Identification of small molecule inhibitors of botulinum neurotoxin serotype E via footprint similarity

Article abstract of DOI:10.1016/j.bmc.2016.07.031

Botulinum neurotoxins (BoNT) are among the most poisonous substances known, and of the 7 serotypes (A–G) identified thus far at least 4 can cause death in humans. The goal of this work was identification of inhibitors that specifically target the light ch

Full text of DOI:10.1016/j.bmc.2016.07.031

Accepted Manuscript
Identification of Small Molecule Inhibitors of Botulinum Neurotoxin Serotype
E via Footprint Similarity
Yuchen Zhou, Brian E. McGillick, Yu-Han Gary Teng, Krupanandan
Haranahalli, Iwao Ojima, Subramanyam Swaminathan, Robert C. Rizzo
PII:
S0968-0896(16)30538-7
DOI:
http://dx.doi.org/10.1016/j.bmc.2016.07.031
BMC 13144
Reference:
Bioorganic & Medicinal Chemistry
To appear in:
Received Date:
Revised Date:
Accepted Date:
8 April 2016
15 July 2016
16 July 2016
Please cite this article as: Zhou, Y., McGillick, B.E., Teng, Y.G., Haranahalli, K., Ojima, I., Swaminathan, S., Rizzo,
R.C., Identification of Small Molecule Inhibitors of Botulinum Neurotoxin Serotype E via Footprint Similarity,
Bioorganic & Medicinal Chemistry (2016), doi: http://dx.doi.org/10.1016/j.bmc.2016.07.031
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Identification of Small Molecule Inhibitors of Botulinum
Neurotoxin Serotype E via Footprint Similarity
Yuchen Zhou,§, a Brian E. McGillick,§, b,e Yu-Han Gary Teng,c, d Krupanandan Haranahalli,d
Iwao Ojima,c, d Subramanyam Swaminathan,e and Robert C. Rizzo*, a, c, f
a Department of Applied Mathematics & Statistics, Stony Brook University, Stony Brook, NY 11794
b Graduate Program in Biochemistry & Structural Biology, Stony Brook University, Stony Brook, NY
11794
c Institute of Chemical Biology & Drug Discovery, Stony Brook University, Stony Brook, NY 11794
d Department of Chemistry, Stony Brook University, Stony Brook, NY 11794
e Biology Department, Brookhaven National Laboratory, Upton, New York 11973
f Laufer Center for Physical & Quantitative Biology, Stony Brook University, Stony Brook, NY 11794
Received XX, 2016
§ These authors contributed equally to this work
* Corresponding author e-mail: rizzorc@gmail.com
Keywords: Botulinum Neurotoxin, BoNT/E, Docking, Virtual screening, DOCK, Footprint similarity,
Scoring functions, Molecular dynamics
Abbreviations: BoNT/E, Botulinum neurotoxin serotype E; FPS, footprint similarity; DCE, DOCK
Cartesian energy; VDW, van der Waals; ES, electrostatic; MD, molecular dynamics;
1

ABSTRACT: Botulinum neurotoxins (BoNT) are among the most poisonous substances known, and
of the 7 serotypes (A-G) identified thus far at least 4 can cause death in humans. The goal of this work
was identification of inhibitors that specifically target the light chain catalytic site of the highly
pathogenic but lesser-studied E serotype (BoNT/E). Large-scale computational screening, employing
the program DOCK, was used to perform atomic-level docking of 1.4 million small molecules to
prioritize those making favorable interactions with the BoNT/E site. In particular, "footprint similarity"
(FPS) scoring was used to identify compounds that could potentially mimic features on the known
substrate tetrapeptide RIME. Among 92 compounds purchased and experimentally tested, compound
C562-1101 emerged as the most promising hit with an apparent IC50 value three-fold more potent than
that of the first reported BoNT/E small molecule inhibitor NSC-77053. Additional analysis showed the
predicted binding pose of C562-1101 was geometrically and energetically stable over an ensemble of
structures generated by molecular dynamic simulations and that many of the intended interactions seen
with RIME were maintained. Several analogs were also computationally designed and predicted to
have further molecular mimicry thereby demonstrating the potential utility of footprint-based scoring
protocols to help guide hit refinement.
2

1. Introduction
Botulinum neurotoxins (BoNT), secreted by the anaerobic bacterium Clostridium botulinum, are
among the most lethal biological substances known to humans and estimated to have a median lethal
dose (LD50) of 1 ng/kg of body weight.1 There are seven BoNT serotypes (labeled A-G) and all target
members of the Soluble N-ethylmaleimide-sensitive factor Attachment Protein Receptor (SNARE)
protein family for cleavage which results in the paralysis of muscle tissue.2,3 In particular, serotypes A,
B, E, and F all cause human botulism with A, B, and E being the most common.1 Given their potential
for use as bioweapons,3 the Centers for Disease Control and Prevention (http://www.cdc.gov) have
designated BoNTs as a Category A biological agent, underscoring the importance of studies aimed at
developing effective therapeutics.
All of the serotypes share high amino acid sequence homology and activity mechanism.4 As
shown in Figure 1 the toxin is organized as two chains comprised of: (i) a large ~100 kDa "heavy chain"
(Figure 1a) consisting of a binding domain (BD) and translocation domain (TD), and (ii) a smaller ~50
kDa "light chain" (Figure 1b) catalytic domain (CD) that includes a zinc dependent active site for
substrate cleavage dependent on serotype.4,5 In terms of mechanism, under normal physiological
conditions, members of the SNARE family that include synaptobrevin, syntaxin, and synaptosome-
associated protein of 25 kDa (SNAP-25) help assemble the synaptic fusion complex thereby allowing
synaptic vesicles to fuse with neuron cell membranes resulting in ordered release of neurotransmitters.
In the presence of the toxin, BoNT heavy chains target gangliosides and a specific protein receptor at
the presynaptic membrane which results in endocytosis of the toxin. A significant conformational
change in the toxin4 enables the release of BoNT light chains into the surrounding neuronal cytosol
resulting in serotype specific cleavage of SNARE proteins. As a consequence the synaptic fusion
complex cannot be formed and neurotransmitter release is blocked. Serotypes A, C, and E cleave
different sites on SNAP-25 while serotypes B, D, F, and G cleave various sites on synaptobrevin.
Serotype C also has activity against syntaxin.3,6
3

