Integration of Enhanced Sampling Methods with Saturation Transfer Difference Experiments to Identify Protein Druggable Pockets

Article abstract of DOI:10.1021/acs.jcim.7b00733

Saturation transfer difference (STD) is an NMR technique conventionally applied in drug discovery to identify ligand moieties relevant for binding to protein cavities. This is important to direct medicinal chemistry efforts in small-molecule optimization processes. However, STD does not provide any structural details about the ligand-target complex under investigation. Herein, we report the application of a new integrated approach, which combines enhanced sampling methods with STD experiments, for the characterization of ligand-target complexes that are instrumental for drug design purposes. As an example, we have studied the interaction between StOASS-A, a potential antibacterial target, and an inhibitor previously reported. This approach allowed us to consider the ligand-target complex from a dynamic point of view, revealing the presence of an accessory subpocket which can be exploited to design novel StOASS-A inhibitors. As a proof of concept, a small library of derivatives was designed and evaluated in vitro, displaying the expected activity.

Full text of DOI:10.1021/acs.jcim.7b00733

Subscriber access provided by - Access paid by the | UCSB Libraries
Article
Integration of enhanced sampling methods with Saturation Transfer
Difference (STD) experiments to identify protein druggable pockets.
Joana Magalhaes, Giannamaria Annunziato, Nina Franko, Marco
Pieroni, Barbara Campanini, Agostino Bruno, and Gabriele Costantino
J. Chem. Inf. Model., Just Accepted Manuscript • DOI: 10.1021/acs.jcim.7b00733 • Publication Date (Web): 26 Feb 2018
Downloaded from http://pubs.acs.org on February 28, 2018
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 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 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 Chemical Information and Modeling 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 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
Integration of Enhanced Sampling Methods with Saturation Transfer Difference (STD)
Experiments to Identify Protein Druggable Pockets.
Joana Magalhães,†,■ Giannamaria Annunziato,†,■ Nina Franko,‡ Marco Pieroni,† Barbara
Campanini,‡ Agostino Bruno,†,#,* and Gabriele Costantino†
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
†
P4T group, Food and Drug Department, Parco Area Delle Scienze 27/A ꢀ 43124, Parma, Italy
‡Laboratory of Biochemistry and Molecular Biology, Food and Drug Department, Parco Area Delle
Scienze 23/A ꢀ 43124, Parma, Italy
#
Experimental Therapeutics Program, IFOM – The FIRC Institute for Molecular Oncology
Foundation, Via Adamello 16 ꢀ 20139 Milano, Italy
These authors equally contributed to the work
■
* Corresponding Authors.
Eꢀmail: agostino.bruno@unipr.it; agostino.bruno@ifom.eu
1
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 2 of 54
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
Abstract
Saturation Transfer Difference (STD) is an NMR technique conventionally applied in drug
discovery to identify ligand moieties relevant for binding to protein cavities. This is important to
direct medicinal chemistry efforts in smallꢀmolecule optimization processes. However, STD does
not provide any structural details about the ligandꢀtarget complex under investigation. Herein, we
report the application of a new integrated approach, which combines enhanced sampling methods
with STD experiments, for the characterization of ligandꢀtarget complexes that are instrumental for
drug design purposes. As an example, we have studied the interaction between StOASSꢀA, a
potential antibacterial target, and an inhibitor previously reported. This approach allowed us to
consider the ligandꢀtarget complex from a dynamic point of view, unrevealing the presence of an
accessory subꢀpocket which can be exploited to design novel StOASSꢀA inhibitors. As a proof of
concept, a small library of derivatives was designed and evaluated in vitro, displaying the expected
activity.
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
ACS Paragon Plus Environment

Page 3 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
Introduction
1
Saturation Transfer Difference (STD) is a relatively straightforward NMR technique based on the
exploitation of the nuclear Overhouser effect (nOe) observed during the interaction between a
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
2
protein and a small molecule ligand. STD methods can be applied only when particular conditions
occur, that is when the residence time of the ligand into a protein binding site is long enough to
permit spin diffusion through nOe, but, at the same time, short enough to allow ligand dissociation
and interchange between bound and unbound ligand states within the timeframe of the NMR spin
d
relaxation. Therefore, STD methods can be applied to weak ligands, usually with a K between 1
and 100 µM.
Under these conditions only ligand protons in close contact with the protein atoms will receive spin
saturation by nOe and a difference spectrum can be recorded by subtracting the offꢀresonance (IO)
2
spectrum from the onꢀresonance (ISAT) spectrum. Since nOe occurs through space, only those
protons close enough to protein atoms will show higher signal differences, while protons distant
from protein atoms will display low or null signal differences.
STD has received interest in the field of drug design because it can be conveniently employed as a
2–4
rapid and cheap screening technique. Indeed, small molecules which do not bind to the target
protein will not produce STD signals, since no spin saturation will be transmitted to the ligands;
also aspecific binders can be easily eliminated, since the fast dissociation time will preclude
effective spin saturation transfer. STD experiments only require a few milligrams of the soluble
protein and the ligand, while the spectra can be produced in a few hours by using routine NMR
instrument. Taken together, all these characteristics make STD a reliable tool for screening
mediumꢀsized libraries of fragments or small molecules.
The main limitation of this technique, which has hampered its wider use in classical drug design
pipeline, is the lack of specificity for the obtained information. The detection of a STD signal only
indicates that a given ligand binds to the protein, but no information about the nature of interactions
3
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 4 of 54
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 provided and information about the interaction with specific binding pocket(s) can be obtained
only if STD experiments are complemented with competition experiments. Therefore, STD can
only be used for identifying hit structures and ligand moieties important for binding through Group
Epitope Mapping (G.E.M.), which helps directing medicinal chemistry efforts in the hit expansion
and hit to lead optimization process. Indeed, the identification of the binding epitopes can be used
to enable complementary approaches. In one of such cases, when the 3D structure of the target
protein is known, the experimentally recorded STD intensity can be used for guiding (or ranking)
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
6,7
docking poses obtained by docking algorithms (CORCEMA, PLANTS, ). In the common case of
multiple binding poses with similar scores, the ambiguity can be solved by ranking the poses
according to their coherence with STD signals. Operatively, this can be carried out by introducing a
penalty function in the docking scoring function, which assigns low scores to those poses which are
not in agreement with the experimental STD signals. The approach is however not largely
applicable, since it requires at least preliminary knowledge about the binding pocket and the mode
of interaction of the ligand with the target protein, a condition that seldom takes place. Taking into
account all of these considerations, in this work we present a new method to integrate the STD
technique with enhanced sampling methods, in order to broaden the scope of STD application to the
hit optimization process.
5
6,7
While in the aforementioned approaches (CORCEMA, PLANTS ) the STD signals are used to
bias the docking function, we decided to use molecular dynamics (MD) to reconstruct the STD
spectrum. The rationale behind this idea is that STD signals are the result of magnetization transfer
from protein atoms to ligand protons (through space, not through chemical bonds). Therefore, only
those ligand protons residing close enough and in contact with protein atoms, and for a time
comparable with the spin saturation transfer timeframe will receive spin diffusion. The information
derived from the observed nOe effects can be used to derive interꢀatomic spatial information (e.g.
8
distance between atoms). Indeed, the higher the intensity of the STD signal, the closer is the ligand
proton to the protein environment, and in turn higher is the number of contacts between that ligand
4
ACS Paragon Plus Environment

