Subscriber access provided by UNIV OF NEW ENGLAND ARMIDALE
Molecular Mechanics
OPLS3e: Extending Force Field Coverage for Drug-Like Small Molecules Katarina Roos, Chuanjie Wu, Wolfgang Damm, Mark Reboul, James M. Stevenson, Chao Lu, Markus K. Dahlgren, Sayan Mondal, Wei Chen, Lingle Wang, Robert Abel, Richard A Friesner, and Edward D. Harder J. Chem. Theory Comput., Just Accepted Manuscript • DOI: 10.1021/acs.jctc.8b01026 • Publication Date (Web): 15 Feb 2019 Downloaded from http://pubs.acs.org on February 15, 2019
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.
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 53 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
Journal of Chemical Theory and Computation
OPLS3e: Extending Force Field Coverage for DrugLike Small Molecules Katarina Roos2,3, Chuanjie Wu1, Wolfgang Damm1, Mark Reboul1, James M. Stevenson1, Chao Lu1, Markus K. Dahlgren1, Sayan Mondal1, Wei Chen1, Lingle Wang1, Robert Abel1, Richard A. Friesner2, Edward D. Harder*1 1Schrodinger,
Inc., 120 West 45th Street, New York, New York
10036, United States,
2Department
of Chemistry, Columbia
University, 3000 Broadway, New York, New York 10027, United States, 3Department of Cell and Molecular Biology, Uppsala University, Biomedical Centre, BOX 596, SE-751 24 Uppsala, Sweden
ABSTRACT
Building upon the OPLS3 force field we report on an enhanced model, OPLS3e, that further extends its coverage of medicinally relevant chemical
space
transferability.
by
addressing
limitations
in
chemotype
OPLS3e accomplishes this by incorporating new
ACS Paragon Plus Environment
1
Journal of Chemical Theory and Computation 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
Page 2 of 53
parameter types that recognize moieties with greater chemical specificity
and
integrating
an
on-the-fly
approach to the assignment of partial charges.
parameterization As a consequence,
OPLS3e leads to greater accuracy against performance benchmarks that assess small molecule conformational propensities, solvation and protein-ligand binding.
I. Introduction Free
energy
approaches
that
utilize
all-atom
molecular
dynamics simulations are emerging as a powerful tool in the design of new molecules and materials. Spurring interest in these methods is the promise of a high degree of predictive accuracy owing to the
physical
rigor
with
which
these
methods
account
for the
composite atomistic and configurational complexity of molecular systems. ability to
In principle, their accuracy is only limited by the achieve
adequate
phase
space
sampling
and
by the
veracity of the model potential. Because of their speed, the class of approximate molecular models commonly referred to as force fields offer a tractable means for applying free energy methods to the study of phenomena in complex biological and materials systems. A notable example is the application of free energy perturbation theory (FEP) to calculate ligand binding potencies in structurebased drug discovery projects.1-5
The adoption of this technology
in
traction
active
projects
is
gaining
ACS Paragon Plus Environment
as
advances
to
the
2
Page 3 of 53 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
Journal of Chemical Theory and Computation
technology result in greater automation, delivering more robust results
with
greater
expediency.6-11
As
the
capacity
of
the
technology continues to improve and sees greater usage on a wider regime of systems it is imperative that the development of the force field continues apace to ensure meaningful and predictive results. Force fields aim to strike a balance between computational expediency and physical salience and accomplish this goal by employing simple potential functions, such as, harmonic terms to represent bond vibrations and fixed-point charges to approximate electrostatic interactions.12-13 These terms depend explicitly upon sets
of
empirically
determined
pertinent physical properties.
parameters
tuned
to
reproduce
Target parameterization data is
typically comprised of quantum mechanical and experimental data for representative and diverse small organic molecules. effective,
a
force
field
needs
to
provide
an
To be
accurate
parameterization of all components of the system that can impact a property of interest. For ligand-protein binding, a particularly noteworthy challenge is the development of a force field that achieves a uniformly high level of accuracy (on the order of 1 kcal/mol) over the many diverse chemistries found within drug-like molecules. Early efforts to build force fields for drug-like small molecules have tended to focus on parameter generalization.14-17 That
is,
automating
the
process
by
which
ACS Paragon Plus Environment
a
set
of
existing
3
Journal of Chemical Theory and Computation 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
Page 4 of 53
parameters can be translated into assignments for any arbitrary ligand.
Less effort has been placed on building the extensive
basis of parameters needed to cover drug-like chemical space at a high
level
of
accuracy.
The
recent
development
of
OPLS318
represents perhaps the largest concerted parameterization effort aimed at traversing this space and it forms the starting point for the work discussed herein.
To review, notable features of the
OPLS3 model include: 1. An improved protein force field. 2. Extensive
parameterization
of
valence
and
torsional
terms found in drug-like molecules supplemented by an automated version of the fitting protocol (FFBuilder) used to fill coverage gaps. 3. Improvements to the charge model including virtual (offatom centered) sites to better represent lone pair and sigma hole charge distributions. These advances translated to significant improvements in proteinligand binding accuracy using FEP.
Relative to OPLS2005, OPLS3
resulted in a 30% improvement in aggregate RMS error.
Despite
these gains, the current prediction error continues to exceed the expected
experimental
indicating
additional
uncertainties room
for
by
a
improvement
significant in
the
margin,
technology
remains.
ACS Paragon Plus Environment
4
Page 5 of 53 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
Journal of Chemical Theory and Computation
The aim of the work presented herein is to address additional outstanding
sources
of
force
field
error
significantly to the residual FEP error.
that
contribute
In particular, we’ve
found that sub-optimal parameter transferability across shared chemotypes
impacting
parameterizations
can
both
the
torsion
significantly
impact
and FEP
partial
charge
accuracy.
To
address this, a new version of the force field, OPLS3e, has been developed.
Notable improvements include:
A more accurate and transferable torsional parameterization via an extensive optimization of torsion types.
Integration of a ligand-specific, on-the-fly partial charge assignment approach.
The paper is organized as follows.
In section 2, we present
two representative cases that illustrate the source of force field error resulting in significant binding free energy mispredictions. Section 3 discusses the OPLS3e development.
Part 3.A provides a
brief overview, Part 3.B presents the new torsion model including the process of refining torsion types, the extended data sets that were
required
to
train
the
model
and
the
performance
on
conformational energy benchmarks. Part 3.C presents the new charge assignment method and also includes a discussion of the simple condensed phase data sets used to develop and validate the model. In Section 4 we discuss the protein-ligand binding validation. This data set includes a significant extension from previous
ACS Paragon Plus Environment
5
Journal of Chemical Theory and Computation 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
reports,1,
18
Page 6 of 53
with new ligand and receptor data collected with the
expressed aim of presenting challenging chemotype transferability cases for the force field.
Section 5 then summarizes the results
and discusses future directions.
II.
OPLS3 Force Field Outliers Suboptimal
assignment
of
torsion
and
partial
parameters can lead to erroneous free energy calculations.
charge Panel
A of Figure 1 shows an example of an FEP+ misprediction that arose in a drug discovery program that was due to sub-optimal torsion parameters (only details that directly pertain to the outlier investigation are presented). of
an amide substituent
The ligands differ by a replacement
with an ester.
Experimentally, the
relative binding free energy of the ligands is approximately zero. However,
the
OPLS3-based
FEP+
prediction
significantly
underestimates the amide-pyrrole binding free energy relative to the ester.
An indicator of the underlying model issue driving
the erroneous prediction is shown in Panel B.
The top fragment
represents the substructure of the amide-pyrrole ligand in the FEP calculation. The plot contrasts the associated QM and MM potential energy
scanning
the
dihedral
angle
connecting
the
amide
substituent to the pyrrole ring defined by the NCcn atoms (aromatic atoms in lower case).
The predominant rotamer state in solution
corresponds to an NCcn angle ~ 0.
The predominant state in the
ACS Paragon Plus Environment
6
Page 7 of 53 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
Journal of Chemical Theory and Computation
complex corresponds to an NCcn angle ~ 180 which brings the NH of the amide and pyrrole into close proximity facilitating a bidentate contact with a nearby histidine.
As such, to adopt the bioactive
conformation,
derivative
the
amide-pyrrole
degree of conformational work.
needs
to
pay
some
Based on the potential energy
comparison to QM, OPLS3 appears to significantly overestimate this contribution affinity.
consistent
with
the
observed
To verify this hypothesis,
under-predicted
we re-ran the
binding
experiment with a slight modification to OPLS3 (denoted OPLS3*) where the amide-pyrrole torsion of 2-amide-3 methyl pyrrole is refit to match QM. The associated relative amide-to-ester free energy of
OPLS3*
is
-0.23
kcal/mol,
in
excellent
agreement
with
experiment. To explain why OPLS3 errs in matching QM for this dihedral, the training set molecule used to fit the OPLS3 amide-pyrrole dihedral, is shown in panel B of Figure 1 (bottom fragment). Though chemically similar to the ligand substructure of interest, the training set molecule does contain two crucial differences that
ultimately
result
in
poor
quantitative
transferability to the moiety of interest.
parameter
The pyrrole nitrogen
is capped by a methyl group instead of hydrogen and the training set molecule lacks an additional ortho-methyl group.
