mutant HK853 D411A.
10
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 12 of 52
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
Inhibition loss was also seen by Zhao and coworkers in PhoQ (Salmonella enterica) when
3
4
conserved D416 was mutated to alanine. Apart from inhibitor binding, mutation of this Asp
3
5
residue results in complete loss of kinase activity, which further suggests the importance of
D411 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
Evaluation of Additional Pyrimidine Substitutions
In addition to the 6ꢀN and 9ꢀN positions, we evaluated SAR for other positions on the
adenine ring using a combination of commercially available and synthesized adenine/adenosine
analogs. First, 2ꢀaminoadenine (13) was tested for HK853 inhibition and surprisingly, it proved
to be much worse (13, IC 2.6 mM) than 9 (Figure 3a, Table 1). Addition of a single –NH
50
2
group at the 2ꢀC position resulted in 6ꢀfold less potency, which suggests that this moiety might
make the adenine electron rich, thereby weakening other nonꢀcovalent interactions. A similar
trend was observed with aminoadenosine (14, IC 1.8 mM), which again suggested that ribose
5
0
plays no role in effective binding and that the crucial Hꢀbond to D411 is through 6ꢀCꢀNH and
2
not 9ꢀNH in these analogs (Figure 3a, Table 1). To investigate these hypotheses, 2ꢀchloroadenine
(
15) and its riboside, 2ꢀchloroadenosine (16) were tested as this group adds bulk on the 2ꢀC
position, yet is instead electron withdrawing. Interestingly, the chloro group enhanced potency of
and 9 by 10ꢀfold. Similar to the chloro, the fluoro analogs, 2ꢀfluoroadenine (17) and 2ꢀ
fluoroadenosine (18) also exhibited superior potencies to 8/9 albeit to a lesser extent. Fluoro and
8
36
chloro are both mildly deactivating groups and lower the pKa of the bases by at least 2.5 times.
A pKa decrease due to Cl may strengthen the Hꢀbond of 6ꢀCꢀNH with D411, observed as
2
increased affinity of the inhibitor. Increased affinity may also be influenced by: 1) the
differences of the size of the group and/or its lipophilicity, or 2) the EWG –Cl favors the
37
conserved πꢀπ stacking of the adenine ring. Nonetheless, further investigation was necessary,
11
ACS Paragon Plus Environment

Page 13 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
particularly as to whether increased affinity is specific to –Cl or is common to other deactivating
groups. Docking of 15 and 16 showed Hꢀbonding between D411 and 6ꢀCꢀNH , πꢀπ stacking with
2
Y384 (Figure 3b, S4b), and Cl occupying the conserved hydrophobic region constituting L446,
I/L377 and F472. In contrast, 13 and 14 failed to dock into the receptor. A possible explanation
for this result is that the hydrophobic area as seen in docked 15/16 clashes with the polar –NH2
group in 13/14. Apart from the total score, SYBYL SurflexꢀDock also provides a crash score
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
(
degree of inappropriate penetration by the ligand into the protein and of interpenetration
between ligand atoms or selfꢀclash that are separated by rotatable bonds). A crash score close to
is favorable, while negative numbers indicate penetration. These scores can be highly useful to
0
have deeper understanding of favorable or failed docking outcomes. For both 13 and 14, the
crash value was ≤ –4.0, which suggests that the ligand atoms clashed with the residues in the
ATP domain (Table S2).
12
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 14 of 52
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. a) HK853 inhibitory activity of adenine compounds with pyrimidine ring
modifications. Docking of compounds 15 (b) and 19 (c). The top images show the docking pose
of the compound in the receptor cavity. For clarity, only the cavity (green ribbons), and not the
full protein, is shown, including important residues (sticks) for ligandꢀprotein binding. The
ligands are shown as sticks with C = magenta, N = blue, Cl = green. These images were
produced with PyMOL after docking was performed in SYBYL SurflexꢀDock. The lower
ligand–protein interactions were generated with MOE program. Legend of the possible ligand
interactions generated by MOE.