Page 5 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
proton and the protein environment. This concept is similar to that already implemented in
6,7
PLANTS, where a defined distance cutꢀoff is used to generate distance hierarchies so as to score
the docking poses.
MD simulations have proven to be successful tools in classical drug discovery projects, allowing
implementation of SBDD strategies that fully account for the dynamic properties of relevant
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
9
ꢀ12
proteinꢀligand complexes.
In this context, we reasoned that if the molecular dynamics of the
ligand within the protein cavities is simulated for a sufficient amount of time, a distance cutꢀoff can
be used to define the number of contacts that each proton atom is establishing with the protein
environment. The normalization of the number of contacts for the ligand proton characterized by
the highest number of contacts (the proton closest to the protein atoms) provides with a measure
useful to rank the ligand protons, which can be compared with the experimental STD.
In an ideal setting, if the simulation time is prolonged to infinitum, the ligandꢀprotein complex will
be able to explore every configurational microstate and the most stable state will be populated for a
longer time, according to the Boltzmann distribution and ergodic hypothesis. This state will
contribute to the largest extent to the number of contacts between ligand protons and protein atoms
and this, in turn, will drive the computed STD profile. If the computed STD is in agreement with
that experimentally determined, this means that the MD simulations converged correctly, sampling
the configurational space, and thus it can be employed for further ligand optimization purposes. The
obvious pitfall of this approach is that, to ensure convergence, an unbiased molecular dynamics
simulation should cover the timeframe of the spin saturation transfer effect, which is in the range of
milliseconds to seconds time. This limitation can be overcome by using enhanced sampling
13
14
15
techniques (e.g. steered molecular dynamics, umbrella sampling, metadynamics and its derived
16
17–21
forms, ) which can be used to describe at atomic scale the behaviour of complex systems.
22
Among these, we decided to exploit FunnelꢀMetadynamics (FM, a derived metadynamics form),
an emerging technique that has proven to be successful in accelerating the number of backꢀandꢀ
2
3–25
forth events between the ligand bound and free states.
In FM, the introduction of a funnelꢀ
5
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 6 of 54
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
shaped restraining potential limits the exploration of the solvated state outside the binding pocket,
without affecting the sampling into the binding region, thus allowing the thorough characterization
22
of the ligand binding process in a reasonably short wall clock time. Most importantly, in the FM
framework no a priori information about the ligand binding mode is required.
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
We thus used FM to reconstruct the relative STD profile for the real case of a proteinꢀligand
complex and used the so obtained data to gain insights into the binding mode of a series of ligands.
We decided to test our integrated approach using as the reference system Oꢀacetylserine
sulfhydrylase (OASS), a PLPꢀdependent enzyme for which we have already produced a set of
26–29
information, such as a consistent SAR for inhibitors and STD data.
OASS is a potential
antibacterial target, as it catalyses the last step of the cysteine biosynthesis pathway in bacteria, and
therefore the inhibition of the enzyme is expected to reduce the fitness of the bacteria under
3
0
oxidative stress conditions. Our group recently reported a series of OASS inhibitors based on the
2
6–28
cyclopropylꢀcarboxylate scaffold.
While the 3D structure of StOASS is known, no Xꢀray crystal
structure of StOASS in complex with smallꢀmolecule inhibitors has been reported so far and the
binding mode of inhibitors can only be predicted by docking experiments. However our experience
in the field indicates OASS as a particularly challenging system for docking algorithms, in view of
26–28
the wideness and flexibility of the binding pocket.
Thus, we selected OASSꢀA as a reliable test
case for our integrated STD/FM approach and we have selected one of our previously reported
31
31
inhibitors
as
a
test
ligand.
Indeed,
compound
12b
((1R,2S)ꢀ1ꢀethylꢀ2ꢀ
phenylcyclopropanecarboxylic acid, Chart 1) has an experimental K of ≃ 13 µM and the STD
d
31
profile with StOASSꢀA was previously reported, making it a suitable ligand to be used to test our
approach.
Therefore, we firstly proceeded to the reconstruction of the whole binding process between 12b and
StOASSꢀA, by applying the FM technique, which allowed us to accurately predict the binding free
energy for such inhibitor (with a 0.95 kcal/mol difference with respect to the experimental one) and
to obtain hundreds of thousands binding conformations. The binding conformations were rescored
6
ACS Paragon Plus Environment

Page 7 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
by an inꢀhouse built software (STDc) able to compare the experimental STD profile with that
recalculated for each FMꢀderived binding conformation. This approach also allowed us to consider
the ligandꢀtarget complex from a dynamic point of view. Indeed, to better clarify the nature of the
complex under investigation a small ensemble of proteinꢀligand complexes (3 structures) were
extrapolated by means of STDc. These representative structures allowed us to rationalize the SAR
profile of previously published StOASSꢀA inhibitors and, more importantly, to identify a newly
druggable subꢀpocket, which is conserved across different OASS isoforms and in different
orthologues of the enzyme. These findings were used to synthesize a number of derivatives that
displayed the expected activity. More importantly, the disclosure of a new druggable subꢀpocket
paves the way to further extend the chemical space of our class of inhibitors, addressing at the same
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
32
time the soꢀcalled multiꢀparametric optimization process.
Results and Discussion
Molecular Modelling.
Unbiased MD simulation. In a previous work we reported a docking derived complex between
3
1
StOASSꢀA and compound 12b. The StOASSꢀA/12b complex was obtained by combining the
complete relaxation and conformational exchange matrix analysis of saturation transfer
5
(
CORCEMAꢀST) theory and docking studies. Even if this complex was instrumental to rationalize
preliminary SAR studies, it suffered from some limitations, which made the model useless in the
view of further ligand optimization. Indeed, differences between the experimental STD profile and
31
the calculated STD were observed. Such differences were ascribed to the fact that some of the
ligand moieties (particularly the aromatic ring) were in close proximity to a flexible loop delimiting
the binding pocket, which can adopt different conformations in solution. STD signals are sensitive
to protein flexibility, since they are the result of an average signal throughout time. On the contrary,
in the docking experiments the conformation of the aforementioned loop was fixed. Therefore, to
accurately take into account the effect of the protein dynamics on the calculated STD and to better
7
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 8 of 54
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
clarify the interaction mechanism of compound 12b, we decided to consider the StOASSꢀA/12b
complex from a dynamic point of view, by performing preliminary MD simulations studies.
Unfortunately, preliminary 120ns MD simulations revealed that the selected complex was not stable
under our simulation conditions, since the ligand adopts a different binding conformation after few
ns (see Figure S1). Ligand instability is likely due to the fact that StOASSꢀA is a highly flexible
enzyme that upon ligand binding can adopt significantly different conformations. So far, three
StOASSꢀA Xꢀray crystal structures have been disclosed: (i) a full open ligand free conformation
(pdb code: 1OAS) (ii) StOASSꢀA in an intermediate conformation in complex with chloride and
sulfate anions (pdb code: 1FCJ (chain A)); and (iii) a K41A mutant close conformation (pdb code:
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
1D6S). In a previous work we demonstrated that StOASSꢀA can adopt different conformations with
28
respect to the fully open state, depending on the presence and the nature of the inhibitors. Such
conformations are significantly different from those observed in the Xꢀray crystal structures. Thus,
we cannot firmly exclude that compound 12b can adopt a peculiar binding mode into the StOASSꢀ
A active site, which cannot be properly described by simply relying on the known Xꢀray crystal
structures. In this context, enhanced sampling methods such as FM can be useful in elucidating the
binding mechanism of a ligand to a target of interest, providing a better picture of the dynamic
behavior of such complex.
Analysis of the Ligand Unbinding Pathway and Free-Energy Calculation Using FM simulations. In
order to obtain preliminary hints on the most probable ligand unbinding pathway, and, in turn, on
how the funnelꢀrestrained potential would be positioned in the next stages, Random Accelerated
33
Molecular Dynamics (RAMD) were conducted. This method allows for the identification of the
most probable unbinding pathway offering two relevant advantages: (i) low computational cost and
(ii) objective pathway search. RAMD was successfully applied to disclose unbinding pathways for
34–38
several protein targets.
In Figure 1A a cartoon representation of the most probable unbinding
pathway for compound 12b is reported. In particular, in six out of ten RAMD simulations the ligand
8
ACS Paragon Plus Environment