Because both
features lie one bond away from the atoms used to define the OPLS3 torsion, the same parameters are assigned to both motifs.
ACS Paragon Plus Environment
The
7
Journal of Chemical Theory and Computation 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
Page 8 of 53
modified version of the force field, OPLS3*, which was trained to the 2-amide-3-pyrrole fragment, shows a similar level of error when
applied
to
the
OPLS3
training
set
molecule,
further
illustrating that a common set of parameters can’t reconcile the energetics
of
both
moieties.
To
resolve
this
incompatibility
issue, the assignment algorithm would need to distinguish between these motifs by taking into account an extended chemical regime that includes these ortho positioned features. Additional
examples
emerged
following
a
similar
theme:
differences in chemistry unaccounted for by the assigned torsion parameter type lead to sub-optimal energetic transferability to a moiety of interest resulting in an FEP+ mis-prediction.
However,
the chemical features driving these errors varied from case to case.
In an effort to identify and minimize instances like this,
in future applications, we set out to systematically refine the chemical basis that defines torsion parameter types to better delineate groups that necessitate distinct parameter assignments. In section 3 we discuss that effort. Our second example illustrates an FEP+ misprediction due to sub-optimal
assignment
of
partial
charges.
We
observed
poor
predictive performance using OPLS3 over a set of inhibitors with cyclic amidine and guanidine cores binding to the two catalytic aspartates of Bace-1.19,
20
As shown in Figure 2, the predicted
binding free energy for replacing an aromatic carbon for a nitrogen
ACS Paragon Plus Environment
8
Page 9 of 53 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
Journal of Chemical Theory and Computation
(aza derivatives) in the amino-isoindole core is underestimated by 3-4 kcal/mol.
After reviewing the parameters assigned to the
amidine motif we noticed that the partial charges of the aminoaza-isoindole cores differ significantly and consistently from those of the amino-isoindole cores, especially for the amidinium atoms HNC marked in the Figure. We validated this error hypothesis by
fitting
QM charges to
the
ligands
directly
“OPLS3+QMcharges” method outlined in Section 3C.
following the The fitted
partial charges of the amidine motif are in qualitative agreement with the OPLS3 charges for the amino-isoindole core, and consistent for the aza derivatives.
The FEP+ predicted free energy using
OPLS3 with these QM fitted charges leads to good agreement with experiment (Figure 2).20 The reason for this OPLS3 error follows from dependencies of the method on the composition of the training set.
OPLS3 employs
a CM1A-BCC based charge model that combines Cramer-Truhlar CM1A charges21 with specifically fit bond charge correction terms (BCC). The
BCC
corrections
are
primarily
fit
to
QM
electrostatic
potentials (ESP) (using HF/6-31G*) but also include additional refinements against experimental solvation free energies22.
These
BCCs in turn depend on the molecule composition of this training set. The OPLS3 training set includes the bicyclic core of compound 11.
However, the associated aza derivatives are not part of the
OPLS3 training set and the BCCs assigned to the cores in compound
ACS Paragon Plus Environment
9
Journal of Chemical Theory and Computation 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
Page 10 of 53
13 and 16 instead fall back to parameters based on the next best available
match.
In
this
case,
the
next
best
match
is
a
combination of pyridine and acyclic amidine which transfers poorly to the fused ring substructure of interest. The example illustrates a general need for better equipping the charge model when it encounters ligand substructures that lie outside the force field training set.
To address this, we develop
a ligand-level charge model that assigns partial charges for each new compound based on a combination of fitting to QM derived ESP and restraints to OPLS3 and describe a framework that can easily be further updated in the future.
III.
OPLS3e Developments
A. Overview The functional form of the force field and the development of bond stretching, angle bending and van der Waals parameters is discussed
in
the
OPLS3
unchanged in OPLS3e.
reporting
publication18
and
remains
Changes with respect to OPLS3 include
revisions to the small molecule torsional and charge potentials covered in parts B and C, below. B.
Refined Torsion Types
Torsional
parameters
are
needed
to
describe
the
quantum
mechanical effects associated with stereoelectronic effects and resonance structures.
They play a critical role in determining
ACS Paragon Plus Environment
10
Page 11 of 53 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
Journal of Chemical Theory and Computation
molecule
conformational
protein-ligand
binding
energies by
which
modulating
the
sensitively
impacts
conformational
work
needed for a ligand to adopt its bioactive conformation.23
In
OPLS3e, this component of the potential is represented by the same truncated
Fourier
series
described
previously
(see
eq
1
in
reference 18) where the associated V1-4 parameters are determined for each torsion type by fitting to model molecule quantum chemical torsion energy profiles. The profiles were generated as follows: 1. Force field-based torsion scans (in 30 degree increments) are developed from low energy
conformers
and
used
as
the
starting
geometries
in
a
subsequent QM optimization. 2. The quantum chemical profile is resolved via restrained optimization of the molecules at the B3LYP/6-31G* level for each discretized value of the dihedral angle. 3.
Single point M06-2X/cc-pVTZ(-f) calculations are used
to resolve the reference energy surface.
This represents a change
in the QM level used relative to earlier work, which was motivated by
a benchmark comparison
of various
CCSD(T)/CBS, summarized in Table S1.
QM methods relative to
This level is used in the
fitting of all OPLS3e small molecule torsions and is QM reference for all subsequent force field comparisons reported in the paper. The most significant change from OPLS3 is in the definition of the
constituent
torsion
types.
Motivated
by
erroneous
calculations similar to the example discussed in Section 2, we set
ACS Paragon Plus Environment
11
Journal of Chemical Theory and Computation 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
out
to further optimize
the type definitions
Page 12 of 53
to improve the
recognition of chemical environments that should be treated as distinct.
Our approach can be summarized as follows:
1. for each OPLS3 dihedral of interest, identify additional model molecules that share the same torsion types but differ in their chemical composition; 2. develop associated quantum chemical torsion profiles for the set of molecules; 3. test parameter transferability by training to one member of the molecule set and assessing the accuracy relative to QM for the other molecules; 4. identify the distinguishing chemical features that lead to a significant loss in accuracy; 5. revise the torsion type definitions to encapsulate the new chemical features; and 6. refit the model with the new basis of torsion types. An example illustrating this process is summarized in Table 1. Here, the dihedral of interest is the amino rotation of dimethyl aniline.
For OPLS3, the training set molecule is the otherwise
unsubstituted phenyl derivative.
The test set includes para
substituted methyl, fluoro, chloro, amino, hydroxyl, enolate, nitrile and phenyl.
The parameter transferability quality is
indicated by the root mean square error with respect to QM, over the rotamer energy surface, provided for each fragment.
ACS Paragon Plus Environment
Good
12
Page 13 of 53 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
Journal of Chemical Theory and Computation
transferability
is
seen
for
para-methyl,
chloro
and
phenyl
derivatives, where the substituents have little impact on the QM barrier to rotation.
The π-electron fluoro, amino, hydroxyl and
enolate donors significantly weaken the QM barrier to rotation. π-electron
acceptor,
significantly strengthens the QM barrier to rotation.
The trend
The
nitrile
derivative,
which
is
a
is consistent with previous spectroscopic studies.24 The impact these substituents have on the conjugation properties of the torsion in question is not captured by the force field leading to a significant drop in accuracy for these groups. this
limitation,
OPLS3e
encapsulates
this
To address
information
by
delineating its parameter assignment based on the presence of the particular
para
transferability.
substituents
that
significantly
impact
As a consequence, OPLS3e achieves a RMS error
no greater than 0.3 kcal/mol for this data set.
To the 11,845
fragment molecule torsional profiles that constitute the OPLS3 training set18 we added an additional 11,161 fragment profiles to conduct similar parameter transferability tests. In
total,
the
parameter
type
refinements
made
in
OPLS3e
encompass variations along 4 main chemical classes: heteroatom composition,
ortho
substituents,
exocyclic
and
meta/para
substituents.
Figure 3, shows a representative pair of molecules
for each class, with their associated rotamer energies.
Where
in OPLS3, the indicated dihedral parameters are shared for each
ACS Paragon Plus Environment
13
Journal of Chemical Theory and Computation 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
Page 14 of 53
pair, OPLS3e now treats each moiety as distinct. One consequence of refining the definition of torsion types to be more chemotype specific is that it impacts to what extent chemical substructures will be fully covered by pre-defined well parameterized OPLS3e torsion types.
In Harder et al. we estimated that the fraction
of torsional bonds found in drug-like molecules that would match existing trained OPLS3 torsion types was 93%.