Next, we sought to investigate if swapping the 2ꢀCꢀCl and 6ꢀCꢀNH groups would have
2
13
ACS Paragon Plus Environment

Page 15 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
effects on inhibitor activity. The purine, 2ꢀamino,6ꢀchloropurine (19 IC 118 µM) has better
5
0
affinity than adenine (5), and is slightly more potent than the 2ꢀCl analog 15 (Figure 3a, Table 1).
Compound 19 was then docked using the same protocol (Figure 3c). Interestingly, a binding pose
more similar to that of leads 1 or 3 where 9ꢀNHꢀ makes polar contacts with D411 was observed,
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
even in the presence of a free primary amine at 2ꢀCꢀNH . If this interaction is present, then the
2
riboside, 2ꢀamino,6ꢀchloropurine ribose (20) should show no activity, which is what is observed
in our biochemical assay (IC 1.4 mM) and docking experiments (ligand failed to dock; Figure
5
0
3). This result reinforce that the crucial Hꢀbonding interaction to the Asp in HKs is through the
ꢀNH and not through 2ꢀCꢀNH , unless the ribose causes unfavorable interactions. A similar low
9
2
affinity was seen with 2ꢀaminopurine (21), which is not substituted at 6ꢀC and has a 2ꢀCꢀNH2,
thereby indicating that this amine is incapable of binding to D411. Finally, if the inhibitor binds
through the 9ꢀNH, other binding events to the conserved G415 and N380 as seen with ADP, can
occur through the nearby N atoms (like 3ꢀN or 7ꢀN). We estimate that these additional contacts
make the flipped pose of inhibitors 1, 3 or 19 highly favorable and possibly in 2/4. In summary,
these studies indicate that a chloro substituent on the adenine greatly enhances its activity, and is
preferred on both the 2ꢀC and 6ꢀC positions, although the binding poses are different for the
resulting inhibitors.
Since the lead compounds (1, 2, 3, 4) and the superior inhibitor 19 had 6ꢀC
functionalization, we next explored this position on the ring. Examination of the 15 additional
adenineꢀlike hits for the HTS indicated that large, hydrophobic groups at Cꢀ6 often result in
protein aggregation at low concentrations (~100 µM; compounds referred to as nonꢀleads, NLꢀ;
representative examples in Figure 4a, all in Figure S7). Thus, we pursued variants that are similar
in size to lead 4 such as the bioisosteric analog 22 (Figure 4a, Table 1). Bioisosteres or structural
14
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 16 of 52
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
motifs that are similar in shape and function have proven to enhance potency and/or
3
8
pharmacological properties. Gratifyingly, we observed that placement of an aniline group on 6ꢀ
C, 6ꢀanilinopurine (22) had superior potency (IC 7.3 ꢁM) compared to 3 and 9 by nearly 20ꢀ
5
0
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 60ꢀfold, respectively. This increase can be attributed to the electron delocalizing effect
resulting in Hꢀbond and/or enhanced hydrophobic interactions with the aniline group. To
determine if the analog binds to D411 through 9ꢀNH like 3, we synthesized the riboside, 6ꢀ
anilinopurine ribose (23). This analog failed to inhibit the HK, confirming our hypothesis that
when the 6ꢀN position is functionalized with an alkyl group, the binding pose preferred by the
adenine base is likely one where it donates a Hꢀbond to the Asp through 9ꢀNH. Upon docking
22, a similar observation was made with 9ꢀNH–D411 and an additional areneꢀH interaction was
seen with conserved V431 in the ATP lid (Figure 4b, Table 1). Importantly, this compound does
not aggregate HK853 even at ≥1.25 mM (See aggregation analysis section in SI).
15
ACS Paragon Plus Environment

Page 17 of 52
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 4. a) HK853 inhibitory activity of adenineꢀlike compounds with pyrimidine ring
modifications. b) Docking pose of compound 22. The left image shows the docking pose of the
compound in the receptor cavity. For clarity purposes, only the receptor cavity (green ribbons)
and not the full protein, along with the important residues (sticks) participating in the ligandꢀ
protein binding, are shown. The ligands are shown as sticks with C = magenta, N = blue. The
right image was produced with PyMOL after docking was performed in SYBYL SurflexꢀDock.
The images are ligandꢀprotein interactions were generated with MOE program. Legend of the
16
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 18 of 52
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
possible ligand interactions generated by MOE.