Page 9 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
egressed following similar unbinding pathways, in three simulations partial unbinding events, along
the same unbinding route, were obtained. Finally, in one simulation no unbinding events were
observed (Figure S2). After that, the funnelꢀrestrained potential was positioned so that the
cylindrical portion of the funnel could include the most probable unbinding pathway (Figure 1B).
FM allows the wellꢀcharacterization of the freeꢀenergy profile since it increases the number of
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
22
binding events observed during the sampling. Moreover, when FM is applied, no a priori
information about the ligand binding mode is required, thus the exploration of the binding site is
22
possible also considering slow protein motions and solvent effects. At the end of the simulations,
the free energy landscape can be derived from the historyꢀdependent bias potential added on a few
degrees of freedom of the system, namely collective variables (CVs). The FM simulations were
performed for 1.5 µs applying the wellꢀtempered metadynamics bias on two CVs: (i) the distance
between the N backbone atom of Thr72 and the C atom of the carbonyl group of 12b and (ii) the
dihedral angle defined by N(Thr72)ꢀN(Asn71)ꢀC2(12b)ꢀC1(12b) atoms (Figure 1C and
Supplementary Information). The Gaussian deposition rate and heights were chosen so as to
smoothly conduct FM simulations (see Materials and Methods section), thus allowing protein
adaptation upon ligand binding. The statics of two furthers CVs, the projection along the main z
axis of the funnel and the distance from that axis, were collected along the FM simulations to
evaluate the conformational space explored by the ligand inside the funnel. As shown in Figure 2Aꢀ
B, several reꢀcrossing events (bound to unbound and viceversa) were observed (See also
1
Supplementary Movie). Indeed, during simulation the ligand visits several times the bound (CV <
1
14 Å) and the unbound states (33 <CV > 36 Å). Convergence of the calculations was ensured by
0
23–25
monitoring the bias deposition and ꢁG
b
(Figure 2CꢀD), as reported elsewhere.
FES (Figure 2B)
can be reconstructed as a function of the bias added on the system and once the calculation is
converged the absolute binding freeꢀenergy can be calculated as a function of the bias added on the
system and the absolute binding freeꢀenergy can be calculated.
9
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 10 of 54
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
Absolute Binding Free-Energy. The quality and reliability of the free energy profile is tightly linked
22
to the number of reꢀcrossing events observed between different visited states. As depicted in
1 1
Figure 2A the system explores several times both bound (CV < 14 Å) and unbound (33 < CV >
0
3
6 Å) states. ꢁG was calculated as the freeꢀenergy difference of the system with the ligand in the
b
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
bound and unbound state (Figure 2C) and from the potential mean force W(z) (Figure 2D) obtained
0
b
from metadynamics (see Material and methods and Figure S3). The estimated ꢁG
for compound
12b converges to ꢀ12.15 ± 1.09 kcal/mol (Figure 2). Considering the analytical correction of 3.8
kcal/mol due to the funnelꢀrestrained potential (see eqs. 1 and 2 in Material and Methods), the final
0
b
free energy is ꢀ8.35 ± 1.09 kcal/mol. The calculated ꢁG is in good agreement with that derived
from eq. 4 (see Material and Methods), which corresponds to ꢀ9.30 kcal/mol. Moreover, the
8.35 ± 1.09 kcal/mol value is in line with the experimental ꢁG values measured for ligands having
a K falling in the high nmꢀlow µM range, which is typical for weak binders as 12b.
ꢀ
b
2
2,39
d
STDc re-scoring and hit-target complexes ensemble. We previously reported the experimental STD
3
1
profile of compound 12b. In that work, the saturation received by the different ligand protons was
expressed as Group Epitope Mapping (G.E.M.), e.g., the percentage of STD signals normalized
with respect to the most saturated signal. The higher the value, the more intimate is the recognition
40
of the ligand portion by the protein binding pocket. Even if such data revealed relevant
information about which ligand portions mostly interact with the enzyme active site, the structural
details of the complex still remain elusive. Nevertheless, the experimental STD profile can be used
to extract, from the FM derived bound state minima, the conformation(s) that closely resemble the
experimental profile, providing us with reliable binding conformation(s). In principle, it might be
possible to calculate the STD profile for each binding conformation derived from the FM bound
state and to compare the calculated G.E.M. with the experimental ones. To date, two main methods
have been reported in literature for the in silico calculation of the STD/G.E.M. profile. The first
method is mainly based on the full calculation of the relaxation and conformational exchange
1
0
ACS Paragon Plus Environment

Page 11 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
5
matrix (CORCEMA, and CORCEMAꢀST). Such method proved to be effective in identifying
41–43
reliable binding modes for different ligandꢀprotein complexes.
Nevertheless, CORCEMA and
CORCEMAꢀST have the limitation to be extremely timeꢀconsuming (several hours for one ligandꢀ
protein complex), and they consider the proteinꢀligand complexes from a static point of view, thus
preventing their application for challenging ligandꢀprotein complexes, where flexibility plays a
dominant role. The second method is represented by the PLANTS docking scoring function, where
an userꢀdefined empirical scalar value (WSTD) is introduced to define several cutꢀoff distances,
which are then used to classify proteinꢀligand contacts with respect to the experimental G.E.M.
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
6,7
values. The PLANTS methods, although faster compared to CORCEMA and CORCEMAꢀST,
present relevant limitations, since (i) flexibility is not explicitly treated and (ii) it strongly depends
on docking scoring functions, which are known to have limitations in ligand binding mode
identification and ranking. We managed to overcome these limitations by developing STDc, an inꢀ
house built script able to process thousands of binding conformations in few hours. The binding
conformations are then ranked by their root mean square errors (R.M.S.E) with respect to the
experimental G.E.M. values. STDc finally provides the structure of the binding conformations with
the closest G.E.M. values to the experimental ones and an ensemble of conformations which better
resemble the experimental STD profile. In the latter case the ensemble of ligand binding
conformations is projected as density surface into the enzyme active site, so as to better describe the
binding cavity regions likely explored by the ligand during the STD experiments (for full details see
Material and Methods and Supplementary Information).
First, we proceeded to the reconstruction of the FES along different CVs, using the reweighting
44
algorithm developed by Bonomi et al. This method allows the reconstruction of the FES using
CVs different from those biased during the metadynamic simulations. The projection along the
main z axis of the funnel and the distance from that axis were used as new CVs (Figure 3A). This
procedure allowed us to better identify all the bound states falling into a minima of ≃ 2 kcal/mol
(
74747 structures). These structures were then clustered for a total of 462 families, which were
1
1
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 12 of 54
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
processed by STDc. In Figure 3B, the calculated vs the experimental G.E.M. values are reported for
the single binding conformation, which closely resemble the experimental values and for the
ensemble. The plot reported in Figure 3B highlights how the combination of the FMꢀSTDc leads to
a general improvement of the G.E.M. profile reconstruction. Indeed, the average deviation from the
experimental G.E.M. profile (blue bars) is, respectively, equal to 0.11, 0.10 and 0.22 for the G.E.M.
profile derived from the single conformation (orange bars), ensemble (cyan bars), and that derived
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
1
from the docking studies (yellow bars), previously reported. Therefore, by applying our protocol,
we were able to computationally calculate a STD/G.E.M. profile with an error from the
31
experimental profile reduced by 50% with respect to our previous work. Most importantly, there
is quite a good agreement between the experimental profile and that derived from the ensemble.
This allowed us to collect a series of findings that previously could not be unravelled. It was
possible to: (i) identify H as the proton that shows the strongest interaction with the enzyme active
c
3
1
site, whereas in the previous work this information was missing; (ii) better characterize the
location of the pendant aromatic ring in position 2, for which it is possible to identify the following
trend H >H ≃H (proton atoms of the aromatic ring), whereas in the previous work the signal of
2
,6
3,5
4
H was not properly reproduced. This information is instrumental to correctly plan at which position
4
of the aromatic ring functional groups can be introduced; (iii) accurately rank the signals of the
protons of the cyclopropane ring with respect to their interaction with the active site, that can be
3
1
described as such: H
c b a
>H >H (in our previous work this was not possible due to the limitations of
considering the protein as a rigid entity); and (iv) the CH G.E.M. of the ethyl branch is properly
2
3
reproduced, differently from the CH that appears to be the only one which is completely missꢀ
ranked. This is likely due to the great degree of freedom of this group, which can freely rotate
inside the StOASSꢀA binding site, as highlighted by high standard deviation of the measurement.
The application of the FMꢀSTDc approach allowed us to identify the binding mode reported in
Figure 3C, whose stability was verified over 100ns of standard MD simulations (Figure 3D). The
complex resulted significantly more stable with respect to that used at the beginning of our study
1
2
ACS Paragon Plus Environment