Figure 1. (a) Ribbon structure of BoNT/E showing the heavy chain (orange) consisting of the binding domain
(BD) and the translocation domain (TD), and the light chain (gray) comprised of the catalytic domain (CD),
coordinates from PDB code 3FFZ.7 (b) Close-up view of a BoNT/E light chain (gray) co-crystallized with a
natural substrate derivative tetrapetide RIME (green), coordinates from PDB code 3D3X.8 (c) Structures of NSC-
77053 (top) and C562-1101 (bottom).
While a number of prior studies focused on vaccine development,9-12 considerable efforts to
develop BoNT inhibitors have also been undertaken. These previous studies focused primarily on
serotypes A and B and resulted in the discovery of several classes of inhibitors.2,13 One such class are
polypeptides designed to bind to the BoNT/A light chain active domain. Highlighting the importance of
a P1’ arginine position for tight binding, Schmidt et al14 designed the peptide inhibitor CRATKML with
a Ki of 2 μM, and subsequently improved the Ki to 330 nM by replacing the cysteine with 2-mercapto-3-
phenylpropionyl.15 Further refinement led to the design of an extremely potent peptide inhibitor DNP-
DAB-RWT-DAB-ML with a Ki of 41 nM. Kumaran et al16 as well as Kumar et al17 developed a series
of tetrapeptide inhibitors displaying low micromolar activity,13 the most potent comprised of the
sequence RRGF, with an IC50 of 0.9 μM. Other efforts included the identification of two natural
product inhibitors that target the BoNT/A exosite, D-Chicoric acid by Silhar et al18 with a Ki of 0.7 μM,
and Lomofungin identified by Eubanks et al19 with a Ki of 6.7 μM.
Past efforts to develop small molecules targeting the light chain active site have focused
primarily on serotype A.2,13 Examples include the discovery of para-chloro-cinnamic hydroxamate with
an IC50 of 15 μM and a subsequently improved analog ortho-para-chloro-cinnamic hydroxamate with an
IC50 of 0.41 μM.20 Silvaggi et al21 determined the crystal structure of ortho-para-chloro-cinnamic
4

hydroxamate revealing the hydroxamic acid chelates the active site zinc. Building on this finding,
Thompson et al22 designed and tested a series of hydroxamic acid derivatives, among which the most
potent compound showed an IC50 of 4.6 μM. Quinolinol based small molecules are another important
class of compounds showing low micromolar inhibitory activity, with recent examples identified by
Burnett et al23 showing IC50 values as low as 10 μM. Additionally, Videnovic et al24 reported a second
generation 4-amino-7-chloroquinoline inhibitor with a Ki value of 103 nM and Caglic et al25 recently
tested a series of quinolinol inhibitors with the most active compound having an IC50 of 0.8 μM.
Despite its importance, efforts to develop potent BoNT/E inhibitors have been minimal.
Exceptions include work by Agarwal et al8 who reported a co-crystal structure of the BoNT/E light
chain catalytic domain in complex with the tetrapeptide RIME (see Figure 1b) and Kumar et al26 who
subsequently used this complex to identify candidate small molecule inhibitors through computer-based
virtual screening. Among the computational candidates, compound NSC-77053 (Figure 1c top) showed
an experimental Ki of 1.29 μM using an HPLC-based assay and represents the first reported small
molecule inhibitor identified that targets the BoNT/E catalytic site.26 Recent follow up work by the
same group,27 employing pharmacophore modeling, led to the identification of four roughly equipotent
analogs. Importantly, although the RIME peptide is a derivative of the natural substrate SNAP-25
(residues 180 through 183 of SNAP-25 spanning the scissile bond) it is not cleavable by BoNT/E.28
Thus, the BoNT/E-RIME crystallographic complex8 represents a native-like substrate-bound state and is
a reasonable starting point for continued structure-based inhibitor design.
The goal of the present work is to expand upon previous BoNT/E studies, utilizing a much larger
virtual screen against the BoNT/E catalytic site in order to identify new small molecule inhibitor
candidates. The primary objectives are fourfold: (1) perform flexible ligand docking of ~1.4 million
commercially available small molecules from the ZINC29 database, (2) employ a recently updated
version of the screening program DOCK30,31 containing enhanced scoring functions32,33 that allow for
identification of compounds that make similar interaction patterns as the bound RIME substrate, (3)
purchase and experimentally test a subset of the top-scoring compounds, and (4) perform detailed
5

structural and computational analysis on compounds showing experimental activity. As described
below, this workflow has led to the identification of a novel BoNT/E small molecule inhibitor C562-
1101 / ZINC20284316 (Figure 1c bottom) which is about three-fold more potent than inhibitor NSC-
77053 (Figure 1c top) when tested in the same GFP-based assay.
2. Methods
2.1 Receptor selection. The crystallographic coordinates selected for virtual screening were based on
the structure reported by Agarwal et al8 consisting of a BoNT/E light chain at a resolution 2.25 Å (PDB:
3D3X) co-crystalized with a substrate SNAP-25 derivative, tetrapeptide RIME. Importantly, use of the
substrate-bound structure for virtual screening helps to ensure that the target sites are "preformed" to
accommodate a bound ligand. And in the case of RIME, the substrate also provides a baseline reference
to help guide identification of compounds that may make similar interactions with the active site zinc
ion and surrounding residues as outlined below. For screening, chain B of the structure was selected
over chain A given that the tetrapeptide ligand RIME in chain A was not completely resolved, there
were 4 missing residues (235-238) in chain A that were in close proximity to the binding site, and chain
B had only two missing residues (59, 60) and these were relatively distant from the binding site (> 15
Å). It should be noted that the experimental BoNT/E structure8 shows the N-terminus of RIME
coordinated directly to the active site zinc and therefore unlikely to be protonated. Consequently, the
coordinating RIME substrate was modeled as having an uncharged (free NH2) terminal amine group.
2.2 Docking setup. Docking setups, procedures, and protocols for DOCK generally followed those
previously reported.30,31,34 Briefly, water molecules and sulfate ions were deleted from the initial x-ray
coordinates and the antechamber and tleap programs in the AMBER 14 package35 were used to assign
ff99SB36 (protein) and GAFF37 plus AM1-BCC38,39 (RIME reference) force field parameters. Hydrogen
atoms in the complex were then minimized using the sander module in AMBER in conjunction with
heavy restraints (1000.0 kcal mol-1 Å-2) on all non-hydrogen atoms. Protein and ligand coordinates
6