At this stage of
the OPLS3e development that estimate dropped to 68%. To recover a comparable level of statistical coverage an additional 19,558 torsional profiles, for model molecules with missing coverage, were resolved and fit using OPLS3e. That is 30,719 more torsional rotamer sets than what we used for developing OPLS3.
The
concomitant refinements resulted in a total of 146,669 torsion types compared to the 48,142 types in OPLS3. we
provide
an
automated
version
of
And, as with OPLS3,
our
fitting
protocol
(FFBuilder) that can be used to fill in outstanding gaps in OPLS3e coverage. Table 2 contrasts the rotamer energy RMS error for OPLS2005, OPLS3
and
OPLS3e.
The
comparison
excludes
molecules
whose
constituent torsion types are not presently covered by OPLS3. The evaluation is split by molecules used to train and test OPLS3e.
Though OPLS3 (RMSE~1 kcal/mol) performed significantly
better than its older counterpart, OPLS2005 (RMSE~2 kcal/mol) the
ACS Paragon Plus Environment
14
Page 15 of 53 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
Journal of Chemical Theory and Computation
new
OPLS3e
parameterization
performs
significantly
better
(RMSE~0.5 kcal/mol).
C. Ligand-Based On-the-Fly Charge Model The OPLS3e charge model retains the same Coulombic fixed charge functional form utilized by OPLS3.
Additional virtual (off-atom
centered) sites for aromatic ring nitrogens and aryl halogens (excluding fluorine) that were added to OPLS3 are retained in OPLS3e.
The difference with respect to OPLS3 is in the method
used to assign partial charges.
As discussed in Section 2, OPLS3
employs a CM1A-BCC scheme that depends on a set of pre-tabulated bond charge corrections predominantly fit to recapitulate the HF/6-31G* electrostatic potential of 11,845 model compounds with additional general,
refinements
this
model
against
has
solvation
demonstrated
an
free
energies.
acceptable
level
In of
accuracy when applied to molecules that lie outside its training set..6-11,
18
However, as evidenced by the Bace-1 outlier shown in
Section 2, application to untrained heterocycles can prove to be problematic. Two approaches that could be applied to ameliorate this issue are: (1) extend the BCC training set to include a growing list of heterocycles25 or (2) parameterize the partial charges for each new ligand by fitting directly to the target QM data (i.e. HF/6-31G* ESP’s) on-the-fly.
Option (1) is akin to the approach taken in
ACS Paragon Plus Environment
15
Journal of Chemical Theory and Computation 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
refining the torsion model.
Page 16 of 53
For charges, however, we expect this
route to present a greater challenge owing to the expansive nature of drug-like chemical space and the particular sensitivity of charges to their chemical environment. method
maximally
encapsulates
all
Because the on-the-fly the
pertinent
chemical
information of the constituent ligands, we’ve chosen to focus on this approach.
One potential downside is that the compute time
required to generate the QM ESP can be prohibitive for certain FFbased applications such as virtual screening or small molecule conformational search.
However, for MD based applications, such
as FEP, the requisite assignment time (~10 min) remains small relative to the time required for the production simulations. Though HF/6-31G* based charges, in general, perform well when combined with the OPLS force field, there are a limited set of known chemical moieties where QM based charge fitting does not translate
well
addressed
by
to
the
phase.26
condensed
additional
refinement
available condensed phase data.22
of
In
BCC
OPLS3,
this
parameters
was
against
To retain these refinements in
OPLS3e we use a restrained QM fitting approach, where the reference values
are
based
on
OPLS3
charges
and
the
restraints can vary by chemical substructure.
strength
of
the
In principal, this
would allow the model to hew more closely to this reference for moieties
that
required
additional
freely to QM, otherwise.
refinements
and
adapt
more
The restraint values are developed and
ACS Paragon Plus Environment
16
Page 17 of 53 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
Journal of Chemical Theory and Computation
validated against a set of small molecule experimental hydration and transfer free energies. The hydration free energy data set is comprised of the Shivakumar et al.22 set of 240 molecules, as well as
an
additional
178
molecules
from
the
FreeSolv
database27
selected to cover a variation of chemical groups while excluding larger compounds with multiple rotatable bonds.
Applying the same
selection criteria, we further expanded our validation sets to include a sub-set of water-toluene and water-chloroform transfer free energies from references 28-30.
FEP+ solvation free energies
were calculated by annihilation of the compound from solvent. Simulations times were 4ns, charge and van der Waals perturbations were run sequentially and transfer free energies were calculated as the difference between annihilation free energies in water and in chloroform/toluene respectively. As a baseline, we explored a charge model with zero restraints to reference values (OPLS3 using QM charges).
Results for these
solvation free energy benchmarks are shown in Table 3. of
the
hydration
free
energy
benchmark
was
The error
observed
to
be
significantly worse using this model versus standard OPLS3 (1.39 kcal/mol versus 1.02 kcal/mol).
Upon closer inspection, the
regression was concentrated in the following functional groups (see Table 4): amines, formates/esters, nitros and the indicated sulfur derivatives.
Representative fragments for these groups are
shown in Figure S1. Amines present a well-known issue where the QM
ACS Paragon Plus Environment
17
Journal of Chemical Theory and Computation 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
Page 18 of 53
ESP derived partial charges fail to describe the trend in solvation energies of primary, secondary and tertiary amines.26 As discussed previously,
OPLS3
has
addressed
this
by
further
refining
parameters against additional condensed phase data. After
iterative
empirical
testing
against
condensed
phase
benchmarks we settled on a two-tier restraint scheme, with a relatively large restraint for the functional groups mentioned above
and
a
relatively
small
restraint
for
everything
else.
Including low restraints helps ameliorate the problems that the partial charges can suffer from the underdetermined nature of fitting to ESP’s for ligand sized molecules31 and ensure physically relevant charges (e.g. non-negative hydrogens).
The hybrid model,
OPLS3e, compares well to OPLS3 for hydration free energies and shows a marked improvement in the transfer free energy benchmarks (Table
3).
In
particular,
the
chemical
groups
identified
as
problematic to describe with an unrestricted fit to the QM ESP are well described in OPLS3e (Table 4).
It is important to emphasize
here that this scheme is predicated on a posteriori knowledge of (and the availability of adequate existing parameters for) the set of moieties where QM electrostatic potential fitting leads to poor transferability to the condensed phase.
This list of moieties
will require revision if, with the acquisition of new experimental data, additional chemistries are identified that show a similar lack of transferability.
ACS Paragon Plus Environment
18
Page 19 of 53 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
Journal of Chemical Theory and Computation
We need to further address the preparation of the molecular conformation used to fit the partial charges. Because fixed charge models do not explicitly account for intramolecular polarization effects, their charge parameters will depend on the particular conformation(s)
used
when
fitting
to
the
QM
ESP.
This
is
especially pertinent for ligand sized molecules in a FEP+ like context.
Here we employ the putative bioactive conformation used
in the FEP calculation. extended charges)22
conformation but
found
As an alternative, we also attempted an (employed
better
by
OPLS3
performance
to using
calculate the
CM1A
bioactive
conformation. As an initial step before fitting charges, the given input
molecule
geometries
are
pre-minimized
with
a
torsional
constraint, to preserve the initially provided conformation while avoiding strain.
In principal, systems where this fixed charge
approach results in parameters that are incongruous with other pertinent
ligand
conformations
(eg.
in
water
solvent)
would
necessitate an electrostatic treatment that explicitly accounts for these polarization effects.32,33
However, we note that in the
FEP+ systems studied in the present work we did not observe significant errors as a consequence of the present approximation.
IV. Application to Protein-Ligand Binding In reference 18, we used a large data set of protein-ligand binding FEP calculations to assess the impact of improvements made
ACS Paragon Plus Environment
19
Journal of Chemical Theory and Computation 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
to the force field.
Page 20 of 53
Comparing OPLS3 to earlier variants of OPLS
we observed a significant systematic reduction in FEP RMS error with improvements to the quality of the force field.
Here we aim
to do the same comparison for OPLS3e and OPLS3. The data set includes the 8 systems originally reported in reference 1.
We also collected additional test series designed to
more stringently probe challenging chemical transformations akin to the outlier cases presented in Section 2.
The class of
perturbations most commonly associated with both the charge and torsion
transferability
issues
discussed
composition of conjugated ring systems.
involved
changes
in
This addition includes
the Chk1, Bace and FXa series first presented in Harder et al. Additional ligand series were sourced for FXa from references 3437,, Bace from reference 38 and another series focused on changes to an amidinium core, coordinating the catalytic aspartates of Bace
in
reference
19,
excluding
compounds
with
uncertain
protonation state or multiple possible tautomeric states. representative
ligand
for
each target is
shown
A
in Figure 4.
Additional data sets, representing a similar chemistry focus, was sourced from Schrodinger collaboration projects, denoted project A, B, C and D in Table 5.