Page 13 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
(
Figure S1). From the analysis of the complex it can be appreciated how the ligand is well
accommodated into the enzyme active site, establishing the following key interactions: (i) the
carboxylic moiety Hꢀbonds the NH proton of the backbone residue Thr72, Asn71 and the OH
26,27,31
groups of Thr72 and Thr68 sidechains, in agreement with what previously hypothesized;
(ii)
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
the pendant aromatic ring in position 2 protrudes in an accessory hydrophobic pocket previously
2
6
described; (iii) the pendant phenyl ring establishes a πꢀHꢀbond interaction with the Asn69
sidechain, (iv) the ethyl branch accommodates well into a hydrophobic pocket defined by the
methyl group of Thr72 and Thr177 sidechains and the LysꢀPLP coꢀfactor. Of greater interest is the
observation that during the FM simulations the enzyme readapts upon ligand binding, assuming a
fully closed conformation (Figure 3E). Such a conformational change is typical for transaminases
(foldꢀtype I) and the members of the βꢀfamily of PLPꢀdependent enzymes (Fold type II). The latter
class of enzymes is characterized by the presence of a large and a small flexible domain, separated
45–46
by a hinge region, which rearranges upon substrate binding.
The structural comparison of the
FMꢀSTDc derived complex with other members of the same PLPꢀfamily for which Xꢀray crystal
structures are available (Figure S4) supports the notion that the enzyme assumes a fully closed
conformation upon ligand binding. The reꢀadaptation of the enzyme structure observed during the
FM simulation was possible since we chose to perform a highlyꢀsmoothed FM protocol (see
Material and Methods), allowing protein adaptation during ligand binding. Indeed, a similar
28
conformational change was observed also in our previous work, where we applied standard MD
simulations, supporting the notion that such event is likely to occur at low energetic cost, if the
ligand properly occupies the enzyme active site. So far, the Xꢀray crystal structure of wildꢀtype
StOASSꢀA in complex with a competitive inhibitor in its closed conformation has never been
reported. To the best of our knowledge, our protocol allowed for the first time to propose a reliable
model structure describing StOASSꢀA in its closed conformation in complex with an inhibitor. Such
structural details are instrumental in order to aid further ligand optimization strategies, paving the
way for the identification of improved inhibitors.
1
3
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 14 of 54
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
SAR profile of previously reported inhibitors. To properly validate our protocol, we firstly
proceeded to the rationalization of the SAR profile of previously reported StOAASꢀA inhibitors.
The analysis of the G.E.M. profile derived from the ensemble reported in Figure 3B (cyan bars),
and the projection of the density surface of the same ensemble (Figure 4A, see STDc section in
Material and Methods for details) revealed that two positions can be exploited for functionalization:
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
the CH group on the ethyl branch and the aromatic ring on position 2 of the cyclopropane core.
3
31
Regarding the modification of the methyl group, this was already exploited in our previous work,
leading to the identification of more potent inhibitors. Of great relevance is the observation that our
protocol effectively identified such position as one of the most suitable to introduce bulkier
substituents, strongly supporting the reliability of the method. In Supplementary Figure S5 the
3
1
binding mode of previously reported OASS inhibitors is depicted along with a short description of
the SAR (See supplementary Information).
Design, Synthetic strategies and generation of derivatives. Once rationalized the SAR profile of
previously reported inhibitors we looked for a robust proof that the structural details obtained by
our integrated approach could be used to design new StOASSꢀA inhibitors with a desirable
pharmacological profile.
The analysis of the density surface reported in Figure 4A revealed the presence of an additional
accessory pocket, which can be filled by the introduction of bulkier substituents at the 3’ꢀposition.
Noteworthy, modifications of the pendant phenyl ring are not reported so far, and our protocol
identifies a possible region for further ligand derivatization and optimization. Moreover, the
comparison of the Xꢀray crystal structures so far available for the OASS enzymes revealed that such
pocket is conserved across different orthologues and isoforms (Figure S6).
To support the notion that the newly identified pocket (Figure 4) can be exploited to further
increase the chemical space of our ligand class, we compared the selected binding mode and the
1
4
ACS Paragon Plus Environment

Page 15 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
extrapolated density surface with the Xꢀray crystal structures of OASSꢀA from different organisms
and in complex with inhibitors (Figure 4BꢀD and Figure S6). The analysis revealed that in other
48
49
OASS from different organisms (e.g. Haemophilus influenzae and Mycobacterium tuberculosis )
this accessory pocket is conserved and it can be occupied by bulkier and hydrophobic substituents.
Encouraged by these results, we decided to set up a synthetic strategy to easily and quickly prepare
a small focused library of derivatives bearing substituents of different nature in position 3’, in order
to prove this position as the most prone to modification and therefore validate our approach. The
easiest strategy to modify compound 12b in position 3’ is the introduction of another aromatic
function (Figure 4E) by means of a Suzuki coupling reaction. In this way, a small library of
derivatives can be easily generated by using different boronic acid fragments. All the purchasable
boronic fragments from SigmaꢀAldrich and Maybridge were downloaded as sdf files and using the
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
50
Reactor tool of ChemAxon a library of 12b derivatives was generated. A total of 113 derivatives
were generated and docked into the StOASSꢀA conformation obtained from the STDꢀFM approach
(See Material and Methods for details and Figure S7). Among the aforementioned derivatives only
4 compounds were selected, taking into account their docking scores and visual inspection (See
Material and Methods for selection strategies and Figure S7). The dissociation constants for the
selected compounds are reported in Table 1. Interestingly, compounds 7a,c and d show comparable
activities against StOASSꢀA, while 7b shows a reduced ability to bind StOASSꢀA. To better
characterized our newly designed inhibitor we also resolved the racemic mixture of compound 7a,
leading to compounds 10a and 11a, which show a slight affinity improvement if compared to 12b.
3
1
Interestingly, in our previous work we have demonstrated that StOASSꢀA ligand requires a
defined stereochemistry in order to be properly accommodated in the active site, since usually one
31
of the enantiomers is more active than the other (12b vs. 12c, Table 1). We were pleased to notice
that the modifications planned at position 3' of the phenyl ring allowed to surmount such stereoꢀ
chemical requirements, since both enantiomers display equal affinity towards OASS (10a and 11a
vs. 12b and 12c). At this regard, we can speculate that the introduction of bulkier substituents at the
1
5
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 16 of 54
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
’ position can lead to energetically favourable contacts with the protein active site, overcoming the
stereoꢀchemical requirements of the pocket. Indeed, we have previously reported that StOASSꢀA
preferentially binds compounds with specific stereochemistry, whereas the other enantiomers
showed only residual affinity. This suggests the possibility that such geometrical requirements can
be surmounted if proper ligand modifications are introduced. Since the chiral resolution of the
racemate or an enantioselective synthesis are no longer required, a significant enhancement in the
preparation of analogues can be reached to expand the series and refine the SAR.
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
Chemistry.
Styrene oxide 2 was obtained from the commercially available 3ꢀbromostyrene 1 by oxidation with
51
mꢀchloroperbenzoic acid, according to well established protocol (scheme 1). Alkylation of triethyl
phosphonoacetate 3 with bromoethane in the presence of sodium hydride afforded ethyl 2ꢀ
(diethoxyphosphoryl)ꢀ3ꢀ(pꢀtolyl)propanoate 4 in good yield. WittigꢀHorner reaction between
precursors 2 and 4, in anhydrous dimethoxyethane at 90 ᵒC and in the presence of n-buthyllithium
as the base, allowed us to obtain the required key intermediate trans ethyl 2ꢀ(3ꢀbromophenyl)ꢀ1ꢀ
ethylcyclopropanecarboxylate 5. SuzukiꢀMiyaura crossꢀcoupling reaction between the proper
boronic acid and compound 5 gave derivatives 6aꢀd, that were treated with LiOH at 100 degrees
under microwave irradiation to afford the final compounds 7aꢀd in good overall yields.
Diastereoisomers 8a and 9a were obtained from racemic 7a, that was reacted in dichloromethane at
room temperature with chiral precursor (R)ꢀ(−)ꢀ2ꢀPhenylglycinol, according to a standard amide
synthesis procedure (scheme 2). After separation of the two diastereoisomers by flash column
chromatography, hydrolysis in acidic conditions of 8a and 9a gave the desired enantiomers 10a and
11a in good yields.
1
6
ACS Paragon Plus Environment