were then extracted and separately saved in MOL2 format for use with DOCK. A final energy
minimization allowing the RIME ligand to relax in the context of the DOCK infrastructure was then
performed using a tethered restraint (in this case 5.0 kcal mol-1 Å-2). As expected, the DOCK
preparation protocol yielded little deviation in coordinates relative to the initial x-ray structure for
residues near (defined as ≤ 6 Å from RIME) the targeted binding site (0.008 Å rmsd) or for the RIME
reference itself (0.39 Å rmsd).
2.3 Binding site preparation. Following the receptor preparation and energy minimizations steps, the
sphgen40 module in the program DOCK631 was used to generate spheres to help guide anchor placement
for docking, and only spheres within 8.0 Å to the reference ligand were retained. An energy grid
consisting of van der Waals (VDW) and electrostatic (ES) terms was generated using the program grid41
which pre-computes the receptor contributions which speeds up the docking. Grid calculations
employed a box surrounding spheres within an 8.0 Å margin in each dimension from the RIME
reference and a grid point separation of 0.4 Å. Following standard practice,34 VDW terms were
computed using a 6-9 Lennard-Jones potential which softens the intermolecular energy landscape and
ES terms were computed using Coulombs law with a distance dependent dielectric (ddd) = 4r which
improves DOCK outcomes.
2.4 Virtual screening protocol. A subset of the ZINC library,29 consisting of ~ 1.4 M commercially
available drug-like small organic molecules, was then screened against the BoNT/E light chain active
site using the program DOCK6,31 with the best grid-based energy pose being retained for each
successfully docked compound. The grid-based poses were then energy minimized using the standard
DOCK Cartesian energy (DCE) function consisting of Coulombic (ddd =4r) plus Lennard-Jones (in this
case 6-12) terms in Cartesian space to reach a more refined local minimum energy binding pose and
permit footprint-based terms to be computed as described below.
7

It should be emphasized that all ZINC libraries were docked as provided in terms of their
protonation states or isomeric form. Molecules with different protonation states may exist as separate
entries in ZINC libraries but these typically will have the same ZINC code. As standard practice, all are
docked individually, however only the best scored molecule with a given ZINC code is retained in later
post-processing steps. In terms of optical isomers, these typically have unique ZINC codes and thus
from the standpoint of the virtual screening protocol are treated as unique molecules. However, if a
compound is identified as active through experimental testing, and the physical vendor sample contains
multiple isomeric states, all of the isomers should be more thoroughly studied as exemplified by the
discussion presented in sections 2.10 and 3.2 regarding R versus S enantiomers.
2.5 Footprint similarity (FPS) scoring. A recently-developed footprint similarity (FPS) scoring
function32,33 was subsequently used to rescore the docked compounds. FPS score quantifies the
similarity of interaction patterns made by candidate compounds with that of a reference compound
known to bind to the target, using a per-residue decomposition of the intermolecular VDW and ES
energies (based on the DCE function). The goal in this case was to identify compounds that make
similar footprint patterns as the tetrapeptide RIME. Prior work in our laboratory has employed FPS
scoring alone, and in combination with other functions, to successfully identify active compounds
targeting HIVgp41,42-44 FABP,45,46 and BoNT/A.47 Figure 2 graphically illustrates the overall DOCK
setup comprised of (a) the BoNT/E light chain (gray) with active site region bounded by the docking
grid (magenta), docking spheres (orange), and RIME reference (green), as well as (b) the VDW (top)
and ES (bottom) footprints made by bound RIME used to guide compound selection.
8

Figure 2. DOCK setup showing (a) the BoNT/E light chain site being targeted through virtual screening. Protein
surface in gray, tetrapeptide RIME in green stick model, docking grid in magenta, and docking spheres in orange.
Panel (b) shows the per-residue footprints (top 30 residues) made by bound RIME decomposed into VDW (top)
and ES (bottom) contributions.
2.6 Compound selection. To improve diversity in the types of compounds chosen for experimental
testing, the top-scoring compounds from standard docking were clustered using a pairwise clustering
algorithm based on MACCS48 fingerprint similarity score, as implemented in the program MOE,49 with
a Tanimoto similarity cutoff of 0.75. To help prioritize molecules for purchase, five different rank-
ordered lists were then prepared by re-ranking the resulting clusterheads: (1) DCEVDW+ES, DOCK
Cartesian energy score consisting of van der Waals plus electrostatic terms, (2) FPSVDW+ES, footprint
similarity score consisting of van der Waals plus electrostatic terms, (3) FPSVDW, footprint similarity
score consisting of only the van der Waals term, (4) FPSES, footprint similarity score consisting of only
the electrostatic term, and (5) Total Score, a linear combination of DCEVDW+ES and FPSVDW+ES
.
Following 3D visual inspection of compounds in the binding site ~ 29 representative members from
each list were purchased (vendor: ChemDiv, Inc.) for experimental testing.
2.7 Experimental SNAP-Etide assay. Testing of compounds identified through the virtual screen was
performed using a commercially available SNAP-Etide substrate (List Biological Inc., CA) as employed
9

in prior work,50 following the manufacture recommendations.51 Assays were carried out in a 100 μL
reaction mixture containing an optimized buffer solution containing 20 mM Hepes pH 7.4, 2 mM DTT,
2.5 nM of BoNT/E light chain, and 100 μM of each test compound. A few compounds were tested at
lower concentrations due to reduced solubility although none of these showed activity and were not
pursued further. Compounds were dissolved in 100% DMSO and added to the reaction mixture to a
final concentration of ≤ 1% DMSO and incubated with the BoNT/E light chain for 15 minutes. The
reaction was initialized by the addition of 10 μM SNAP-Etide substrate and conducted at 37°C in 96-
well, black, flat-bottomed, non-binding plates (Corning® Costar®) and monitored using a Tecan
Saphire2 microplate reader. The excitation and emission wavelengths were set at 320 nm and 420 nm,
respectively. The compounds were tested in sets of 10 and were run simultaneously with two DMSO
controls. Initial velocities (Vi) of the control runs were averaged and compared to the initial velocity of
each test compound with percent inhibition of each compound being calculated as [1 – (TEST(Vi)
÷
DMSO(Vi))] x 100%.
2.8 Experimental GFP based assay. Compounds showing activity in the SNAP-Etide assay were re-
tested using a secondary assay based on the work of Hines et al52 that contains a Green Fluorescent
Protein (GFP) reporter and employs a larger substrate than SNAP-Etide. Briefly, the assay utilizes a
hybrid VAMP2/SNAP-25 substrate with the SNAP-25 portion consisting of residues Ala128-Gly206.
The VAMP2 and SNAP-25 domains are connected to one another by a short linker sequence and the
entire construct contains an N-terminal AviTag and C-terminal GFP. After biotinylation of the AviTag,
the substrate is fixed to Streptavidin coated plates. Subsequent cleavage of substrate by addition of
buffer containing BoNT/E light chain liberates GFP into bulk solution allowing for the visualization of a
normal enzymatic progress curve. Reactions were performed in 96-well, black, streptavidin coated
plates (PierceTM) in a buffer solution containing 20 mM Hepes pH 7.4, 2 mM DTT, 25 nM BoNT/E
light chain, and compound. Optimization of the reaction required an approximate 10-fold increase in
the BoNT/E LC concentration compared to the SNAP-Etide assay which is likely due to differences
10