At this stage, placement of an anilino group at 6ꢀC and chloro at 2ꢀC resulted in superior
activities for adenine inhibitors. However, these compounds employed opposite docking poses to
interact with D411 indicating that the combination of these two features may yield a poor binder.
We synthesized 2ꢀchloro,6ꢀanilinopurine (24) by an electrophilic substitution on 1, 2ꢀ
dichloropurine with aniline and as predicted, found it to be far less potent (IC 1.3 mM) than 22
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
5
0
or 15 (Figure 4a, Table 1). Even addition of a small group, such as methyl, at 6ꢀCꢀNH in
combination with the 2ꢀCl yielded an inferior inhibitor, 2ꢀchloro,6ꢀNꢀmethyladenine (25; IC50
837 µM) in comparison to 15 or 9. Consistent with these data, when ligand 24 was docked into
HK853, it attained a unique binding pose in which the 9ꢀNH weakly interacts with Y429 and
contacts with D411 are no longer observed (Figure S4c). During the preparation of 24, we
obtained the diꢀsubstituted adenine derivative (S1) as a byꢀproduct. We tested the potency of this
compound (Table S1) and found it to exhibit much worse inhibition than 24 (IC50 7 mM),
suggesting that the additional bulky group at the 2ꢀC position is highly unfavorable.
Evaluation of Imidazole Substitutions
After exploring the pyrimidine ring of the adenine core, we next focused on the pyrrole
ring system. The 7ꢀN of the pyrrole ring of ADP is predicted to form a waterꢀmediated Hꢀbond to
conserved N380 and we postulated that this residue is vital for nucleotideꢀbinding as it likely
2
+
also interacts with ribose, the phosphates and Mg (Figure 1b). Mutation of the analogous amino
acid, N347D, in EnvZ (E. coli) by Tanaka and coworkers demonstrated that substituting this
3
9
important residue causes complete loss of ATP binding and autophosphorylation activity. We
do not observe these interactions in our docking studies as they were performed following the
removal of water molecules from the active site. Nevertheless, residues N380 and G415 are often
17
ACS Paragon Plus Environment

Page 19 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
situated in close proximity to the docked poses of our leads (within 3ꢀ7Å). To further assess the
importance of 7ꢀN to productive binding, we examined analog 7ꢀNꢀmethyladenine (26), in which
the nitrogen has been methylated (Figure 5a, Table 1). As expected, 26 failed to show any
activity indicating that the free 7ꢀN is vital. We also noted that the 2ꢀN,Nꢀdimethyl amino version
(Figure S8) was a nonꢀhit (NH) in our HTS pilot screen. When the methyl group is instead
positioned on 9ꢀN, the resulting compound 9ꢀNꢀmethyladenine (27) is more potent than
unmodified base 9 and possibly interacts with the active site in a similar orientation. These data
suggest that the combination of Nꢀmethylation and a chloro group at Cꢀ2 (e.g., 15) may yield a
more active molecule. As expected, methylation at 7ꢀN resulted in an inactive compound (2ꢀ
chloro,7ꢀNꢀmethyladenine, 28). Instead, when 9ꢀN was methylated to provide 2ꢀchloro,9ꢀNꢀ
methyladenine (29), we found it to be superior to both 15 and 27 (IC 95 µM; Figure 5a, Table
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
0
1). This observation confirms that in addition to Hꢀbonding interactions with D411, molecule
binding is dependent upon interactions with residues such as N380 and G415.
18
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 20 of 52
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 5. (a) Adenine inhibitors with functionalization at 7ꢀN, 9ꢀN and 2ꢀC positions and their
activity for HK853 inhibition. (b) Adenine inhibitors with functionalization at 8ꢀC positions and
their activity for HK853 inhibition.
Inhibitor Selectivity Assessment
Inhibitor selectivity is an essential feature of lead compounds, thus, we sought to examine
the ability of adenineꢀbased compounds to interact with two important classes of ATPꢀbinding
mammalian proteins, HSPs and EKs. EKs or mammalian (Ser/Thr, Tyr) kinases share common
features within their catalytic site including several βꢀsheets that together form a conserved Nꢀ
lobe, while αꢀhelices constitute the conserved Cꢀlobe and in between these lobes, there is “hinge”
4
0, 41
region where ATP binds (Figure S9a).