In total 194 additional ligands were
added to the original 199 ligand benchmark. data
set
includes
various
targets
and
The final combined
compounds,
including
examples of kinases, proteases, tyrosine phosphatase, peptidase,
ACS Paragon Plus Environment
20
Page 21 of 53 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
Journal of Chemical Theory and Computation
positive/negative/neutral ligands, and variations of both cores and R-groups. The FEP+ protocol was similar to that previously described in reference 1, with a few notable differences. All simulations were run for 25ns, instead of 5ns, to improve convergence and enable us to
focus
on
force
field
related
incomplete loop was modeled in.
differences.
For
P38,
an
For FXa, after close inspection
of crystal structure densities, we noticed an ion binding site close to the ligand which was previously assigned as water.34,
36, 37
Consistent with the reported assay conditions, we replaced this water with a sodium ion.35,
37
We also refined ligand starting
geometries using a similar strategy to that discussed in reference 39.
When there was ambiguity about the pose, such as for symmetric
R-groups with meta or ortho substituents that could flip 180 degrees, we used FEP+ to assess and select the pose with the lowest free energy.
Starting pose coordinates for ligands and proteins
can be found in supplementary material. Associated PDB codes are provided in Table S2. The overall RMS error is improved by approximately 20% with OPLS3e.
That
improvement
is
primarily
improvement in the extended test series.
driven
by
accuracy
On the original FEP+
benchmark (rows 1-8), both OPLS3 and OPLS3e perform similarly well. The series that sees the greatest improvement using OPLS3e is ProjectC (OPLS3 RMSE = 2.2 kcal/mol, OPLS3e RMSE = 0.9 kcal/mol).
ACS Paragon Plus Environment
21
Journal of Chemical Theory and Computation 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
Page 22 of 53
The improvement is a product of both the improved torsions and charges.
An example perturbation improved by the OPLS3e torsions
is shown in Figure 5.
It involves the breaking of a fused ring
system in the core of the molecule leaving a dangling amide substituent.
Experimentally, breaking the fused ring results in
a 1.5 kcal/mol loss of potency.
In contrast, OPLS3 predicts a
loss of potency of 4.5 kcal/mol.
The simulated distribution of
the torsion connecting the amide substituent to the ring shown in the figure indicates the source of the error.
In both OPLS3 and
OPLS3e, the predominant state in the complex corresponds to a NCcc angle of approximately 180 degrees which situates the amide group in
the same plane as
the parent heterocycle
and
facilitates
formation of inter-molecule hydrogen bonds with the backbone of the protein.
The same conformation is the primary state sampled
by OPLS3e in solution.
However, in OPLS3, the primary state in
solution has the plane of the amide group orthogonal to the core heterocycle (NCcc=90).
Thus, the OPLS3 representation incorrectly
suggests that this variant will incur an additional conformational strain upon binding which leads to the observed under-prediction of its affinity versus the experimental reference data. Consistent with this trend, the QM and OPLS3e energy required to rotate this group to the NCcc=90 state is approx. 14 kcal/mol. OPLS3 under-estimates this energetic penalty by 7 kcal/mol which leads to the erroneous solvent population observed here.
ACS Paragon Plus Environment
Fixing
22
Page 23 of 53 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
Journal of Chemical Theory and Computation
just the torsion issues addresses a significant portion of the OPLS3 error over this series.
Using a model that combines OPLS3
charges with OPLS3e torsions reduces the series RMSE from 2.2 kcal/mol to 1.5 kcal/mol (Table 6).
The remaining improvement in
accuracy is due to the improved partial charge treatment with OPLS3e for the varied heterocyclic systems in this series. The set of amidine inhibitors of Bace-1 (Bace_cores) is another data set that shows significant improvement with OPLS3e.
The data
set includes the outlier case discussed in Section 2 and is comprised of a set of relatively uncommon heterocyclic cores (Figure 6) that form key interactions with catalytic aspartates in the binding site.
To accurately reflect how these variations
modulate the strength of their interaction with the catalytic aspartates, is a challenge for the force field.
The predictions
for this data set are relatively insensitive to the details of the torsion model and are primarily driven by changes in OPLS3e to the partial charges. The data set consists of two maps: one covering a series of monocyclic compounds (cmpd 1-10) and another covering a set of fused bicyclic compounds (cmpd 11-24).
For the monocyclic sub-
set the RMS error improves from 1.7 kcal/mol with OPLS3 to 1.3 kcal/mol.
The basis for the improvement is subtle and manifested
over the full data set, illustrating the benefit of OPLS3e for novel heterocycle design.
For the fused bicyclic subset the RMS
ACS Paragon Plus Environment
23
Journal of Chemical Theory and Computation 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
error improves from 2.2 kcal/mol to 1.5 kcal/mol.
Page 24 of 53
The improvement
is driven by better partial charge modeling of the aza derivatives in the series. unsubstituted
Table 7, gives the binding affinity for each
pyridyl
and
pyrmidyl
derivative
relative to the amino-isoindole core (cmpd 11).
in
the
series
As discussed in
Section 2, the poor OPLS3 predictions are a product of poor parameter transferability from the OPLS3 training set for the naryl substituted compounds. OPLS3e addresses this issue by fitting directly to QM leading to significantly better performance over this set. Despite these improvements the RMS error for the Bace_core data set (1.39 kcal/mol) remains quite high.
The remaining outlier,
which is mainly responsible for the residual error, is cmpd 23. As shown in Figure 7, relative to the amino-isoindole core (cmpd 11) cmpd 23 introduces an additional OMe group to the core ring. As indicated in Figure 7, the addition of the OMe substituent introduces a significant electron rich moiety in close proximity to ASP228 which leads to a significant drop in potency for cmpd 23.
However, both OPLS3 and OPLS3e significantly over-estimate
the associated drop in potency by approx. 3 kcal/mol.
In our FEP+
simulations the two catalytic aspartates in Bace-1 are assumed to be unprotonated in the holo state, in agreement with previous publications.19,
40
However, given the present observation we
conjectured that the introduction of substituents with significant
ACS Paragon Plus Environment
24
Page 25 of 53 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
Journal of Chemical Theory and Computation
electronegativity at this position may shift the preferred state of ASP228 to be neutral.
To test this hypothesis, we estimated
the pKa of this residue in the presence of both compounds #11 and #23 to be 4 and 7, respectively. SI Section 3.
The pKa protocol is discussed in
Given the associated experimental pH of the assay
is 4.5, this implies that the neutral form of ASP288 should be predominant for cmpd 23. Correcting the predicted binding affinity of cmpd #23 relative to cmpd #11, using Equation S6, gives +2.0 kcal/mol which is in-line with the experimentally measured value of +2.6 kcal/mol. We are presently pursuing strategies to include pKa effects, like this, in FEP+ calculations. Likewise, for the monocyclic molecules the binding affinity of core number 5 is still overestimated with OPLS3e. All Bace-1 cores in the data set are assumed to have a protonated amidine motif both in the protein binding site and in the reference state. Experimental core contribution to pKa is lower for core 5 than the other monocycles.19 The same is true for the calculated pKa, which for core 5 with two phenyl R-groups is 4.8,
close to the assay
conditions of pH=4.5. There is a possibility that core 5 is unprotonated before binding, which would result in an energetic penalty upon binding, and could explain the too high predicted affinity. In summary, there is still room for improvements, but overall OPLS3e
performs
well
for
all
tested
systems,
ACS Paragon Plus Environment
and
especially
25
Journal of Chemical Theory and Computation 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
Page 26 of 53
contributes a significant improvement to OPLS3 for challenging heterocycles. V.
Conclusion
We here report the development and validation of a new force field, OPLS3e, that leads to improved performance in predicting protein-ligand binding affinities.
Over an expansive set of
protein targets and ligand chemistries, the model achieves a RMS error of approx. 1 kcal/mol.
The model accomplishes this by
addressing limitations in chemotype transferability that affects small molecule torsions and charges.
Better torsion parameter
transferability is achieved with an expanded basis of torsion types that
distinguish
moieties
with
greater
chemical
specificity.
Better charge parameter transferability is achieved by directly fitting
ligand
specific
electrostatic potential.
charges
to
the
target
QM
derived
Ligand transformations that modulate the
composition of conjugated heterocycles are particularly sensitive to
these
effects
and
display
the
most
pronounced
accuracy
improvements. With regards to future development, we anticipate that additional parameter refinements to the present development (eg. adding new torsion types) will be necessary as more reference data becomes available.
We also expect that, as more aggressive explorations
of chemical and configurational space are enabled by continued advances in sampling efficiency, that will, in turn, expose new
ACS Paragon Plus Environment
26
Page 27 of 53 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
Journal of Chemical Theory and Computation
challenges to the fidelity of the force field.