between the two substrates, fluorophores, and other conditions. As before, potential inhibitors were
allowed to incubate with BoNT/E light chain for 15 minutes and the solutions were subsequently added
to the plate wells to initiate the reactions. All reactions were conducted at 37°C and were monitored
using a Tecan Saphire2 microplate reader at excitation and emission wavelengths of 460 nm and 506 nm,
respectively.
2.9 Molecular dynamics simulations and free energy calculations. Molecular dynamics (MD)
simulations were also performed for the most promising hits identified in both assays, as well as several
analogs, to examine the overall stability of the predicted binding geometries over an ensemble of
structures. Analogous to the DOCK setup protocols described above, MD-ready models were similarly
constructed using the AMBER 14 package35 with the protein being assigned ff99SB36 parameters and
docked ligands being assigned GAFF37 parameters augmented by AM1-BCC38,39 partial charges.
Systems were solvated using the TIP3P53 explicit water model and the AMBER pmemd module was
used to perform energy minimizations and MD in the NPT ensemble at a temperature of 298.15 K to
mimic the experimental conditions.
Prior to data collection (20 ns in the present work), a nine-step minimization and dynamics
equilibration protocol was employed that enforced gradually decreasing restraints on both the receptor
and ligand to relax the complex in a controlled manner. Similar to past MD protocols employed by our
laboratory,43,54,55 the data collection phase employed no restraints on the ligand and only a relatively
weak restraint (0.1 kcal mol-1 Å-2) on protein backbone heavy atoms. To evaluate the geometric
stability of predicted binding poses, post-MD analysis included ligand root mean squared deviation
(RMSD), distance measurements, and ensemble-based footprint similarity calculations (i.e. FPS scores)
which were all calculated using the AMBER cpptraj56 utility.
Free energies of binding (Gbind) were also estimated using the implicit solvent MM-GBSA57,58
"single-trajectory" approach, from periodically saved protein-ligand snapshots extracted from the
original explicit-solvent MD trajectories, following protocols outlined in previous work.54,55,59 It should
11

be emphasized that since the trajectories were generated with explicit solvent, the effects of structural or
ordered waters should be reflected in the resultant ensemble of protein binding site and ligand
conformations used for the free energy calculations, although, no detailed water analysis was performed
in this work.
2.10 Chemical synthesis of C562-1101. Although the R enantiomer of C562-1101 (ZINC20284316)
was the isomer originally identified from the DOCK virtual screen, it was unclear if the sample
provided by the vendor for experimental testing was a racemic mixture, or the R or S enantiomer. Thus,
in order to clarify the stereochemical integrity of the sample from the vendor and to examine the activity
of R and S enantiomers, as well as to establish a feasible synthetic route to new analogs for optimization,
(R)- and (S)-C562-1101 were synthesized starting from commercially available (R)- and (S)-Boc-Leu-
OH, and 2-benzoxazolinone based on the retrosynthetic analysis shown in Scheme 1.
Scheme 1. Retrosynthetic analysis of (R)- and (S)-C562-1101
Synthesis of (R)-C562-1101 is illustrated in Scheme 2. 3-Methylbenzoxazolinon-6-ylsulfonyl
chloride (3), was synthesized from 2-benzoxazolinone (1), in two steps in high yield. First, the
methylation of 1 using dimethyl sulfate gave 3-methylbenzoxazolinone (2), which was then reacted with
chlorosulfonic acid to afford 3 in excellent yield. The coupling of (R)-Boc-Leu-OH (4) with 2,3-
12

dimethylaniline via mixed anhydride gave (R)-Boc-Leu-anilide 5 in virtually quantitative yield. (R)-H-
Leu-anilide 6, obtained by deprotection of the Boc group with trifluoroacetic acid (TFA), was reacted
with 3 in the presence of trimethylamine (TEA) to give the desired (R)-C562-1101 (SB-BTE-001) in
good yield. The corresponding S enantiomer and (R)-Ile analog of C562-1101 (SB-BTE-002) were also
synthesized using the same procedure in similar overall yields. See Supplementary Information for
details.
Scheme 2. Synthesis of (R)-C562-1101 (SB-BEI-001)
2.11 Chiral HPLC analysis of C562-1101 obtained from the vendor. Chiral HPLC analysis of the
purity and stereochemistry of the sample obtained from the vendor was carried out using a chiral HPLC
column and R and S enantiomers of C562-1101, synthesized above, as standards. The analysis indicated
that the sample of C562-1101 from the vendor was a ~90:10 mixture of S and R enantiomers (~80%
e.e.). Thus, the original C562-1101 vendor sample was the S enantiomer arising from (S)-Leu-OH. See
the Supplementary Information for details.
13

3. Results and Discussion
3.1 Virtual screening. Figure 3 outlines the overall virtual screening procedure consisting of four
primary steps (gray boxes) used to obtain smaller, more focused, sets of results ultimately resulting in
92 compounds being purchased. From the initial ZINC library of ~1.4M compounds docked to BoNT/E,
the 100,000 top-scoring molecules (DCEVDW+ES score) were then processed and clustered into groups
having similar chemically identity according to the MACCS48 fingerprinting method. Post processing
also included computation of per-residue footprints for each compound and a variety of drug-like
descriptors including Lipinski-like terms, number of rotatable bonds, and formal charge, among others,
to assist with compound prioritization. Using the five different scoring criteria described above in
methods (DCEVDW+ES, FPSVDW+ES, FPSVDW, FPSES, Total Score) the resultant clusterheads were then re-
ranked and the top 2,000 from each list were retained to aid compound selection (see Figure 4). The
goal is twofold: (i) maximize the diversity of compounds prioritized for purchase by using clusterheads
of a given family of related compounds, and (ii) employ different ranking functions to eliminate bias
from any one particular scoring method. In general, each of the 5 lists should contain a diverse set of
compounds that will be unique from those in different ranked lists.
Figure 3. Virtual screening protocol for targeting the BoNT/E light chain active domain.
14