These active site features are quite distinct from the
19
ACS Paragon Plus Environment

Page 21 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Bergerat fold that is present in HKs. Further exploring one of EKꢀadenine inhibitor crystal
4
2
structures (PDB:4XX9, Figure S9b,c) and comparing that to ADP–bound HK structures reveal
many differences, providing prospects for the design of selective inhibitors.
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
Notably, several clinicallyꢀutilized drugs that target the EKs are adenineꢀbased and
provided the ideal platform to determine if EK inhibitors also interact with the HKs (Figure
4
1
6
a). These compounds all contain an NH or NH at the 6ꢀC position and most display
2
substituents on the 7, 8, 9ꢀpositions of the pyrroleꢀtype ring. The coꢀcrystal structure of B43
(similar to Ibrutinib) with Bruton’s tyrosine kinase (BTK) shows Hꢀbonds between the 6ꢀNH and
1ꢀN on the 4ꢀamino pyrrolopyrimidine ring to backbone residues, Glu475 and Met477
4
3
(
PDB:3GEN, Figure S10b), whereas Idelalisib interacts with phosphoinositide 3ꢀkinase delta
(PI3Kδ) by forming an Hꢀbond between V828 and Nꢀ3 of the pyrimidine ring (PDB: 4XE0,
44
Figure S10a). Although these structures do not depict a generalized picture of how these two
kinase drugs might interact with other EKs, the distinct differences in the inhibitor interactions
compared to that of HKs can be observed. Nevertheless, we purchased Idelalisib and Ibrutinib
and tested their activity for HK853 inhibition. Both of them failed to show any inhibition even at
1.25 mM, providing evidence that selective inhibition can be attained (HK853 inhibition section
in SI). We utilized the structures of these adenineꢀlike kinase drugs to yield a picture of their
commonalities as shown in Figure 6a. The pink circle highlights the free amine, the yellow
circles show the Nꢀatoms within the scaffold required for protein interactions and the blue arrows
indicate the positions on the ring that display substituents. Given the distinct fold that EKs use to
bind ATP and the inability of these EK inhibitors to affect the HKs, the likelihood that selectivity
for the latter is achievable is high.
We next focused on the GHKL family, specifically HSPs. In the past decade, HSP90
20
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 22 of 52
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
45
chaperone proteins have become of paramount interest in the field of cancer drug discovery. As
HSPs are prime drug targets in humans, attaining selectivity against this class of proteins is of
great importance. HSPs and HKs share the Bergerat fold as their adenineꢀnucleotideꢀbinding
pocket indicating that inhibitors of one enzyme type could bind the other favorably (Figure S11).
Moreover, within the GHKL family, HKs are structurally closest to HSPs due to their similarities
in the flexible ATPꢀlid. These features prompted us to investigate coꢀcrystal structures of HSP90
to gather information about their ligandꢀreceptor binding. Not surprisingly, many HSP90
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
inhibitors are derived from a purineꢀscaffold and several coꢀcrystal structures are available
45
(
Figure 6b). Chiosis and coworkers have performed pioneering work in the development of
potent adenineꢀbased HSP90 inhibitors and one is in clinical trials as an antitumor agent (Figure
4
6
6b, PU-H71). Pharmaceutical companies, such as Vernalis, have conducted fragment screens
and discovered purineꢀbased hits that are similar to some of our analogs (Figure 6b, VFꢀseries).
For example, VF1 is similar to 27 and compound VF3 displays a 6ꢀCꢀNH feature as seen in 3
4
5
and 29. Inspection of purine inhibitorꢀHSP90 coꢀcrystal structures highlights several key
interactions, including: 1) Hꢀbond of 9ꢀNH to D93, 2) waterꢀmediated Hꢀbonds with conserved
G97 and N51 by the 3ꢀN and 7ꢀN on adenine ring (Figure 7c). These interactions directly
correlate with those predicted by our docking and SAR studies with the HKs. In addition, the
flipped binding pose that we predicted for analogs 1 and 3 was observed with HSP90 bound to
VF2-3, where 9ꢀNH interacts with the conserved aspartate (Figure S12). As described in this
45
report, this binding mode enables Gly:Asp:Asn coupling with the 7ꢀN:9ꢀN:3ꢀN triad as seen
with the related adenine sites, 1ꢀN:6ꢀCꢀNH :7ꢀN (Both triads are shown in Figure S13 with
2
postulated interactions). The strong similarities between our models of adenine inhibitorꢀHK
binding and published HSP90ꢀligand structures provides additional evidence for the accuracy of
21
ACS Paragon Plus Environment

Page 23 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
our SARꢀdocking results.