Page 17 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
Conclusion
We have herein reported for the first time the development of an integrated method based on the
combination of FM experiments and STD approaches. This method can be used to identify reliable
ligandꢀtarget complexes which can be exploited to drive and implement drug design strategies. As a
test case, we have applied the newly developed FMꢀSTD approach, to study the interaction of
StOASSꢀA and a weak inhibitor previously identified by us. Several limitations of the STD method
when used alone were clearly overcome by the use of this combined approach, which indeed
allowed to (i) elucidate the binding mechanism of 12b to StOASSꢀA and calculate the binding free
energy for such complex which is in good agreement with the experimental one; (ii) identify a
binding mode able to rationalize previously reported biological data; (iii) identify a subꢀpocket
conserved among different OASS orthologues and isoforms, which can be exploited to further
functionalize this chemical class. Unexpectedly, this study led to the synthesis of inhibitors that,
differently from the parent compounds, are not affected by strict stereochemical requirements,
firmly simplifying the overall synthetic process.
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
The results obtained support the notion that besides classical drug design strategies, the STDꢀFM
approach can be applied to relevant pharmaceutical targets for which only weak binders are
reported and structural details about the ligandꢀprotein complex are unknown. Moreover, our STDc
script can be easily implemented in classical MD simulations or docking experiments as rescoring
function in order to better rank different binding conformations of the same complex, and in turn, to
identify reliable ligandꢀtarget complexes. Overall, these findings suggest that STDꢀFM can be of
extremely high interest when applied to some of the most common medicinal chemistry cases, such
as in fragmentꢀbased drug design approaches or right after carrying out HTS campaigns that
furnishes a number of weakly binders. Moreover, STDꢀFM can be complementary applied in more
challenging contexts such as the identification of binding cavities of modulators for intrinsically
52
disordered proteins, for which conventional structureꢀbased approaches cannot be applied.
1
7
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 18 of 54
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
Nowadays almost all medicinal chemistry lab are integrated with NMR and computational facilities,
where STDꢀFM could be easily performed without any further technological improvements or cost,
providing at the same time useful structural hints to speed up the drug optimization process.
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
Experimental Section
Computational Methods.
StOASS-A/12b complex generation and Plain MD simulation. The initial StOASSꢀA/12b was
obtained by combining the complete relaxation and conformational exchange matrix analysis of
5
26
saturation transfer (CORCEMAꢀST) theory and docking studies, as previously reported. In
particular, the 1OAS Xꢀray crystal structure was used. The complex was prepared by applying the
53
protein preparation protocol available in Maestro9.1. The complex was further minimized using
5
3
54
Macromodel, the PRCG method, OPLSSꢀ2005 as force field and water as implicit solvent
model.
55
The system, thus obtained, was parameterized using ff14SB and gaff as force fields. 12b atomic
single charges were computed using Gaussian 09, HF/6ꢀ31** as theory level, (http://gaussian.com)
and fitted with restrained electrostatic potential (RESP). Atom type and parameters for 12b were
55
28
retrieved from gaff. The PLP cofactor was simulated in its neutral state as previously reported,
the partial charges of the PLP were assigned using Gaussian 09, UB3LYP/6ꢀ31G* as theory level,
http://gaussian.com) and fitted with restrained electrostatic potential (RESP). Atom type and
(
5
5
parameters for PLP were retrieved from gaff. The system was solvated using leap and the TIP3P
water model.
5
6
Plain MD simulations were performed using NAMD2.10 and a relaxation protocol based on six
steps. The system was initially equilibrated performing 50ns in the NPT ensemble, after that a
production phase of 120 ns was performed in the NVT ensemble.
1
8
ACS Paragon Plus Environment

Page 19 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
Random Accelerated Molecular Dynamics simulations (RAMD). Recently RAMD have been
successfully applied to disclose putative ligand unbinding pathways for relevant biological
34–38
targets.
In the RAMD framework a randomly oriented force (f) is applied to the center of mass
of a group of atoms (ligand) for a defined short amount of timeꢀsteps (N). Currently, RAMD can be
performed in two schemes: (i) pure RAMD simulations in which the randomlyꢀoriented
acceleration is applied continuously, and (ii) combined RAMDꢀMD simulations in which RAMD
steps alternate with standard MD steps. For the StOASSꢀA/12b complex the combined RAMDꢀMD
simulations scheme was applied. The starting configuration was represented by the complex
generated after the 120 ns of plain MD simulations, the constant force was applied on all the ligand
heavy atoms and ten independent RAMD simulations were performed. Each ligand unbinding
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
57
pathway was drawn by tracing the center of mass of the ligand heavy atoms using VMD1.9.2. The
funnelꢀrestrained potential was applied so as to include the most probable unbinding route.
Funnel-metadynamics. The starting configuration for FM was represented by the last frame of the
57
120ns of the production phase. PLUMED1.3 plugin was used to run metadynamics calculations.
Since the funnelꢀrestrained potential is fixed in the space, the StOASSꢀA diffusion was avoided by
imposing positional restrains on seven Cɑ atoms (Ala29, Ile44, Glu153, Val170, Val202, Val246,
1 2 1 2
Ile297). The bias was added on a distance (CV ) and dihedral angle (CV ). For CV and CV
Gaussian widths of 0.05 Å and 0.09 rad were respectively applied. The Gaussian deposition
ꢀ1
frequency of 1 Kcal mol every 2 ps was initially applied, and gradually decreased on the basis of
the adaptive bias with a ꢁT of 2700 K. During FM simulations, an upper limit of 31 Å and a lower
limit of ꢀ3 Å were imposed along the funnel z axis, so as to avoid ligand exit from the funnelꢀ
5
7
restrained potential. Trajectories were analysed with VMD1.9.2.
Absolute binding free energy. In FM simulations when the ligand is inside the cone portion of the
restraint potential, no external potential is applied, and the system works under the standard wellꢀ
1
9
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 20 of 54
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
tempered metadynamics framework. In this way, the potential does not affect the exploration of the
enzyme active site by the ligand. When the ligand is in the unbound state z > z a cylindrical
cc
2
2
restraint potential is applied. Using this protocol the binding constant K
b
in presence of the
restraint is given by
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
ꢀꢁ = ꢂꢃꢄꢈ
ꢊꢋꢌꢍꢎ[ꢏ
ꢍꢏꢓꢔꢕ
]
[1]
ꢐꢑꢒ
ꢉ
ꢅꢆꢇ
ꢖꢗꢘꢙ
2
where πRcyl is the surface of the cylinder used as restraint potential while the potential W(z) and its
value in the unbound state, Wref can be obtained from the PMF (Potential Mean Force) obtained
ꢀ
1
from FM calculation. β= (k
temperature of the system. K
b
b
T) is constant, where k is the Boltzman constant and T the
0
b
b
and ꢁG
are directly related through the formula reported below:
ꢜ
ꢛ
ꢛ
Δꢚ =ꢂ−ꢂ ln ꢂꢐ ꢝ ꢀ ꢒ
[2]
ꢁ
ꢁ
ꢎ
0
0
ꢀ3
where ꢁG
b
is the proteinꢀligand binding free energy, K
b
the binding constant and C = 1/1660 Å
is the standard concentration 1M. As reported in ref. 22 using eq. 1 in FM simulation eq. 2 can be
rearranged
ꢜ
ꢛ
ꢈ
ꢛ
Δꢚ = ꢂΔꢚ −ꢂ ꢞꢟꢠꢃꢄ ꢝ ꢡ
[3]
[4]
ꢁ
ꢅꢆꢇ
ꢎ
ꢛ
Δꢚ =ꢂ−ꢄꢢꢞꢟꢀ
ꢙꢣ
ꢁG is the free energy difference between the bound and unbound states, and the absolute proteinꢀ
0
0
b
ligand binding free energy (ꢁG ) is equal to ꢁG minus the analytical correction in eq. 3. ꢁG was
derived from eq. 4 using RT=0.57
2
0
ACS Paragon Plus Environment