3.1.1 Ligand properties. Figure 4 emphasizes the unique characteristics of each of the five ranked sets
of ligands in terms of nine different descriptors: (a) DCEVDW+ES energy, (b) DCEVDW energy, (c) DCEES
energy, (d) FPSVDW+ES distance, (e) FPSVDW distance, (f) FPSES distance, (g) Total Score, (h) molecular
weight, and (i) number of rotatable bonds. As expected, use of a specific scoring function will show
bias for specific chemical or physical properties. For example, clusterheads selected by the DCEVDW+ES
function always have the most favorable DCEVDW+ES scores (Figure 4a black line) as well as strongly
favorable DCEVDW (Figure 4b black line) and DCEES (Figure 4c black line) terms which are the
components of the DCEVDW+ES score. Similarly, compounds selected using FPSVDW+ES (Figures 4d
green), FPSVDW (Figures 4e blue), or FPSES (Figures 4f red) functions yield more compounds with more
favorable scores in each of their respective histograms. Compounds selected with Total Score, intended
to provide a balance between DCEVDW+ES (which favors compounds with the most favorable receptor
interactions) and FPSVDW+ES (which favors compounds with the most similar interactions as a reference
molecule), analogously yields the "largest peak" in its respective histogram (Figure 4g orange). The
clusterheads selected using Total Score yield population distributions roughly in between that obtained
with DCEVDW+ES (Figure 4a orange versus black or green) and FPSVDW+ES (Figure 4d orange versus
black or green) indicative of docked poses with both favorable energy scores and reasonable molecular
mimicry in terms of the RIME reference.
15

Figure 4. Histograms for top-ranked compounds (N=2000 clusterheads) prioritized by 1 of 5 different DOCK
scoring functions (DCEVDW+ES = black, FPSVDW+ES = green, FPSVDW = blue, FPSES = red, Total Score = orange)
plotted as 9 different descriptors: (a) DCEVDW+ES energy, (b) DCEVDW energy, (c) DCEES energy, (d) FPSVDW+ES
distance, (e) FPSVDW distance, (f) FPSES distance, (g) Total Score, (h) molecular weight, and (i) number of
rotatable bonds.
As expected, use of the standard DOCK function alone (i.e. DCEVDW+ES) biases towards larger
compounds compared to that of FPS functions alone as illustrated by higher molecular weights (Figure
4h black) and greater number of rotatable bonds (Figure 4i black). Somewhat surprisingly, use of Total
Score, containing both DCE and FPS terms, yields nearly the same trends for both descriptors (Figure
4h,i black vs orange) despite the fact the compound sets have quite different energetic properties (Figure
4h,i black vs orange). In sharp contrast, use of the three FPS functions alone shows much broader
distributions of molecular weight and down shifted numbers of rotatable bonds indicative of smaller
molecules.
16

3.1.2 Compounds in common. Table 1 shows the number of molecules in common between different
rank-ordered lists for a smaller subset consisting of the top-scoring 200 compounds (a smaller more
practical number that one might reasonably consider for purchase). For these experiments, it is apparent
that use of footprint similarity scoring provides greater diversity in terms of which compound would be
chosen, as opposed to if only the DCEVDW+ES function was used. Specifically, there are no compounds
in common between the top 200 DCEVDW+ES compounds and any of lists derived from the three footprint
similarity scores. Not surprisingly, there is overlap between the DCEVDW+ES and Total Score lists
(N=72 compounds). Among the three different footprint similarity derived lists, there is also diversity
although not nearly as extreme. Use of FPSVDW+ES yields between ~20-25% in common with either the
van der Waals or electrostatic interaction footprint similarity scores alone. Interestingly, use of
FPSVDW+ES yields only a small number of compounds in common with Total Score (N=5), in sharp
contrast to the value of 72 noted above for the other-way-round, despite the fact that Total Score is a
combination of DCEVDW+ES plus FPSVDW+ES. Given these observations, with the present protocols, Total
Score appears to be dominated by the standard DCEVDW+ES score although this could easily be adjusted
in future experiments by increasing the "weight" coefficient used when combining the two values.
Table 1
Number of overlapping compounds between top 200 compounds from each rank ordered list.
DCEVDW+ES
200
FPSVDW+ES
FPSVDW
FPSES
0
Total Score
DCEVDW+ES
FPSVDW+ES
FPSVDW
0
0
72
5
200
48
38
200
2
0
FPSES
200
6
Total Score
200
Figure 5 shows the ensemble results visually in terms of binding geometries. As expected, the
ensemble prioritized using DCEVDW+ES (Figure 5a), while still within the general area of the binding site,
shows compounds significantly spread out over a larger surface. In contrast, compounds identified by
17

FPSVDW+ES (Figure 5b) are more tightly clustered in the BoNT/E binding site, as they are more likely to
make contacts with amino acid side chains in a similar manner as the RIME reference used in
prioritization. Notwithstanding the discussion above regarding the importance of individual score
components, the use of Total Score to derive an ensemble of 200 compounds (Figure 5c) qualitatively
leads to molecules occupying space in the binding site roughly "in-between" that of the other two
functions (Figures 5a or 5b), as was the original intent.
Figure 5. Overlay of the top 200 scoring compounds in the BoNT/E site prioritized using (a) DCEVDW+ES, (b)
FPSVDW+ES, or (c) Total Score functions.
3.1.3 Compound selection. BoNT/E is a zinc dependent protease thus good coordination to the active
site zinc ion was emphasized during compound prioritization in an attempt to identify compounds that
"mimic" known substrates. This was primarily done through visual inspection of DOCK predicted
binding poses for candidate compounds as well as requiring a strong electrostatic interaction peak with
zinc ion in footprint similarity (FPS) scoring plots. Observed zinc coordinating groups included
carbonyl oxygens, nitrogens, sulfurs, sulfones, and carboxylic acids. Additional criteria for eliminating
compounds included number of stereo centers, formal charges, molecular weight, and number of
rotatable bonds. In total, 122 compounds were prioritized for experimental testing from a roughly even
breakdown in terms of which of the five scoring functions was used for prioritization: DCEVDW+ES = 29,
FPSVDW+ES = 26, FPSES= 27, FPSVDW= 26, and Total Score = 30. As was illustrated in Table 1, in some
18