In addition, we
are also actively looking into accuracy improvements that could be derived from modifications to the functional form of the model (eg. multi-dimensional dihedral terms, better VdW terms etc.) and improved treatment of regimes of chemical space that lie outside the
present
testing
data
(eg.
metalloenzymes).
Beyond
improvements to the force field, we are also actively pursuing strategies to account for the impact that changes in protonation state can have in FEP+ calculations. these
prospective
activities,
we
As a basis for prioritizing continue
to
investigate
outstanding prediction outliers to gain insights into the most important improvements that need to be made. Figures and Tables
Figure 1. relative
Left panel shows the experimental and FEP-predicted binding
free
energy
in
kcal/mol
distinguished by the ester-/amide-pyrrole groups.
ACS Paragon Plus Environment
between
ligands
The right panel 27
Journal of Chemical Theory and Computation 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
Page 28 of 53
shows the associated potential energy of the NCcn dihedral angle for an amide derivative derived from the FEP ligand (top) and the molecule used to train the associated torsion parameters (bottom). The
predominant
angle=180.
state
in
the
complex
corresponds
to
a
NCcn
The predominant state in solution corresponds to a
NCcn angle=0.
OPLS3* is a modified version of OPLS3 that re-
parameterizes the amide-pyrrole dihedral against
the fragment
analog of the amide-pyrrole substructure present in the left ligand.
Figure 2. Bace-1 inhibitor charge outliers. compound 11 in the binding site.
Left panel shows full
Bicyclic core variations with
OPLS3/QM amidine charges overlaid shown at right. Table summarizes the relative experimental and predicted binding free energy of the corresponding full compounds to Bace-1 in kcal/mol. The amidine
ACS Paragon Plus Environment
28
Page 29 of 53 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
Journal of Chemical Theory and Computation
motif forms key hydrogen bonds to the two catalytic aspartates of Bace-1. The predicted binding free energy for replacing one or two aromatic carbons for nitrogens (13, 16) in the amino-isoindole core (11) is consistently underestimated with OPLS3. Ligand series are
from
reference
19,
FEP+
calculations
employed
the
4DJW
structure. Table 1. RMS Error of with respect to QM for dimethyl aniline torsion profiles with varying para substituents (kcal/mol)a Para group
OPLS3 paramet er typea
OPLS3e paramet er typea
OPLS3 RMSE
OPLS3e RMSE
-H
A
A
0.0
0.0
-CH3
A
A
0.3
0.3
-F
A
B
0.6
0.0
-Cl
A
A
0.1
0.1
-NH2
A
C
1.4
0.0
-OH
A
D
1.2
0.1
-O-
A
E
4.1
0.2
-C#N
A
F
1.8
0.1
-Ph
A
A
0.3
0.2
a Letter symbols represent associated dimethyl aniline parameter type. All molecules share the same type in OPLS3 and the parameter is trained against -H. OPLS3e delineates the assignment for para -F, -NH2, -OH, -O- and -C#N functional groups.
ACS Paragon Plus Environment
29
Journal of Chemical Theory and Computation 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
Page 30 of 53
Figure 3. Rotamer energy comparisons for molecule pairs along the indicated dihedral.
Each pair (A-D) is a representative example
of the 4 chemical classes impacted by the OPLS3e torsion parameter type refinements.
In OPLS3, each pair shared the same parameter
assignment which now differ in OPLS3e. Table 2. Torsion profile RMS Error with respect to M06-2X/ccpVTZ(-f) (kcal/mol)a Model
all molecul
trainin g molecul
test molecule s: well
ACS Paragon Plus Environment
30
Page 31 of 53 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
Journal of Chemical Theory and Computation
es (25890)
es (20646)
covered (5244)
OPLS_200 5
1.9
1.9
1.8
OPLS3
1.2
1.2
1.3
OPLS3e
0.4
0.3
0.8
aNumber
of molecules for each set is shown in parentheses.
Table 3. Error summary for hydration free energy results, and transfer free energies to chloroform and toluene. Data set
No. cmpd s
RMSE (kcal/mol) OPLS3
OPLS3 + OPLS3e QM charges
418
1.02
1.39
0.98
Water-chloroform free energies
transfer 242
1.12
1.10
0.99
Water-toluene free energies
transfer 163
0.93
0.74
0.71
Hydration free energies
Table 4. Error summary for hydration free energies of chemical groups assigned high restraints in the new ligand based charge model. RMSE (kcal/mol) Chemical Group
No. cmpds
OPLS3
OPLS3 + OPLS3 QM e charges
Amines
55
1.39
2.17
1.30
Formates/este
49
0.89
2.04
1.16
ACS Paragon Plus Environment
31
Journal of Chemical Theory and Computation 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
Page 32 of 53
rs Nitros
29
1.05
1.36
0.80
Sulfoxides/su lfones/sulfat es/aryl thiols
5
0.66
2.45
0.76
ACS Paragon Plus Environment
32
Page 33 of 53 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
Journal of Chemical Theory and Computation
Figure 4. Examples from series of compounds from the extended data set focused on variations of heterocycles. Table 5. Relative binding free energy results.
ACS Paragon Plus Environment
33
Journal of Chemical Theory and Computation 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
No.
No.
Page 34 of 53
RMSE (kcal/mol)
System
cmpd s
Seri es ref
BACE
36
41
58
1.11
1.10
CDK2
16
42
25
1.07
1.00
JNK1
21
43
34
0.81
0.88
MCL1
42
44
71
1.25
1.17
P38
34
45
56
0.89
0.88
PTP1B
23
46
49
0.70
0.63
Thrombin
11
47
16
0.85
1.08
Tyk2
16
48, 49
24
0.73
0.65
FXA
40
1417
60
1.47
1.29
CHK1
19
5055
18
1.20
0.98
BACE core 23
19
30
1.97
1.39
BACE heterocyc 21 les
38, 41
28
1.47
1.04
ProjectA
13
a
11
0.91
0.80
ProjectB
11
a
18
2.11
1.31
ProjectC
22
a
28
2.22
0.90
ProjectD
45
a
63
1.23
1.01
Total Weighted Average
393
589
1.29
1.04
Pert OPLS3 .
OPLS3 e
a In-house data.
ACS Paragon Plus Environment
34
Page 35 of 53 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
Journal of Chemical Theory and Computation
Figure 5. Left panel shows the experimental and FEP-predicted relative
binding
free
energy
in
kcal/mol
between
ligands
distinguished by the breaking of the fused lactam ring from the projectC series.
The right panel shows the simulated population
of the NCcc dihedral angle of the acyclic derivative. Table 6. ProjectC Relative Binding Free Energy Error Model
RMSE (kcal/m ol)
OPLS3
2.2
OPLS3 + 1.5 New Torsions OPLS3e
0.9
ACS Paragon Plus Environment
35
Journal of Chemical Theory and Computation 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
Page 36 of 53
Figure 6. Illustration of the chemical diversity of the challenging monocyclic and bicyclic core variations of Bace-1 inhibitors from reference 19. The description of the heterocycle variations and their effect on the key interactions were limited with OPLS3 and improved with OPLS3e. Table 7. Experimental and FEP Predicted Relative Binding Free Energies for Bace_core ligandsa Ligand perturbat iona
ddG (exp.)
ddG(OPLS3 )
ddG(OPLS3 e)
ACS Paragon Plus Environment
36
Page 37 of 53 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
Journal of Chemical Theory and Computation
11 12
0.7
5.3
2.3
11 13
0.0
4.0
1.0
11 14
0.2
4.0
0.5
11 15
0.5
4.5
1.6
11 16
0.9
3.3
0.9
a Specified ligands are illustrated in Figure 6. in kcal/mol.
Free energies
Figure 7. The experimental and FEP-predicted relative binding free energy in kcal/mol between ligands distinguished by the indicated substituent change (methoxy for hydrogen).
Substitution position
is in close proximity to ASP228. Tabulated FEP predictions are based on the receptor state indicated here where both ASP32 and ASP228 are modeled in their deprotonated form. ASSOCIATED CONTENT
ACS Paragon Plus Environment
37
Journal of Chemical Theory and Computation 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
Page 38 of 53
FEP results, input structures and OPLS3e parameters are included with the supplementary information. AUTHOR INFORMATION Corresponding Author *
[email protected] Notes K.R. and C.W. contributed equally to this work.
The authors
declare the following competing financial interest(s): R.A.F. has a significant financial stake in, is a consultant for, and is on the Scientific Advisory Board of Schrodinger, Inc. K.R. was funded by the Swedish Research Council. REFERENCES (1)
Wang, L.; Wu, Y.; Deng, Y.; Kim, B.; Pierce, L.; Krilov,
G.; Lupyan, D.; Robinson, S.; Dahlgren, M. K.; Greenwood, J.; Romero, D. L.; Masse, C.; Knight, J. L.; Steinbrecher, T.; Beuming, T.; Damm, W.; Harder, E.; Sherman, W.; Brewer, M.; Wester, R.; Murcko, M.; Frye, L.; Farid, R.; Lin, T.; Mobley, D. L.; Jorgensen, W. L.; Berne, B. J.; Friesner, R. A.; Abel, R.