Page 21 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
The reweighting algorithm. To identify the main ligand binding conformations falling into a
4
4
minima of ≃ 2 kcal/mol we applied a recently developed reweighting procedure. This method
allows the reconstruction of the FES using CVs different from those biased during the
metadynamics simulations. Once the metadynamics simulation has converged, using the newly
computed probability distribution, the FES can be reconstructed as a function of the newly selected
CVs. The projection along the main z axis of the funnel and the distance from that axis were used as
new CVs.
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
STDc. STDc is an inꢀhouse built script able to calculate the STD/G.E.M profile for a given ligandꢀ
target complex and to compare it with the experimental one. It is not a standalone script and it
5
5
59
58
works in conjunction with cpptraj, catdcd and plumed.
In the first step a MD trajectory is clustered using cpptraj and user defined parameters. In our case
the averageꢀlinkage cluster algorithm with a R.M.S.D. cutꢀoff of 1 Å (calculated on the ligand
heavy atoms and the Cɑ atoms of the StOASSꢀA small domain and the flexible loop defined by
215ꢀ240 residues) were used. Cpptraj turns out with representative structures of each cluster
obtained. Subsequently, every representative structures generated at the previous step is processed
58
by plumed. For each ligand hydrogen atom the contact with the protein environment are estimated
by using eq. 5.
ꢬ
ꢧ
ꢨꢩ
ꢜꢍꢦ
ꢫ
ꢫ
ꢓꢪ
ꢤꢐꢥꢒ =ꢂ∑ꢮ
ꢭ
ꢧ
ꢨꢩ
ꢜꢍꢂꢦ
ꢓꢪ
[5]
The switching function reported in eq. 5 takes in consideration two groups of atoms, the ligand
proton atom (i) and all the protein atoms (j). In this way s(d) measures if the ligand proton atom i is
2
1
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 22 of 54
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
found at a certain distance d from protein atoms j. When d is less or equal to r (distance cutꢀoff)
ij
ij
0
s(d) is equal to 1 and the contact with the atoms is formed, while for d > r the function decays
ij
0
smoothly to 0 and the nature of the observed contacts become less relevant.
0
We used n=6, m=12, r =3.0 so as for dij=10Å s(d) becomes 0 (See Supplementary Information and
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 S8). Indeed the distance cutꢀoff for the long range interaction during the MD simulation was
set to 10Å, which is also the distance cutꢀoff through which nOe effects is still observable in STD
2ꢀ6
experiments. At this point, a list reporting each ligand hydrogen atom with the corresponding
total sum of contacts is stored and the values are normalized for the hydrogen atom with the highest
number of contacts (the ligand proton atom closest to the protein atoms). For each hydrogen atom
the R.M.S.E value is calculated with respect to the corresponding experimental value and the
representative structures are ranked on the basis of their average R.M.S.E. Finally, STDc turns out
with the representative structure having the lowest average R.M.S.E. and with an ensemble of
representative structures extrapolated by means of a R.M.S.E. cutꢀoff defined by the users. The
ensemble of structures is then used to project a surface representation of the protein binding site
where the ligands establishes protein contacts that mostly resemble the experimental STD/G.E.M.
profile (See Supplementary Information and Figure S8).
Library Design and Compound selection for synthesis. Two library of commercially available
boronic acid fragments were retrieved from SigmaꢀAldrich (38) and Maybridge (164). These
fragments were optimized with the Standardizer tool of ChemAxon, which allows for the
geometrical optimization and protonation state evaluation. Duplicates were removed and the
optimized fragments were conjugated to 12b by means of Reactor and Suzuki coupling reaction.
The point of substitution on 12b was represented by the 3’ position on the pendant phenyl ring. A
total of 113 derivatives were obtained and used for the subsequent docking studies. The StOASSꢀAꢀ
12b binding conformation extracted with STDc was used as reference structure. The protein was
53
prepared with the protein preparation tool of Maestro and the docking grid was centred on 12b.
2
2
ACS Paragon Plus Environment

Page 23 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
5
3
The compound collection of derivatives was prepared with ligprep each docking run was carried
53
out with Glide using the standard precision (SP) method, and the van der Waals scaling factor of
nonpolar atoms was set to 0.8. For each derivative one binding conformation was generated. A
total of 4 compounds were selected for synthesis and in vitro evaluation on the basis of the docking
score and after visual inspection (Figure S7).
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
Chemistry
General Information. All the reagents were purchased from SigmaꢀAldrich and AlfaꢀAesar at
reagent purity and, unless otherwise noted, were used without any further purification. Dry solvents
used in the reactions were obtained by distillation of technical grade materials over appropriate
dehydrating agents. Reactions were monitored by thin layer chromatography on silica gelꢀcoated
aluminum foils (silica gel on Al foils, Supelco Analytical, Sigmaꢀ Aldrich) at both 254 and 365 nm
wavelengths. Where indicated, intermediates and final products were purified through silica gel
flash chromatography (silica gel, 0.040−0.063 mm), using appropriate solvent mixtures.
1
13
H NMR and C NMR spectra were recorded on a Bruker Avance spectrometer at 400 and 100
1
MHz, respectively, with TMS as internal standard. H NMR spectra are reported in this order:
multiplicity and number of protons. Standard abbreviation indicating the multiplicity was used as
follows: s = singlet, d = doublet, dd = doublet of doublets, t = triplet, q = quadruplet, m = multiplet
and br = broad signal. HPLC/MS experiments were performed with HPLC, Agilent 1100 series,
equipped with a Waters Symmetry C18, 3.5 ꢂm, 4.6 mm × 75 mm column; and MS, Applied
Biosystem/MDS SCIEX, with API 150EX ion source. HRMS experiments were performed with
LTQ Orbitrap XL Thermo.
All compounds were tested as 95−100% purity samples (by HPLC/ MS).
2
3
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 24 of 54
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
2ꢀ(3ꢀbromophenyl)oxirane (2) mꢀCPBA (994 mg, 5.76 mmol) was added to a solution of 3ꢀ
bromostyrene 1 (600.3 µL, 4.61 mmol) in chloroform (6.9 mL) stirring on ice. Reaction mixture
was stirred for 3h on ice and 17h at room temperature. As the reaction wasn't complete further 1 eq
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
(795mg, 4.61 mmol) of mꢀCPBA was added to the reaction mixture that was previously cooled
down in an ice bath. After stirring for another 3h at room temperature it was stopped with the
addition of a 1M solution of NaOH under ice cooling conditions. The product was extracted with
chloroform and the combined organic layers were dried over anhydrous sodium sulphate, filtered
and concentrated under reduced pressure to afford the product as a colourless oil in quantitative
yield.
1
H NMR (400 MHz, CDCl ) δ 7.42 (d, J = 4.3 Hz, 2H), 7.21 (d, J = 4.8 Hz, 2H), 3.89 – 3.72 (m,
3
1H), 3.21 – 3.08 (m, 1H), 2.76 (dd, J = 5.5, 2.5 Hz, 1H).
Ethyl 2ꢀ(diethoxyphosphoryl)butanoate (4): was prepared as previously described and the
51
analytical data matched with those already reported.
Transꢀethylꢀ2ꢀ(3ꢀbromophenyl)ꢀ1ꢀethylcyclopropaneꢀ1ꢀcarboxylate (5): was prepared as
51
previously described and the analytical data matched with those already reported.
General Procedure for SuzukiꢀMiyaura Reaction. Method A:
Into a flask under argon stirred at room temperature was added a solution of compound 5 (1.00
equiv, 0.52 mmol) in a mixture of toluene/methanol/water (80/18/2 v/v, 40mL/mmol),
tetrakis(triphenylphosphine)palladium (0.15 equiv, 0.08 mmol), phenylboronic acid (3.00 equiv,
2 3
1.56 mmol), and K CO 2M (1.6 equiv, 0.832 mmol). The reaction mixture was refluxed overnight
at 110 °C and then filtered through a plug of celite. The filtrate was concentrated under reduced
pressure and the residue purified by flash column chromatography on silica gel. Purity, yields and
elution conditions are reported below.
2
4
ACS Paragon Plus Environment