cases compounds are present in more than one group selected by different prioritizing criteria.
Ultimately, 92 unique commercially available compounds were purchased for experimental testing.
3.2 Experimental testing. During optimization of the experimental SNAP-Etide assay, run-to-run
deviations of initial velocity data between control runs was up to ±10%, therefore a cutoff of >20%
inhibition at 100 µM was used as the criteria for a “positive hit”. We would also emphasize that the
goal of the initial experimental tests was to rapidly eliminate compounds showing no inhibition, even at
100 µM, and our use of a 20% cutoff helps to ensure we will not miss any potential hits. However, at
this relatively high concentration there is an increased risk of false positives due to non-specific
interactions. Therefore, as described below, the initial hits were tested in a secondary assay at a range
of concentrations.
Of the 92 compounds purchased from the vendor based on the computational screen, 9
compounds showed 25-71% inhibition in the initial assay when tested at a concentration of 100 µM as
shown in Table 2 which displays compound identifiers, in silico descriptors, % activity, and 2D
structures. Among the group, two repeating features include a sulfonyl moiety which was present in 3
of the compounds as well as substituted phenylacetamide-like groups in 7 of the compounds.
Interestingly, the compound with the largest % activity (C562-1101) has the second best footprint
overlap (FPSVDW+ES = 14.29) although this is likely to be fortuitous as FPS scores (like DOCK scores)
are not expected to, in a quantitative sense, have strong correlation with experimental activities across a
group of molecules. A noteworthy observation is that the relatively poorer DOCK score for C562-1101
(-65.89 kcal/mol), versus other docked compounds (see Figure 4a), indicates this hit would likely not
have been selected for experimental testing if only the standard DCEVDW+ES function were employed in
the selection process.
19

Table 2
Compound identifiers, in silico descriptors, % activity, and 2D structures for 9 hits out of 92 experimentally
tested compounds using the SNAP-Etide assay.
DCEVDW+ES FPSVDW+ES FPSVDW FPSES
Tot Score Activity
Code
Structure
(kcal/mol)
(d)
(d)
(d)
(kcal/mol)
(%)
C562-1101
-65.89
14.29
5.96
8.33
-51.60
71
8010-8498
C609-0236
-80.20
-78.77
41.40
20.73
9.61
8.60
31.80
12.13
-38.80
-58.04
52
47
C548-1327
C548-0436
F985-1368
-65.21
-67.80
-74.87
16.22
16.42
15.93
8.83
9.40
6.88
7.39
7.02
9.06
-48.99
-51.39
-58.94
42
41
40
G741-1024
8014-7190
-66.56
-76.75
14.11
17.49
5.62
7.75
8.48
9.74
-52.45
-59.27
33
31
M779-0919
-75.28
15.33
7.28
8.05
-59.95
25
20

Visualization of the docked poses for the actives (Figure 6 orange ligands) show that all make
interactions with the catalytic ion via an O --- Zn motif with 7 of the 9 showing very similar
coordination geometries as is made by the C-terminal backbone carbonyl of the native substrate RIME.
Ligand O --- Zn distances (Figure 6 orange ligands) range from 1.8 to 2.5 Å in comparison to RIME
(Figure 6 green ligand) at 2.4 Å.
Figure 6. Three dimensional pose overlays for 9 docked compounds (orange) showing similar interaction
distances (1.8 to 2.5 Å) between oxygen (red) and zinc (magenta) as the native tetrapeptide RIME ligand (green,
2.4 Å).
These 9 hits were also examined using a secondary GFP-based assay (see Methods), in triplicate,
to help alleviate false positives common to fluorescence assays. By using a fluorescent molecule with a
different excitation and emission wavelength than that employed in SNAP-Etide, activity demonstrated
by a candidate inhibitor in both assays is unlikely to be due solely to interference with the fluorophore.
The second assay also employed a larger SNAP-25 construct which is potentially more indicative of the
biologically-relevant substrate. Somewhat surprisingly, only 1 of the 9 compounds (Table 2, compound
C562-1101) active in the first SNAP-Etide assay displayed activity in the second GFP assay. As
expected, the positive control compound NSC-77053 also showed activity. Interestingly, C562-1101
(percent inhibition 71%) was the most potent of all 9 compounds tested in both assays.
As described in the Methods section (vide supra), the most promising hit (C562-1101), was re-
synthesized in-house in its enantiomerically pure (R)- and (S)-forms so that each enantiomer could be
21

tested in the second GFP-based assay. The chiral HPLC analysis of the initial C562-1101 sample
provided by the vendor, using the pure R and S enantiomers as standards, indicated that it was an S
enantiomer with ca. 80% enantiomeric purity (Supplementary Information). In contrast, the form
originally identified from the computational screen was for the R enantiomer. Unexpectedly, the ZINC
database contained unique entries for (R)-C562-1101 (ZINC20284316) and (S)-C562-1101
(ZINC20284319) although both entries code to the same ChemDiv ID of C562-1101. Nevertheless, in
terms of experimental activity, it appears that there is no significant difference in activity between the
pure R and S enantiomers when tested at three different concentrations (Table 3). Interestingly however,
the predicted binding geometries for R and S were very different.
Table 3
Percent inhibition for enantiomers of C562-1101 in the
GFP-based assay.
Concentration
100 M
R enantiomer
84.0% ± 4.0%
58.0% ± 7.0%
46.5% ± 5.5%
S enantiomer
77.5% ± 3.5%
62.5% ± 7.5%
52.0% ± 2.0%
25 M
12.5 M
To investigate why docking the two enantiomers yielded different predicted binding geometries,
the opposite isomer was manually constructed in each case starting from each respective R and S pose.
Manual construction of S, based on the predicted geometry for R, resulted in unfavorable clashes in the
binding site that could not be ameliorated through simple rotamer exploration of the isobutyl group. At
this time, DOCK does not directly support receptor flexibility, thus, with the current binding site
conformation derived from the x-ray coordinates, only the R enantiomer can adopt the predicted
geometry. Conversely, although manual construction of R based on the DOCK-predicted geometry for
S did not result in an unfavorable clash with the protein, the DCE score at only -38.8 kcal/mol was
significantly less favorable than either the docked S (-60.3 kcal/mol) or docked R (-65.9 kcal/mol)
forms. In terms of the experimental results, one possible explanation is that induce fit effects not
currently captured in our molecular modeling approach allow both R and S enantiomers to adopt the
22