We conducted a thorough survey of the published purineꢀbased HSP90 inhibitors that
4
5ꢀ47
have been developed both commercially and academically.
Curiously, the inhibitors
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
discovered by Abbott and Biogen bound to the catalytic D93 of mammalian HSP90 through the
4
5, 48
aminopyrimidine moiety, a position equivalent to 2ꢀC of adenine (Figure 6b).
Importantly,
similar compounds from our library (13, 14, 20, 21) showed negligible activity, with the
exception of 19. A key difference between 19 and the HSP90 analogs is that the main Hꢀbond to
the conserved Asp is through 9ꢀNH, whereas in the HSP90 inhibitors it is through 2ꢀCꢀNH . We
2
decided to investigate this feature with other nucleobases, as many of them have an aminoꢀ
pyrimidine group (Figure 6b). Among these, guanine or guanosine both did not inhibit HK853,
indicating that in the presence of a 2ꢀCꢀNH , D411 does not bind indiscriminately to any
2
available –NH on the purine ring. Several unnatural bases (hypoxanthine, inosine) also failed to
bind, further indicating that HKs are selective to adenineꢀlike nucleobases and that the presence
of a 2ꢀCꢀNH is disfavored. These results were encouraging towards our effort to identify HKꢀ
2
selective leads.
22
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 24 of 52
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 6. (a) FDA approved EK inhibitors and their HK853 inhibition. (b) Published HSP90
inhibitors. Yellow spheres indicate the similar Nꢀscaffold, pink spheres show the conserved ꢀNH,
which forms an Hꢀbond to an Asp residue and the blue arrows indicate substitutions on the ring
commonly found in these inhibitors.
We noted several common features in HSP90 adenine inhibitors including a highly
functionalized phenyl ring attached at 8ꢀC position (Figure 6b). Since the 8ꢀC on the imidazole
core is crucial for HSP90 activity, we sought to determine if such compounds inhibit the HKs. A
number of adenine analogs that are functionalized on 8ꢀC were included in our initial HTS and
23
ACS Paragon Plus Environment

Page 25 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
were all found to be inactive (n=1; representative examples in Figure 5b, all in Figure S8). We
also examined the bioisostere, 8ꢀazadenine (30), to determine if replacement of the carbon with
nitrogen is beneficial and found minimal activity, suggesting that any substitution on the 8ꢀC
position yields unfavorable HK interactions. We anticipate that this feature is likely to aid in the
design of selective inhibitors for HKs, as most mammalian kinase and HSP90 drugs have 8ꢀC
modifications.
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
24
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 26 of 52
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
Figure 7. (a) Representative examples of compounds tested for HSP90α inhibition. (b)
Combined SAR analyses of all inhibition studies on HKs, EKs, HSP90s yielding selective HK
inhibitors. Yellow spheres indicate the similar Nꢀscaffold, pink spheres indicate key NH for Asp
interaction, blue arrows indicate the substitutions on the ring to achieve selective HK inhibitors
and blue arrows with black crosses indicate substitutions that result in poor inhibition for HKs.
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
(c) Coꢀcrystal structure of HSP90 with compound VF3 (PDB:2YEH), showing the active site
residues interacting with the ligand. The residues belonging to different parts of the Bergerat fold
are colorꢀcoded, Blue = Nꢀbox, Purple = Gꢀ1 box, Red = Gꢀ2 box, Brown = short αꢀhelix. The
red and brown regions constitute the ATP lid domain in HSP90s. (d) Docked pose of compound
4
in HK853 showing key conserved residues that interact with the ligand. The residues belonging
to different parts of the Bergerat fold are colorꢀcoded, Blue = Nꢀbox, Purple = Gꢀ1 box, Red = Gꢀ
box, Green = Fꢀbox, Brown = flexible loop. The red, green and brown regions constitute the
ATP lid domain in HKs.