Page 25 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
Transꢀethyl 2ꢀ([1,1'ꢀbiphenyl]ꢀ3ꢀyl)ꢀ1ꢀethylcyclopropaneꢀ1ꢀcarboxylate (6a): Purified by flash
chromatography (petroleum ether/ Ethyl acetate 99:1); colourless oil; 81% yield.
1
H NMR (300 MHz, CDCl
3
) δ 7.66 – 7.54 (m, 2H), 7.51 – 7.29 (m, 6H), 7.18 (d, J = 7.6 Hz, 1H),
4.21 (dq, J = 7.1, 3.7 Hz, 2H), 2.87 (dd, J = 9.1, 7.1 Hz, 1H), 1.81 – 1.54 (m, 2H), 1.30 (t, J = 7.1
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
Hz, 3H), 1.22 (dd, J = 7.1, 4.6 Hz, 1H), 0.98 – 0.85 (m, 3H).
Transꢀethyl 2ꢀ(3ꢀ(2ꢀaminopyrimidinꢀ5ꢀyl)phenyl)ꢀ1ꢀethylcyclopropaneꢀ1ꢀcarboxylate (6b):
Purified by flash chromatography (petroleum/ethyl acetate from 75:25 to 50:50 + 0.1% of
thriethylamine); slightly yellow powder; 96% yield.
1
3
H NMR (300 MHz, CDCl ) δ 8.52 (s, 2H), 7.54 – 7.28 (m, 3H), 7.19 (d, J = 7.0 Hz, 1H), 5.25 (s,
2
H), 4.34 – 4.06 (m, 2H), 2.85 (t, J = 8.0 Hz, 1H), 1.77 – 1.55 (m, 2H), 1.30 (t, J = 7.1 Hz, 3H),
.19 (dd, J = 6.9, 4.8 Hz, 1H), 0.92 – 0.83 (m, 3H).
1
General Procedure for SuzukiꢀMiyaura Reaction. Method B:
A microwave tube was charged with [1,1'ꢀBis(diphenylphosphino)ferrocene]dichloropalladium(II)
(0.02 equiv, 0.0134 mmol), the proper boronic acid pinacol ester (2 equiv, 0.134 mmol), and
caesium carbonate (2 equiv, 0.134 mmol). The tube was purged with argon three times and then a
solution of compound 5 (1 equiv, 0.067 mmol) in a mixture of dimethoxyethane/water 5:1 (175 µL)
was injected in the reaction mixture that was then heated in a microwave reactor for 30 min at
140°C at 100 Watts power. Afterwards, reaction mixture was filtered through a plug of celite, the
filtrate was concentrated under reduced pressure and the residue purified by flash column
chromatography. Purity, yields and elution conditions are reported below.
Transꢀethyl 2ꢀ(3ꢀ(1Hꢀpyrazolꢀ4ꢀyl)phenyl)ꢀ1ꢀethylcyclopropaneꢀ1ꢀcarboxylate (6c): Purified by
flash chromatography (petroleum/ethyl acetate from 75:25 to 50:50 + 0.1% of triethylamine);
slightly orange oil; 42% yield.
1
H NMR (300 MHz, CDCl
3
) δ 7.67 – 7.28 (m, 5H), 7.04 (d, J = 6.5 Hz, 1H), 4.20 (ddd, J = 12.7,
7.1, 3.5 Hz, 2H), 2.87 – 2.69 (m, 1H), 1.77 – 1.59 (m, 2H), 1.35 – 1.24 (m, 4H), 1.17 (dd, J = 7.1,
2
5
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 26 of 54
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
4
.7 Hz, 1H), 0.95 – 0.77 (m, 3H). HRMS (ESI): calculated for C H N O [M+H] 285.15975
17 20 2 2
found 285.15948.
Transꢀethyl 1ꢀethylꢀ2ꢀ(3ꢀ(3ꢀmethylꢀ1Hꢀpyrazolꢀ4ꢀyl)phenyl)cyclopropaneꢀ1ꢀcarboxylate (6d):
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
Purified by flash chromatography (petroleum/ethyl acetate from 75:25 to 50:50 + 0.1% of
triethylamine); pale orange oil; 40% yield.
1
H NMR (300 MHz, CDCl ) δ 7.69 (bs, 1H), 7.37 – 7.26 (m, 3H), 7.12 (d, J = 6.5 Hz, 1H), 4.20
3
(ddd, J = 12.7, 7.1, 3.5 Hz, 2H), 2.99 – 2.93 (m, 1H), 2.47 (s, 3H) 1.79 – 1.73 (m, 2H), 1.35 – 1.24
(m, 4H), 1.17 (dd, J = 7.1, 4.7 Hz, 1H), 0.95 – 0.77 (m, 3H).
22 2 2
HRMS (ESI): calculated for C18H N O [M+H] 298,1681 found 298,3860.
General procedure for ethyl ester hydrolysis: a solution of the ethyl ester compound (1 equiv,
.59 mmol) in a mixture of THF/MeOH/H O in the proportions 3:1:1 was added to a microwave
0
2
tube previously charged with LiOH (4 equiv, 3.54 mmol). Reaction mixture was heated in a
microwave reactor at 100°C for 10 min. Then the solvent was evaporated and the residue was
acidified with HCl 3N. Extractions with ethyl acetate were made and the combined organic layers
were dried over anhydrous sodium sulphate, filtered and concentrated under reduced pressure to
give a crude material, that is purified by flash column chromatography on silica gel to give the
desired acid.
Transꢀ2ꢀ([1,1'ꢀbiphenyl]ꢀ3ꢀyl)ꢀ1ꢀethylcyclopropaneꢀ1ꢀcarboxylic acid (7a): Purified by flash
chromatography (dichloromethane/methanol 99:1); orange oil; 83% yield.
1
3
H NMR (300 MHz, CDCl ) δ 7.64 – 7.53 (m, 2H), 7.50 – 7.33 (m, 6H), 7.20 (d, J = 7.5 Hz, 1H),
3
2
.00 (t, J = 8.2 Hz, 1H), 1.84 – 1.73 (m, 1H), 1.73 – 1.63 (m, 1H), 1.28 (ddd, J = 20.4, 10.6, 5.9 Hz,
H), 1.04 – 0.79 (m, 3H).
2
6
ACS Paragon Plus Environment