Accurate
and
Reliable
Prediction
of
Relative
Ligand
Binding Potency in Prospective Drug Discovery by Way of a Modern Free-Energy Calculation Protocol and Force Field. J. Am. Chem. Soc. 2015, 137, 2695–2703.
ACS Paragon Plus Environment
38
Page 39 of 53 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
Journal of Chemical Theory and Computation
(2)
Zwanzig, R. W. High‐Temperature Equation of State by a
Perturbation Method. I. Nonpolar Gases. J. Chem. Phys. 1954, 22, 1420–1426. (3)
Kollman, P. Free Energy Calculations: Applications to
Chemical and Biochemical Phenomena. Chem. Rev. 1993, 93, 2395–2417. (4)
Perez, A.; Morrone, J. A.; Simmerling, C.; Dill, K. A.
Advances in Free-Energy-Based Simulations of Protein Folding and Ligand Binding. Curr. Opin. Struct. Biol. 2016, 36, 25– 31. (5)
Hansen, N.; van Gunsteren, W. F. Practical Aspects of
Free-Energy Calculations: A Review. J. Chem. Theory Comput. 2014, 10, 2632–2647. (6)
Abel,
R.;
Wang,
L.;
Harder,
E.
D.;
Berne,
B.
J.;
Friesner, R. A. Advancing Drug Discovery through Enhanced Free Energy Calculations. Acc. Chem. Res. 2017, 50, 1625– 1632. (7)
Abel,
R.;
Mondal,
S.;
Masse,
C.;
Greenwood,
J.;
Harriman, G.; Ashwell, M. A.; Bhat, S.; Wester, R.; Frye, L.; Kapeller, R.; Friesner, R. A. Accelerating Drug Discovery through Tight Integration of Expert Molecular Design and
ACS Paragon Plus Environment
39
Journal of Chemical Theory and Computation 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
Page 40 of 53
Predictive Scoring. Curr. Opin. Struct. Biol. 2017, 43, 38– 44. (8)
Ciordia, M.; Pérez-Benito, L.; Delgado, F.; Trabanco, A.
A.; Tresadern, G. Application of Free Energy Perturbation for the Design of BACE1 Inhibitors. J. Chem. Inf. Model. 2016, 56, 1856–1871. (9)
Lovering, F.; Aevazelis, C.; Chang, J.; Dehnhardt, C.;
Fitz, L.; Han, S.; Janz, K.; Lee, J.; Kaila, N.; McDonald, J.; Moore, W.; Moretto, A.; Papaioannou, N.; Richard, D.; Ryan, M. S.; Wan, Z-K.; Thorarensen, A. Imidazotriazines: Spleen Tyrosine Kinase (Syk) Inhibitors Identified by FreeEnergy Perturbation (FEP). ChemMedChem 2015, 11, 217–233. (10)
Jorgensen, W. L. Computer-Aided Discovery of Anti-HIV
Agents. Bioorg. Med. Chem. 2016, 24, 4768–4778. (11)
Lenselink,
E.
B.;
Louvel,
J.;
Forti,
A.
F.;
van
Veldhoven, J. P. D.; de Vries, H.; Mulder-Krieger, T.; McRobb, F. M.; Negri, A.; Goose, J.; Abel, R.; van Vlijmen, H. W. T.; Wang, L.; Harder, E.; Sherman, W.; IJzerman, A. P.; Beuming, T. Predicting Binding Affinities for GPCR Ligands Using FreeEnergy Perturbation. ACS Omega 2016, 1, 293–304. (12)
Hünenberger
classical
P.
interaction
H.;
van
Gunsteren
functions
for
ACS Paragon Plus Environment
W.
F.
molecular
Empirical simulation
40
Page 41 of 53 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
Journal of Chemical Theory and Computation
BT
-
Computer
Theoretical
and
Simulation Experimental
of
Biomolecular
Applications.
In
Systems: Computer
Simulation of Biomolecular Systems; van Gunsteren W. F., Weiner P. K., Wilkinson A. J., Eds.; Springer: Dordrecht, Netherlands, 1997; 3, pp 3-82. (13)
Darden, T. A. Treatment of Long-Range Forces and Potential.
In Computational Biochemistry and Biophysics; Becker, O., MacKerell Jr. A.D., Roux, B., Watanabe, M., Eds.; CRC Press: Boca Raton, 2001. (14)
Halgren, T. A. Merck Molecular Force Field. I. Basis,
Form, Scope, Parameterization, and Performance of MMFF94. J. Comput. Chem. 2018, 17, 490–519. (15)
Wang, J.; Wolf, R. M.; Caldwell, J. W.; Kollman, P. A.;
Case, D. A. Development and Testing of a General Amber Force Field. J. Comput. Chem. 2004, 25, 1157–1174. (16)
Vanommeslaeghe, K.; Hatcher, E.; Acharya, C.; Kundu, S.;
Zhong, S.; Shim, J.; Darian, E.; Guvench, O.; Lopes, P.; Vorobyov, I.; Mackerell Jr. A. D. CHARMM General Force Field (CGenFF): A Force Field for Drug-like Molecules Compatible with the CHARMM All-Atom Additive Biological Force Fields. J. Comput. Chem. 2010, 31, 671–690. (17)
Mobley, D.; Bannan, C. C.; Rizzi, A.; Bayly, C. I.;
Chodera, J. D.; Lim, V. T.; Lim, N. M.; Beauchamp, K. A.;
ACS Paragon Plus Environment
41
Journal of Chemical Theory and Computation 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
Page 42 of 53
Slochower, D. R.; Shirts, M. R.; Gilson, M. K.; Eastman, P. K. Escaping Atom Types Using Direct Chemical Perception. J. Chem. Theory Comput. 2018, 14, 6076–6092. (18)
Harder, E.; Damm, W.; Maple, J.; Wu, C.; Reboul, M.;
Xiang, J. Y.; Wang, L.; Lupyan, D.; Dahlgren, M. K.; Knight, J. L.; Kaus, J. W.; Cerutti, D.; Krilov, G.; Jorgensen, W. L.; Abel, R.; Friesner, R. A. OPLS3: A Force Field Providing Broad Coverage of Drug-like Small Molecules and Proteins. J. Chem. Theory Comput. 2016, 12, 281–296. (19)
Roos, K.; Viklund, J.; Meuller, J.; Kaspersson, K.;
Svensson,
M.
Potency
Prediction
of ß-Secretase
(BACE-1)
Inhibitors Using Density Functional Methods. J. Chem. Inf. Model. 2014, 54, 818–825. (20)
Roos, K.: Friesner, R. A.
Comparing Physics Based
Methods for Protein-Ligand Binding. Manuscript. (21)
Storer, J. W.; Giesen, D. J.; Cramer, C. J.; Truhlar, D.
G. Class IV Charge Models: A New Semiempirical Approach in Quantum Chemistry. J. Comput. Aided. Mol. Des. 1995, 9, 87– 110. (22)
Shivakumar, D.; Harder, E.; Damm, W.; Friesner, R. A.;
Sherman, W. Improving the Prediction of Absolute Solvation
ACS Paragon Plus Environment
42
Page 43 of 53 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
Journal of Chemical Theory and Computation
Free Energies Using the Next Generation OPLS Force Field. J. Chem. Theory Comput. 2012, 8, 2553–2558. (23)
Tirado-Rives,
J.;
Jorgensen,
W.
L.
Contribution
of
Conformer Focusing to the Uncertainty in Predicting Free Energies for Protein−Ligand Binding. J. Med. Chem. 2006, 49, 5880–5884. (24)
Fateley, W. G.; Carlson, G. L.; Bentley, F. F. Phenolic-
OH Torsional Frequency as a Probe for Studying -Electron Distortions in Aromatic Systems. J. Phys. Chem. 1975, 79, 199–204. (25)
Pitt, W. R.; Parry, D. M.; Perry, B. G.; Groom, C. R.
Heteroaromatic Rings of the Future. J. Med. Chem. 2009, 52, 2952–2963. (26)
Rizzo, R. C.; Jorgensen, W. L. OPLS All-Atom Model for
Amines:
Resolution of the Amine Hydration Problem. J. Am.
Chem. Soc. 1999, 121, 4827–4836. (27)
Mobley, D. L. Experimental and Calculated Small Molecule
Hydration
Free
Pharmaceutical
Energies,
2013.
UC
Irvine:
Sciences,
Department
of
UCI.
http://www.escholarship.org/uc/item/6sd403pz (28)
Stephens, T. W.; Loera, M.; Quay, A. N.; Chou, V.; Shen,
C.; Wilson, A.; Acree Jr, W. E.; Acree, W. E.; Abraham, M. H.