Although it is important to test HSP90 inhibitors for HK inhibition, it is critical to test the
2
adenineꢀbased HK inhibitors for HSP90 inhibitory activity. To this end, a commercially available
FPꢀbased binding assay to assess HSP90α inhibition was utilized to test the efficacy of several of
our adenine inhibitors. As outlined in Figure 7a, out of the 13 compounds that were tested, 5
analogs showed inhibition of HSP90 (full compound list in Figure S13; selectivity ratios in Table
S6). Compounds 3, 4 that display long 6ꢀC alkyl or alkylꢀphenyl amines and 2ꢀCl adenosine (16)
were moderately active (IC ≤1.25 mM). Whereas, with a –CH at the 9ꢀNH position in 29, this
5
0
3
inhibition is reduced (IC >>1.25 mM), suggesting that this scaffold might be a starting point for
50
generating selectivity. We compared the inhibition activity of 1, 4, 16, 29 for HKs against
HSP90 and found that selectivity for the former is high (10ꢀ2000 fold; Table S6). Some of the
25
ACS Paragon Plus Environment

Page 27 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
compounds (25, 26) that were ineffective HK inhibitors were also found to be inactive for
HSP90, supporting the similarities in the ligandꢀbinding of these two proteins.
Overall, our studies indicate that one of two adenine NꢀNHꢀN fragments is essential for
binding to HSP90 and HKs and enabled us to generate three purineꢀbased scaffolds that are
potent, as well as selective for bacterial HKs (Figure 7b). The first scaffold (X) has the 1ꢀN:6ꢀCꢀ
NH :7ꢀN design, a –CH substituent at 9ꢀN and a 2ꢀCl group. This core structure can be further
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
3
modified at the 9ꢀN and 2ꢀC positions to improve potency but not at the Cꢀ8 position. The more
potent scaffold (Y) displays the 7ꢀN:9ꢀNH:3ꢀN triad that has a 6ꢀmembered cyclic group at 6ꢀCꢀ
NH. This scaffold can be modified at various positions including bioisosteres at the 6ꢀCꢀNH
position and we decided to investigate it further. First, to rationalize the observed selectivity, we
examined the coꢀcrystallized structure of VF3 with HSP90 as it has a 6ꢀC modification like
compounds 4 and 22 (Figure 7c, S12c). We compared it to the docked pose of 4 in HK853
(Figure 7d), which was found to be inactive for HSP90. As shown in Figure 7cd, the 9ꢀNH in
both compounds forms an Hꢀbond with the conserved Asp (D93, D411) and the exocyclic 6ꢀNH
substituents position themselves near the ATPꢀlid region. It should be noted that although HSPs
and HKs share the Bergerat fold, their ATP lid domains are quite distinct. Specifically, the ATP
lid in HSP90 is long and comprised of short αꢀhelices, Gꢀ2/3 boxes. Whereas in HKs, the ATP
lid is much more flexible and has a conserved hydrophobic Fꢀbox along with Gꢀ2/3 boxes. Our
modeling suggests that the cyclohexylamine in 4 (or the Ph ring in 22) likely accesses the
conserved hydrophobic residues (e.g., V431, I424 and L446) on the lid of HK853, which are not
found in HSP90. Unlike 4, the alkyl ether in VF3 interacts with T109 and L107 from the ATP lid
in HSP90. In HKs, this region is more hydrophobic and appears to prefer 6ꢀmembered cyclic
lipophilic groups. All of these interactions can possibly aid in the selectivity of compounds 4 and
26
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 28 of 52
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
2
2 for the HKs. Moreover, the size of such a cyclic group likely mimics the ribose in the ADPꢀ
pocket, which perhaps allows the optimum positioning of the 6ꢀmembered ring in these
molecules. Finally, although the simple purine scaffold (Z) has some similarities to the
aminopyrimidine HSP90 inhibitors reported by Biogen and Abbott, our data indicate that
selective HK inhibitors could be obtained with careful design.