Page 27 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
1
3
C NMR (75 MHz, CDCl ) δ 181.33, 140.94, 140.66, 136.89, 128.50, 128.36, 127.93, 127.88,
3
127.12, 126.90, 125.49, 33.14, 30.93, 21.34, 18.31, 11.38.
HRMS (ESI): calculated for C18 [M+H] 265.12231 found 265.12341.
18 2
H O
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
Transꢀ2ꢀ(3ꢀ(2ꢀaminopyrimidinꢀ5ꢀyl)phenyl)ꢀ1ꢀethylcyclopropaneꢀ1ꢀcarboxylic acid (7b):
Purified by trituration with diethyl ether; white powder; 50% yield.
1
H NMR (400 MHz, MeOD) δ 8.83 (s, 2H), 7.50 (d, J = 7.7 Hz, 2H), 7.45 (dd, J = 18.9, 11.5 Hz,
1H), 7.30 (d, J = 7.6 Hz, 1H), 2.97 – 2.81 (m, 1H), 1.70 – 1.58 (m, 1H), 1.54 (dd, J = 13.9, 7.0 Hz,
1H), 1.36 (dd, J = 6.9, 4.9 Hz, 1H), 1.02 – 0.90 (m, 1H), 0.91 – 0.74 (m, 3H).
1
3
C NMR (101 MHz, MeOD) δ 178.20, 156.96, 140.07, 133.28, 130.67, 130.32, 128.16, 125.62,
1
24.92, 32.98, 32.42, 22.71, 18.22, 11.84.
HRMS (ESI): calculated for C H N O [M+H] 282.12370 found 282.12488.
16
17
3
2
Transꢀ2ꢀ(3ꢀ(1Hꢀpyrazolꢀ4ꢀyl)phenyl)ꢀ1ꢀethylcyclopropaneꢀ1ꢀcarboxylic acid (7c):
Purified by flash chromatography (dichloromethane/methanol 99:1 + 0.1% of formic acid); slightly
yellow powder; 21% yield.
1
H NMR (400 MHz, CDCl ) δ 9.24 (s, 1H), 7.89 (s, 1H), 7.36 (d, J = 28.4 Hz, 3H), 7.09 (s, 1H),
3
2
.96 (s, 1H), 1.75 (d, J = 23.1 Hz, 1H), 1.25 (dd, J = 20.5, 8.0 Hz, 2H), 0.94 (s, 3H).
1
3
3
C NMR (101 MHz, CDCl ) δ 180.49, 147.76, 146.79, 137.77, 133.41, 132.29, 128.94, 127.50,
127.26, 124.50, 33.22, 31.40, 21.92, 18.49, 11.85.
HRMS (ESI): calculated for C H N O [M+H] 255.11280 found 255.11413.
15
16
2
2
Transꢀ1ꢀethylꢀ2ꢀ(3ꢀ(3ꢀmethylꢀ1Hꢀpyrazolꢀ4ꢀyl)phenyl)cyclopropaneꢀ1ꢀcarboxylic acid (7d):
Purified by flash chromatography (dichloromethane/methanol 99:1 + 0.1% of formic acid); light
yellow powder; 25% yield.
2
7
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 28 of 54
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
1
H NMR (300 MHz, CDCl ) δ 7.71 (bs, 1H), 7.35 – 7.23 (m, 3H), 7.15 (d, J = 6.5 Hz, 1H), 2.97 –
3
2.92 (m, 1H), 2.45 (s, 3H) 1.81 – 1.75 (m, 2H), 1.37 – 1.26 (m, 1H), 1.13 (dd, J = 7.1, 4.7 Hz, 1H),
.93 – 0.74 (m, 3H).
0
13
C NMR (101 MHz, CDCl
27.26, 122.48, 33.22, 31.40, 21.92, 20.23, 18.49, 11.85.
HRMS (ESI): calculated for C H N O [M+H] 270,1368 found 270,3320.
3
) δ 181.47, 147.72, 147.65, 138.23, 134.23, 133.49, 128.94, 127.50,
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
1
16
18
2
2
Synthesis of 2ꢀ([1,1'ꢀbiphenyl]ꢀ3ꢀyl)ꢀ1ꢀethylꢀNꢀ((R)ꢀ2ꢀhydroxyꢀ1ꢀphenylethyl)cyclopropaneꢀ1ꢀ
carboxamide (8a and 9a):
(R)ꢀ(ꢀ)ꢀ2ꢀPhenylglycinol (43.21 mg, 0.315 mmol), TBTU (67.42 mg, 0.21 mmol), EDC*HCl (60.4
mg, 0.315 mmol) and dry triethylamine (29.26 µL, 0.21 mmol) were added to a solution of
compound 7a in anhydrous dichloromethane, stirring on ice and under nitrogen. Reaction mixture
was stirred for 1h at 0°C and about 6h at room temperature until consumption of the starting
material as revealed by TLC. Then, a saturated solution of ammonium chloride was carefully added
and extractions with dichloromethane were performed. The combined organic layers were dried
over anhydrous sodium sulphate, filtered and concentrated under reduced pressure to be purified by
flash column chromatography on silica gel using a gradient of ethyl acetate in petroleum ether (9:1 ꢀ
1
:1).
8a: White solid (45% yield)
1
H NMR (400 MHz, CDCl ) δ 7.60 (d, J = 7.4 Hz, 2H), 7.53 – 7.28 (m, 10H), 7.17 (dd, J = 26.3,
3
8
.5 Hz, 1H), 6.57 (d, J = 6.6 Hz, 1H), 5.14 (dd, J = 11.2, 5.0 Hz, 1H), 3.93 (d, J = 4.8 Hz, 2H), 3.02
(s, 1H), 2.95 – 2.83 (m, 1H), 1.67 (dt, J = 13.9, 7.0 Hz, 1H), 1.56 (ddd, J = 15.6, 12.3, 5.5 Hz, 1H),
1
.24 – 1.08 (m, 2H), 1.00 – 0.88 (m, 3H).
9
a: White solid (16% yield)
2
8
ACS Paragon Plus Environment

Page 29 of 54
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
1
H NMR (400 MHz, CDCl ) δ 7.70 – 7.50 (m, 2H), 7.47 – 7.30 (m, 10H), 7.18 (dd, J = 13.3, 4.9
3
Hz, 1H), 6.55 (d, J = 6.6 Hz, 1H), 5.14 (dd, J = 11.4, 4.9 Hz, 1H), 3.90 (dd, J = 18.1, 5.3 Hz, 2H),
.92 – 2.82 (m, 1H), 1.69 (dt, J = 43.0, 21.5 Hz, 1H), 1.55 (td, J = 14.7, 7.1 Hz, 1H), 1.26 – 1.08
m, 2H), 0.94 (t, J = 7.4 Hz, 3H).
2
(
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
General procedure for amide hydrolysis:
The starting material (8a, 9a) (1 equiv, 0.047 mmol) was solubilized in a mixture 1:1 of 3N H SO /
2
4
Dioxane and stirred overnight at 100°C. Water was carefully added to the reaction mixture after
cool it down until room temperature. Extractions with dichloromethane were performed and the
combined organic layers were dried over anhydrous sodium sulphate, filtered and concentrated
under reduced pressure to be purified by flash column chromatography.
(ꢀ)ꢀTransꢀ2ꢀ([1,1'ꢀbiphenyl]ꢀ3ꢀyl)ꢀ1ꢀethylcyclopropaneꢀ1ꢀcarboxylic acid (10a): purified by
flash column chromatography (dichloromethane/methanol (99:1); colourless oil; 56% yield.
1
H NMR (300 MHz, CDCl
3
) δ 7.61 (d, J = 7.2 Hz, 2H), 7.53 – 7.30 (m, 6H), 7.23 (t, J = 8.6 Hz,
1
H), 3.02 (t, J = 8.1 Hz, 1H), 1.87 – 1.75 (m, 1H), 1.75 – 1.66 (m, 1H), 1.39 – 1.23 (m, 2H), 0.95
(d, J = 2.4 Hz, 3H).
1
3
3
C NMR (101 MHz, CDCl ) δ 182.03, 141.38, 141.10, 137.32, 128.92, 128.78, 128.35, 128.29,
1
27.54, 127.32, 125.91, 33.58, 31.39, 21.76, 18.74, 11.79.
HRMS (ESI): calculated for C18 [M+H] 265.12231 found 265.12332.
18 2
H O
(+)ꢀTransꢀ2ꢀ([1,1'ꢀbiphenyl]ꢀ3ꢀyl)ꢀ1ꢀethylcyclopropaneꢀ1ꢀcarboxylic (11a): purified by flash
column chromatography (dichloromethane/methanol (99:1); Colourless oil (42% yield).
1
3
H NMR (300 MHz, CDCl ) δ 7.59 (d, J = 7.3 Hz, 2H), 7.53 – 7.32 (m, 6H), 7.20 (d, J = 7.6 Hz,
1
0
H), 3.00 (t, J = 8.1 Hz, 1H), 1.86 – 1.74 (m, 1H), 1.74 – 1.63 (m, 1H), 1.37 – 1.24 (m, 2H), 0.99 –
.76 (m, 3H).
2
9
ACS Paragon Plus Environment