ACS Paragon Plus Environment
43
Journal of Chemical Theory and Computation 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
Page 44 of 53
Correlation of Solute Transfer Into Toluene and Ethylbenzene from Water and from the Gas Phase Based on the Abraham Model. Open Thermodyn. J 2011, 5, 104–121. (29)
Sprunger, L. M.; Achi, S. S.; Acree, W. E.; Abraham, M.
H.; Leo, A. J.; Hoekman, D. Correlation and Prediction of Solute Transfer to Chloroalkanes from Both Water and the Gas Phase. Fluid Phase Equilib. 2009, 281, 144–162. (30)
Abraham, M. H.; Platts, J. A.; Hersey, A.; Leo, A. J.;
Taft, R. W. Correlation and Estimation of Gas − Chloroform and Water − Chloroform Partition Coefficients by a Linear Free Energy Relationship Method. J. Pharm. Sci. 1999, 88, 670–679. (31)
Jakalian, A.; Bush, B. L.; Jack, D. B.; Bayly, C. I.
Fast, Efficient Generation of High-Quality Atomic Charges. AM1-BCC Model: I. Method. J. Comput. Chem. 2000, 21, 132–146. (32)
Shi, Y.; Xia, Z.; Zhang, J.; Best, R.; Wu, C.; Ponder,
J. W.; Ren, P. Polarizable Atomic Multipole-Based AMOEBA Force Field for Proteins. J. Chem. Theory Comput. 2013, 9, 4046–4063. (33)
Lopes, P. E. M.; Huang, J.; Shim, J.; Luo, Y.; Li, H.;
Roux, B.; MacKerell, Jr. A. D. Force Fields for Peptides and
ACS Paragon Plus Environment
44
Page 45 of 53 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
Journal of Chemical Theory and Computation
Proteins based on the Classical Drude Oscillator. J. Chem. Theory Comput. 2013, 9, 5430–5449. (34)
Yoshikawa, K.; Kobayashi, S.; Nakamoto, Y.; Haginoya,
N.; Komoriya, S.; Yoshino, T.; Nagata, T.; Mochizuki, A.; Watanabe, K.; Suzuki, M.; . Design, Synthesis, and SAR of Cis-1,2-Diaminocyclohexane Derivatives as Potent Factor Xa Inhibitors.
Part
II:
Exploration
of
6–6
Fused
Rings
as
Alternative S1 Moieties. Bioorg. Med. Chem. 2009, 17, 8221– 8233. (35)
Yoshikawa, K.; Yokomizo, A.; Naito, H.; Haginoya, N.;
Kobayashi, Osanai,
S.;
K.;
Yoshino,
Watanabe,
T.;
K.;
Nagata,
Kanno,
T.;
H.;
Mochizuki,
Ohta,
T.
A.;
Design,
Synthesis, and SAR of Cis-1,2-Diaminocyclohexane Derivatives as Potent Factor Xa Inhibitors. Part I: Exploration of 5-6 Fused Rings as Alternative S1 Moieties. Bioorganic Med. Chem. 2009, 17, 8206–8220. (36)
Zbinden, K. G.; Anselm, L.; Banner, D. W.; Benz, J.;
Blasco, F.; Décoret, G.; Himber, J.; Kuhn, B.; Panday, N.; Ricklin, F.; Risch, P.; Schlatter, D.; Stahl, M.; Thomi, S.; Unger, R.; Haap, W. Design of Novel Aminopyrrolidine Factor Xa Inhibitors from a Screening Hit. Eur. J. Med. Chem. 2009, 44, 2787–2795.
ACS Paragon Plus Environment
45
Journal of Chemical Theory and Computation 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
(37)
Page 46 of 53
Chan, C.; Borthwick, A. D.; Brown, D.; Burns-Kurtis, C.
L.; Campbell, M.; Chaudry, L.; Chung, C-w.; Convery, M. A.; Hamblin, J. N.; Johnstone, L.; Kelly, H. A.; Kleanthous, S.; Patikis, A.; Patel, C.; Pateman, A. J.; Senger, S.; Shah, G. P.; Toomey, J. R.; Watson, N. S.; Weston, H. E.; Whitworth, C.; Young, R. J.;
Zhou, P. J. Factor Xa Inhibitors: S1
Binding Interactions of a Series of N-{(3S)-1-[(1S)-1-Methyl2-Morpholin-4-Yl-2-Oxoethyl]-2-Oxopyrrolidin-3Yl}Sulfonamides. J. Med. Chem. 2007, 50, 1546–1557. (38)
Truong, A. P.; Probst, G. D.; Aquino, J.; Fang, L.;
Brogley, L.; Sealy, J. M.; Hom, R. K.; Tucker, J. A.; John, V.; Tung, J. S.; Pleiss, M. A.; Konradi, A. W.; Sham, H. L.; Dappen, M. S.; Toth, G.; Yao, N.; Brecht, E.; Pan, H.; Artis, D. R.; Lany, R.; Bova, M. P.; Sinha, S.; Yednock, T. A.; Zmolek,
W.;
Permeability
Quinn,
K.
of
Hydroxyethylamine
the
P.;
Sauer,
J-M.
Improving
BACE-1
the
Inhibitors:
Structure–Activity Relationship of P2′ Substituents. Bioorg. Med. Chem. Lett. 2010, 20, 4789–4794. (39)
Kaus, J. W.; Harder, E.; Lin, T.; Abel, R.; McCammon, J.
A.; Wang, L. How To Deal with Multiple Binding Poses in Alchemical
Relative
Protein–Ligand
Binding
Free
Energy
Calculations. J. Chem. Theory Comput. 2015, 11, 2670–2679.
ACS Paragon Plus Environment
46
Page 47 of 53 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
Journal of Chemical Theory and Computation
(40)
Swahn, B.; Holenz, J.; Kihlström, J.; Kolmodin, K.;
Lindström, J.; Plobeck, N.; Rotticci, D.; Sehgelmeble, F.; Sundström, M.; Berg, S. Von; Falting, J.; Georgievska, B.; Gustavsson, S.; Neelissen, J.; Ek, M.; Olsson, L-L.; Berg, S. Aminoimidazoles
as
BACE-1
Inhibitors :
The
Challenge
to
Achieve in Vivo Brain Efficacy. Bioorg. Med. Chem. Lett. 2012, 22, 1854–1859. (41)
Cumming, J. N.; Smith, E. M.; Wang, L.; Misiaszek, J.;
Durkin, J.; Pan, J.; Iserloh, U.; Wu, Y.; Zhu, Z.; Strickland, C.; Voigt, J.; Chen, X.; Kennedy, M. E.; Kuvelkar, R.; Hyde, L. A.; Cox, K.; Favreau, L.; Czarniecki, M. F.; Greenlee, W. J.;
McKittrick,
B.
A.;
Parker,
E.
M.;
Stamford,
A.
W.
Structure Based Design of Iminohydantoin BACE1 Inhibitors: Identification of an Orally Available, Centrally Active BACE1 Inhibitor. Bioorg. Med. Chem. Lett. 2012, 22, 2444–2449. (42)
Hardcastle, I. R.; Arris, C. E.; Bentley, J.; Boyle, F.
T.; Chen, Y.; Curtin, N. J.; Endicott, J. A.; Gibson, A. E.; Golding, B. T.; Griffin, R. J.; Jewsbury, P.; Menyerol, J.; Mesguiche, V.; Newell, D. R.; Noble, M. E.; Pratt, D. J.; Wang,
L.
Z.;
Whitfield,
H.
J.
Cyclohexylmethylguanine Derivatives:
N2-Substituted
O6-
Potent Inhibitors of
Cyclin-Dependent Kinases 1 and 2. J. Med. Chem. 2004, 47, 3710–3722.
ACS Paragon Plus Environment
47
Journal of Chemical Theory and Computation 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
(43)
Page 48 of 53
Szczepankiewicz, B. G.; Kosogof, C.; Nelson, L. T. J.;
Liu, G.; Liu, B.; Zhao, H.; Serby, M. D.; Xin, Z.; Liu, M.; Gum, R. J.; Haasch, D. L.; Wang, S.; Clampit, J. E.; Johnson, E. F.; Lubben, T. H.; Stashko, M. A.; Olejniczak, E. T.; Sun, C.; Dorwin, S. A.; Haskins, K.; Abad-Zapatero, C.; Fry, E. H.;
Hutchins,
C.
W.;
Sham,
H.
L.;
Rondinone,
C.