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
Conclusion:
The World Health Organization (WHO) recently published a report on the world’s 12
most dangerous superbugs that need immediate attention as treatment options are running out.
Most of these antibioticꢀresistant strains like MDR M. tuberculosis or methicillinꢀresistant S.
aureus have TCSs or bacterial kinases that control critical signaling pathways and aid in
pathogenicity. We have previously generated diverse leads that can inactivate multiple histidine
kinases making them potential antibacterial candidates. In this work, we report a detailed
investigation on the SAR studies of one class of these molecules; the adenineꢀbased leads. By
doing structureꢀbased design and exploration of substituents on various positions about the
adenine ring, we discovered more potent inhibitors. Some key residues such as D411, N380,
G415 in HK853 and their interactions with the inhibitors were found to be essential for activity
and must be considered in the design of ATPꢀcompetitive compounds. Since bacterial kinases are
functionally similar to mammalian kinases and structurally related to chaperone heatꢀshock
proteins, the selectivity of the HK inhibitors was also tested. Conducting various inhibition
assays and correlating the function of the tested molecules, we constructed three purineꢀbased
scaffolds that are not only effective but are also selective for HKs. Efforts are ongoing to study
the effects of these inhibitors in both Gramꢀpositive and Gramꢀnegative bacteria.
27
ACS Paragon Plus Environment

Page 29 of 52
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Experimental Section:
Adenine inhibitor library
All molecules were >95% purity as judged by HPLC analysis (Instrument: Agilent HPLC 1200
series, column: Agilent Eclipse XDBꢀC18, 5 ꢁm, 9.4x250 mm, gradient run of 20 min with 95%
water to 100% acetonitrile, modifier: 0.1% Formic acid)
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
Compound Synthesis
4
9
The Nꢀ6 substituted purines were synthesized following previous protocols. In brief, to
chloropurine or chloropurine riboside was added the amine (3 equiv., 3ꢀ5 mmol) and ethanol.
This mixture was refluxed overnight (>16 h). After completion of reaction was confirmed by
TLC, the solvent was evaporated in vacuo and the residue purified by column chromatography
with EtOAc:MeOH (gradient solvent from 12:1 to 9:1) as solvent system. For characterization,
NMRs were taken on 500 MHz Bruker instrument equipped with a cryoprobe. ESIꢀMS was
performed on an Agilent UPLCꢀQTOF instrument in positive ionization mode.
6-N,N dipropylaminopurine riboside 12 (white crystalline solid, MP = 136ꢀ138°C, 72% yield).
1
HNMR (CD CN) H NMR (500 MHz, CD CN) δ 8.14, 7.92 (2s, 2H, adenine ring Hs), 5.96 –
3
3
5.83 (m, 1H, Hꢀ1 of ribose), [5.78 (d, J = 7.1 Hz, 1H), 4.82 (td, J = 6.8, 5.0 Hz, 1H), 4.29 (td, J =
.1, 1.5 Hz, 1H), 4.14 (d, J = 1.6 Hz, 1H), 3.72 – 3.62 (m, 4H), 3.52 (d, J = 3.15 Hz, 1H) ribose
ring Hs, 2x ꢀNꢀCH CH CH ], 1.70 (h, J = 7.4 Hz, 4H, 2x ꢀNꢀCH CH CH ), 0.93 (t, J = 7.4 Hz,
3
2
2
3
2
2
3
1
3
6H, 2x ꢀNꢀCH CH CH ). C NMR (126 MHz, CD CN) δ 155.35, 152.35, 150.25, 139.93,
2
2
3
3
121.68 (5C, adenine ring Cs), 91.43, 88.41, 74.21, 72.89 (5C, ribose ring Cs), 63.61, 11.36 (2C, ꢀ
NꢀCH CH CH ). Due to the fast tumbling of tertiary amines ꢀNꢀCH was not observed in the Cꢀ
2
2
3
2
NMR. ESIꢀMS: expected (M+H) = 351.1907, found = 351.1898.
-Chloro, 6-N anilinopurine riboside 23 (white amorphous solid, 85% yield). HNMR (CD CN):
2
3
28
ACS Paragon Plus Environment