same basic pose in the BotNT/E site which leads to similar activity. Ideally, co-crystallization could be
used to help resolve this question.
From multiple structural and energetic standpoints, the docked R enantiomer was more favorable
than S in terms of: (1) energy score (-65.9 vs -60.3 kcal/mol), (2) footprint similarity with RIME (14.3
vs 27.3 Euclidian distance), (3) volume overlap with RIME (52 vs 30 % overlap), and (4) number of
matched pharmacophore features60 with RIME (5 vs 0). In addition, subsequent molecular dynamics
(MD) simulations of the R enantiomer showed good geometric and energetic stability (see discussion
below). For these reasons, follow-up analysis presented below and computational optimization efforts
have focused on the R enantiomer unless otherwise noted.
Figure 7 plots the dose response curves for (R)-C562-1101 and for comparison NSC-77053,
which were performed in triplicate over a range of compound concentrations under the same assay
conditions. IC50 values were obtained by fitting to a sigmoid dose-response equation using a hillslope =
1. Encouragingly, in the GFP-based assay, the estimated IC50 value for (R)-C562-1101 at 14.2 ± 1.7
µM is about three-fold lower than that of NSC-77053 at 48.5 ± 6.5 µM. It is important however to
emphasize the identification of (R)-C562-1101 represents only a first-stage lead discovery outcome and
that follow-up work to evaluate properties other than activity should be performed including metabolic,
toxic, pharmacokinetic, and related physicochemical descriptors.
Figure 7. Dose response plots for (a) (R)-C562-1101 and (b) control NSC-77053.
Overall, the current results suggest that (R)-C562-1101 is a promising hit compound for
optimization. Since the ligand is considered a mimic of the known substrate tetrapeptide RIME (see
23

Figure 9a) that includes isoleucine in the sequence, a (R)-Ile analog of (R)-C562-1101 (SB-BEI-002)
was also synthesized (see the Methods section). It was found that the inhibitory activity of this analog
was in the same level (IC50 ~16 μM) as that of (R)-C562-1101 (IC50 ~14 μM), i.e., no improvement was
observed. In parallel, as additional early steps towards refinement, several hypothetical analogs were
constructed and computationally evaluated using the predicted DOCK structure as a guide (see section
3.5).
3.3 MD analysis of active compounds. In conjunction with experimental testing, fully flexible
explicit-solvent MD simulations were carried out for each of the 9 compounds identified in the initial
SNAP-Etide assay to help gauge the overall steric and electrostatic stability of the compounds in the
binding site in terms of their predicted binding geometry. Figure 8 shows running block-averaged
RMSD, and estimated free energies of binding (Gbind) using the MM-GBSA method (see section 2.9)
over a 20 ns simulation for each ligand initiated from the docked pose. Interestingly, although a number
of compounds reasonably maintain their predicted binding pose and show stable Gbind values,
compound (R)-C562-1101, which was the most active compound in the SNAP-Etide assay and the only
one that displayed activity in the GFP-based assay, was particularly well-behaved. Specifically, (R)-
C562-1101 maintains a low ligand RMSD of ~2 Å (Figure 8a, orange line) during the 20 ns simulation
and had one of the most favorable Gbind values (Figure 8b, orange line). Visualization of (R)-C562-
1101 geometries taken along the trajectory (Figure 8c) confirms the stability of the binding pose. It
should be emphasized that although such ligand behavior during MD simulations in-and-of-itself does
not guarantee a given compound will be an inhibitor, it does provide additional supporting evidence for
compatibility with the target. A recent study43 by our group examined the effects of using MD
simulations to gauge structural and energetic variability of binding geometries through comparison of
experimentally observed poses vs low-energy decoys over a large test set. The study also examined the
behavior of active and inactive compounds. While additional investigations should be pursued, taken
24

together, the results suggest that, on average, active-like compounds show good overall behavior in
terms of stable and low ligand RMSDs and Gbind values.
Figure 8. (a) Ligand RMSDs and (b) estimated Gbind values vs time for 9 compounds showing activity in the
SNAP-Etide assay. (R)-C562-1101 shown with an orange line. (c) Structural overlays for (R)-C562-1101 from
100 evenly spaced snapshots along the trajectory.
3.4 Binding analysis of (R)-C562-1101. Together, the experimental assay results and computational
analysis suggest (R)-C562-1101 is most promising compound identified from the virtual screen and will
be the focus of the discussions that follow. Notably, (R)-C562-1101 was selected from among top-
ranked compounds in the FPSVDW+ES ranked list and therefore expected to share similar geometric
features and make similar energetic interactions as the reference tetrapeptide RIME in the BoNT/E
catalytic site. Visual overlay of the two binding poses (docked (R)-C562-1101, x-ray RIME) confirms
significant structural overlap (Figure 9a) with several notable features in common (Figure 9b vs 9c): (1)
the carbonyl oxygen of the pentanamide moiety of (R)-C562-1101 occupies a nearly identical position
as the backbone oxygen of arginine on RIME, which displaces the nucleophilic water present seen in the
BoNT/E apo structure, and is situated to make similar contacts with both the catalytic zinc ion and the
hydroxyl of Tyr350 (Figure 9 dashed lines), (2) the hydrophobic isobutyl group of (R)-C562-1101
occupies a nearly identical position as the isoleucine side chain of RIME for which both are buried in
the S1’ hydrophobic pocket8 formed by residues Thr159, Phe191, and Thr208, (3) a sulfonamide
oxygen of (R)-C562-1101 overlays with the backbone oxygen of isoleucine on RIME and positioned to
25