M.;
Trevillyan, J. M. Aminopyridine-Based c-Jun N-Terminal Kinase Inhibitors with Cellular Activity and Minimal Cross-Kinase Activity. J. Med. Chem. 2006, 49, 3563–3580. (44)
Friberg, A.; Vigil, D.; Zhao, B.; Daniels, R. N.; Burke,
J. P.; Garcia-Barrantes, P. M.; Camper, D.; Chauder, B. A.; Lee, T.; Olejniczak, E. T.; Fesik, S. W. Discovery of Potent Myeloid Cell Leukemia 1 (Mcl-1) Inhibitors Using FragmentBased Methods and Structure-Based Design. J. Med. Chem. 2013, 56, 15–30. (45)
Goldstein, D. M.; Soth, M.; Gabriel, T.; Dewdney, N.;
Kuglstatter, A.; Arzeno, H.; Chen, J.; Bingenheimer, W.; Dalrymple, S. A.; Dunn, J.; Farrell, R.; Frauchiger, S.; La Fargue, J.; Ghate, M.; Graves, B.; Hill, R. J.; Li, F.; Litman, R.; Loe, B.; McIntosh, J.; McWeeney, D.; Papp, E.; Park, J.; Reese, H. F.; Roberts, R. T.; Rotstein, D.; San Pablo, B.; Sarma, K.; Stahl, M.; Sung, M.-L.; Suttman, R. T.; Sjogren, E. B.; Tan, Y.; Trejo, A.; Welch, M.; Weller, P.;
ACS Paragon Plus Environment
48
Page 49 of 53 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
Journal of Chemical Theory and Computation
Wong, B. R.; Zecic, H. Discovery of 6-(2,4-Difluorophenoxy)2-[3-Hydroxy-1-(2-Hydroxyethyl)Propylamino]-8-Methyl-8HPyrido[2,3-d]Pyrimidin-7-One
(Pamapimod)
and
6-(2,4-
Difluorophenoxy)-8-Methyl-2-(Tetrahydro-2H-Pyran-4Ylamino)Pyrido[2,3-d]Pyrimidin-7(8H)-One (R1487) as Orally Bioavailible and Highly Selective Inhibitors of p38 MitogenActivated Protein Kinase. J. Med. Chem. 2011, 54, 2255–2265. (46)
Wilson, D. P.; Wan, Z.-K.; Xu, W.-X.; Kirincich, S. J.;
Follows, B. C.; Joseph-Mccarthy, D.; Foreman, K.; Moretto, A.; Wu, J.; Zhu, M.; Binnun, E.; Zhang, Y.-L.; Tam, M.; Erbe, D. V.; Tobin, J.; Xu, X.; Leung, L.; Shilling, A.; Tam, S. Y.; Mansour, T. S.; Lee, J. Structure-Based Optimization of Protein Tyrosine Phosphatase 1B Inhibitors:
From the Active
Site to the Second Phosphotyrosine Binding Site. J. Med. Chem. 2007, 50, 4681–4698. (47)
Baum, B.; Mohamed, M.; Zayed, M.; Gerlach, C.; Heine,
A.; Hangauer, D.; Klebe, G. More than a Simple Lipophilic Contact:
A
Detailed
Thermodynamic
Analysis
of
Nonbasic
Residues in the S1 Pocket of Thrombin. J. Mol. Biol. 2009, 390, 56–69. (48)
Liang, J.; Tsui, V.; Van Abbema, A.; Bao, L.; Barrett,
K.; Beresini, M.; Berezhkovskiy, L.; Blair, W. S.; Chang, C.; Driscoll,
J.;
Eigenbrot,
C.;
Ghilardi,
ACS Paragon Plus Environment
N.;
Gibbons,
P.; 49
Journal of Chemical Theory and Computation 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
Page 50 of 53
Halladay, J.; Johnson, A.; Kohli, P. B.; Lai, Y.; Liimatta, M.; Mantik, P.; Menghrajani, K.; Murray, J.; Sambrone, A.; Xiao, Y.; Shia, S.; Shin, Y.; Smith, J.; Sohn, S.; Stanley, M.; Ultsch, M.; Zhang, B.; Wu, L. C.; Magnuson, S. Lead Identification of Novel and Selective TYK2 Inhibitors. Eur. J. Med. Chem. 2013, 67, 175–187. (49)
Liang, J.; van Abbema, A.; Balazs, M.; Barrett, K.;
Berezhkovsky, L.; Blair, W.; Chang, C.; Delarosa, D.; DeVoss, J.; Driscoll, J.; Eigenbrot, C.; Ghilardi, N.; Gibbons, P.; Halladay, J.; Johnson, A.; Kohli, P. B.; Lai, Y.; Liu, Y.; Lyssikatos, J.; Mantik, P.; Menghrajani, K.; Murray, J.; Peng, I.; Sambrone, A.; Shia, S.; Shin, Y.; Smith, J.; Sohn, S.; Tsui, V.; Ultsch, M.; Wu, L. C.; Xiao, Y.; Yang, W.; Young,
J.;
Optimization
Zhang, of
a
B.;
Zhu,
B.-y.;
4-Aminopyridine
Magnuson,
Benzamide
S.
Lead
Scaffold
To
Identify Potent, Selective, and Orally Bioavailable TYK2 Inhibitors. J. Med. Chem. 2013, 56, 4521–4536. (50)
Lin, N.-H.; Xia, P.; Kovar, P.; Park, C.; Chen, Z.;
Zhang,
H.;
Rosenberg, S. H.;
Sham,
H.
L. Synthesis and
Biological Evaluation of 3-Ethylidene-1,3-Dihydro-Indol-2Ones as Novel Checkpoint 1 Inhibitors. Bioorg. Med. Chem. Lett. 2006, 16, 421–426.
ACS Paragon Plus Environment
50
Page 51 of 53 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
Journal of Chemical Theory and Computation
(51)
Huang, S.; Garbaccio, R. M.; Fraley, M. E.; Steen, J.;
Kreatsoulas, C.; Hartman, G.; Stirdivant, S.; Drakas, B.; Rickert, K.; Walsh, E.; Hamilton, K.; Buser, C. A.; Hardwick, J. Mao, X.; Abrams, M.; Beck, S.; Tao, W.; Lobell, R.; Laura Sepp-Lorenzino,
L.;
Yan,
Y.;
Ikuta,
M.;
Murphy,
J.
Z.;
Sardana, V.; Munshi, S.; Kuo, L.; Reillyd, M.; Mahand, E. Development of 6-Substituted Indolylquinolinones as Potent Chek1 Kinase Inhibitors. Bioorg. Med. Chem. Lett. 2006, 16, 5907–5912. (52)
Oza, V.; Ashwell, S.; Almeida, L.; Brassil, P.; Breed,
J.; Deng, C.; Gero, T.; Grondine, M.; Horn, C.; Ioannidis, S.; Liu, D.; Lyne, P.; Newcombe, N.; Pass, M.; Read, J.; Ready, S.; Rowsell, S.; Su, M.; Toader, D.; Vasbinder, M.; Yu, D.; Yu, Y.; Xue, Y.; Zabludoff, S.; Janetka, J. Discovery of
Checkpoint
Kinase
Inhibitor
(S)-5-(3-Fluorophenyl)-N-
(Piperidin-3-Yl)-3-Ureidothiophene-2-Carboxamide by
Structure-Based
Design
and
(AZD7762)
Optimization
of
Thiophenecarboxamide Ureas. J. Med. Chem. 2012, 55, 5130– 5142. (53)
Zhao, L.; Zhang, Y.; Dai, C.; Guzi, T.; Wiswell, D.;
Seghezzi, W.; Parry, D.; Fischmann, T.; Siddiqui, M. A. Design, Synthesis and SAR of Thienopyridines as Potent CHK1 Inhibitors. Bioorg. Med. Chem. Lett. 2010, 20, 7216–7221.
ACS Paragon Plus Environment
51
Journal of Chemical Theory and Computation 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
(54)
Page 52 of 53
Tao, Z-F.; Wang, L.; Stewart, K. D.; Chen, Z.; Gu, W.;
Bui, M-H.; Merta, P.; Zhang, H.; Kovar, P.; Johnson, E.; Park, C.; Judge, R.; Rosenberg, S.; Sowin, T.; Lin, N-H. StructureBased Design, Synthesis, and Biological Evaluation of Potent and Selective Macrocyclic Checkpoint Kinase 1 Inhibitors. J. Med. Chem. 2007, 50, 1514–1527. (55)
Huang, X.; Cheng, C. C.; Fischmann, T. O.; Duca, J. S.;
Yang, X.; Richards, M.; Shipps, G. W. Discovery of a Novel Series of CHK1 Kinase Inhibitors with a Distinctive Hinge Binding Mode. ACS Med. Chem. Lett. 2012, 3, 123–128.
ACS Paragon Plus Environment
52
Page 53 of 53 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
Journal of Chemical Theory and Computation
Table of Contents Graphic OPLS3e:
Extending
Force
Field
Coverage
for
Drug-Like
Small
Molecules Katarina Roos, Chuanjie Wu, Wolfgang Damm, Mark Reboul, James M. Stevenson, Chao Lu, Markus K. Dahlgren, Sayan Mondal, Wei Chen, Lingle Wang, Robert Abel, Richard A. Friesner, Edward Harder
ACS Paragon Plus Environment
53