make a similar H-bond with Arg347 (Figure 9 dashed lines), and (4) the substituted benzoxazole group
of (R)-C562-1101 overlaps structurally with carboxylic acid on RIME and positioned to make similar
H-bonds with the backbone amide hydrogen of Gly352 and the hydroxyl side chain of Thr246 (Figure 9
dashed lines). Major differences include the ability of RIME to make a salt bridge (Glu158) and extra
hydrogen bond (Glu212) as discussed further below (Figure 9b vs 9c).
Figure 9. (a) (R)-C562-1101 (orange) and tetrapeptide RIME (green) in the BoNT/E light chain active domain
binding pocket with protein residues hidden for clarity. (b) Subset of residues making key interactions with (R)-
C562-1101 (orange). (c) Subset of residues making key interactions with RIME (green). Interactions within
reasonable H-bonding distance (1.7 - 3.3 Å) shown in magenta.
In terms of quantitative energetic similarities, Figure 10 compares the per-residue energy
footprints made by docked (R)-C562-1101 (Figure 10, orange) versus x-ray RIME (Figure 10, green) for
the fifty BoNT/E residues with the most significant van der Waals (VDW) interactions and electrostatic
(ES) interactions shown explicitly, and lesser contributions summed together into the residue labeled
REMAIN. For comparison, the plot also shows an "average" footprint made by (R)-C562-1101 over the
20 ns MD simulation (Figure 10, blue). In terms of the VDW similarities (Figure 10 top, orange vs
green), (R)-C562-1101 well-matches the overall RIME footprint with near perfect overlap in terms of
magnitude at residues Thr159, Asn160, Phe191, His211, Asn242, Glu250, Tyr350, and Gly352. The
principal difference is the favorable VDW peak at Thr246 observed for (R)-C562-1101 for which RIME
is unfavorable (discussed further below). Importantly, the MD-averaged footprint for (R)-C562-1101
(Figure 10 top, blue) shows only minor VDW differences versus the originally docked pose (Figure 10
26

top, orange) thereby demonstrating that RIME-like steric packing interactions can be maintained by a
small organic ligand during normal thermal fluctuations.
Figure 10. Footprint comparison plots with the BoNT/E light chain active domain for the crystallographic
reference pose for tetrapeptide RIME (green), docked pose for (R)-C562-1101 (orange), and average footprint
over an ensemble of conformations from 20 ns MD simulations (blue). The fifty most significant residues are
shown explicitly with remaining interactions combined into the residue labeled REMAIN.
However, in contrast to the high overlap seen in the VDW footprints, (R)-C562-1101 (Figure 10
bottom, orange) does not show the same quantitative correspondence as RIME in terms of the
magnitude of ES peaks with the notable exception of Arg347 and zinc. A primary cause of the
difference is the presence of charged arginine and glutamic acid functionality on RIME which can make
much stronger H-bonds than the neutral functionality of (R)-C562-1101 as well as form salt bridges. As
an example, strong ES interactions (Figure 10 bottom, large green peaks) are observed for R with
Glu158, the N-terminus of R with Glu212, for E with Thr246, and for E with Gly352 with computed
contributions in the range of -2 to -6 kcal/mol (Figure 10 bottom, green peaks). Taken together, the
visual analysis and ES footprints results suggests: (1) a lack of comparable functionality on (R)-C562-
1101 explains the absence of strong interactions with Glu158 and Glu212, (2) a lack of charged
27

functionality explains the small ES peaks at Thr246 and Glu352, and (3) the reduced peak at Tyr350 is
attributed to sub-optimal H-bonding geometry. As described in the next section, a series of (R)-C562-
1101 analogs were devised to try and recapitulate the ES patterns made by RIME, for example, by
adding charged ligand functionality to mimic RIME's H-bonding within the BoNT/E site. In general
the "average" ES footprint for (R)-C562-1101, similar to that seen for the VDW patterns, shows only
minor differences with regards to the originally docked pose (Figure 10 bottom, orange vs blue)
indicating that the ES interactions are also maintained during the MD simulations.
An interesting result to note is that the MD results predict a smaller distance (2.00 Å ± 0.11 Å
MD-average vs 2.39 Å docked) between the (R)-C562-1101 carbonyl oxygen with zinc thus yielding a
stronger ES interaction (Figure 10c bottom, residue ZINC, blue). As a consequence, the isobutyl group
of (R)-C562-1101 is pulled slightly away from the S1' pocket which leads to somewhat less favorable
van der Waals interactions with Thr159 and Asn160 (Figure 10 top, orange vs blue), and the initially
favorable ES interaction with Glu212 becomes unfavorable (Figure 10 bottom, orange vs blue). Such
behavior for VDW footprints is often observed when very strong ES interactions are involved. Stated
another way, although the total interaction energy for a given ligand may be favorable, energy
decomposition can yield generally small unfavorable interactions at specific residues as a result of very
favorable ES interactions often involving ions.
3.5 Computationally designed analogs. To computationally investigate how functional group
alteration would alter ligand behavior in the binding pocket, and to predict if affinity of the initial hit
could be improved, we generated 7 hypothetical analogs of (R)-C562-1101 (Figure 11, compounds
SBA1 to SBA7) and performed MD simulations, footprint similarity analysis, and estimated free
energies of binding (Gbind). As discussed above, the predicted binding geometry of (R)-C562-1101 in
comparison to RIME (Figure 9) shows that although the two molecules coincide at several positions, the
ES patterns are different (Figure 10, green vs orange lines). Thus, while fashioning new analogs, the
initial focus was on placement of functional groups at positions on (R)-C562-1101 that could more
28

directly mimic the electrostatic interaction patterns made by the native substrate RIME through the
addition of H-bond donors and acceptors (compare molecules in Figure 11 to Figure 9c).
Figure 11. 2D structures for seven analogs derived from parent compound (R)-C562-1101.
Specifically (see Figure 11), analogs were designed based with the following hypotheses in mind:
(1) SBA1 replaced the benzoxazole carbonyl oxygen with a carboxylic acid (Figure 11 blue) in an
attempt to mimic the RIME glutamic acid sidechain that makes contacts with Thr244 and the backbone
amide group of Gly352. (2) SBA2 was modified by adding a hydroxyl group (Figure 11 magenta) to
the ortho position of the dimethylphenyl ring to mimic the H-bond observed between Glu212 and the
free amine on the N-terminus of RIME. (3) SBA3 replaced one of the methyl groups at the ortho
position of the dimethylphenyl ring with a guanidinium group (Figure 11 red) to mimic the RIME
arginine side chain that makes a salt-bridge with Glu158. (4) SBA4 was modified to have both a
carboxylic acid (as in SBA1) and a guanidinium group (as in SBA3) in an attempt to strengthen the ES
interactions with Thr244, Gly352, and Glu158. (5) SBA5 combined the carboxylate functionality from
29