Molecular dynamics simulations of viral neuraminidase inhibitors with the human neuraminidase enzymes: Insights into isoenzyme selectivity

Article abstract of DOI:10.1016/j.bmc.2018.05.035

Inhibitors of viral neuraminidase enzymes have been previously developed as therapeutics. Humans can express multiple forms of neuraminidase enzymes (NEU1, NEU2, NEU3, NEU4) that share a similar active site and enzymatic mechanism with their viral counterparts. Using a panel of purified human neuraminidase enzymes, we tested the inhibitory activity of 2-deoxy-2,3-dehydro-N-acetylneuraminic acid (DANA), zanamivir, oseltamivir, and peramivir against each of the human isoenzymes. We find that, with the exceptions of DANA and zanamivir, these compounds show generally poor activity against the human neuraminidase enzymes. To provide insight into the interactions of viral inhibitors with human neuraminidases, we conducted molecular dynamics simulations using homology models based on coordinates reported for NEU2. Simulations revealed that an organized water is displaced by zanamivir in binding to NEU2 and NEU3 and confirmed the critical importance of engaging the binding pocket of the C7–C9 glycerol sidechain. Our results suggest that compounds designed to target the human neuraminidases should provide more selective tools for interrogating these enzymes. Furthermore, they emphasize a need for additional structural data to enable structure-based drug design in these systems.

Full text of DOI:10.1016/j.bmc.2018.05.035

Accepted Manuscript
Molecular dynamics simulations of viral neuraminidase inhibitors with the hu-
man neuraminidase enzymes: Insights into isoenzyme selectivity
Michele R. Richards, Tianlin Guo, Carmanah D. Hunter, Christopher W. Cairo
PII:
S0968-0896(18)30640-0
https://doi.org/10.1016/j.bmc.2018.05.035
BMC 14377
DOI:
Reference:
Bioorganic & Medicinal Chemistry
To appear in:
Received Date:
Revised Date:
Accepted Date:
30 March 2018
17 May 2018
22 May 2018
Please cite this article as: Richards, M.R., Guo, T., Hunter, C.D., Cairo, C.W., Molecular dynamics simulations of
viral neuraminidase inhibitors with the human neuraminidase enzymes: Insights into isoenzyme selectivity,
Bioorganic & Medicinal Chemistry (2018), doi: https://doi.org/10.1016/j.bmc.2018.05.035
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.

Molecular dynamics simulations of viral neuraminidase inhibitors
with the human neuraminidase enzymes: Insights into isoenzyme
selectivity
Michele R. Richards, Tianlin Guo, Carmanah D. Hunter, Christopher W. Cairo*
Alberta Glycomics Centre and Department of Chemistry, University of Alberta, Edmonton, AB
T6G 2G2, Canada
Dedicated to Professor Laura L. Kiessling on the receipt of the 2018 Tetrahedron Prize
*Correspondence, Tel.: 780 492 0377; fax: 780 492 8231; e-mail: ccairo@ualberta.ca

Abstract:
Inhibitors of viral neuraminidase enzymes have been previously developed as
therapeutics. Humans can express multiple forms of neuraminidase enzymes (NEU1, NEU2,
NEU3, NEU4) that share a similar active site and enzymatic mechanism with their viral
counterparts. Using a panel of purified human neuraminidase enzymes, we tested the inhibitory
activity of 2-deoxy-2,3-dehydro-N-acetylneuraminic acid (DANA), zanamivir, oseltamivir, and
peramivir against each of the human isoenzymes. We find that, with the exceptions of DANA
and zanamivir, these compounds show generally poor activity against the human neuraminidase
enzymes. To provide insight into the interactions of viral inhibitors with human neuraminidases,
we conducted molecular dynamics simulations using homology models based on coordinates
reported for NEU2. Simulations revealed that an organized water is displaced by zanamivir in
binding to NEU2 and NEU3 and confirmed the critical importance of engaging the binding
pocket of the C7–C9 glycerol sidechain. Our results suggest that compounds designed to target
the human neuraminidases should provide more selective tools for interrogating these enzymes.
Furthermore, they emphasize a need for additional structural data to enable structure-based drug
design in these systems.
2

1
Introduction
Sialic acids, also known as neuraminic acids, are α-keto nonulosonic acids that are
typically the terminal carbohydrate of glycoproteins and glycolipids.1 The most common sialic
acid in humans is N-acetylneuraminic acid (Neu5Ac 1, Figure 1), though other forms are
known.2 Due to their placement at the periphery of the glycan, sialic acids act as receptors for
pathogens, as well as signaling molecules for the immune system,1 and their expression is
controlled by the interplay of sialyltransferases (SiaTs) and neuraminidase enzymes (NEUs).3
Four human neuraminidases (hNEUs) have been identified: NEU1, NEU2, NEU3, and NEU4.4
The expression of hNEUs has been linked to cancer, diabetes, and cardiovascular disease.5-7
Human NEUs may play roles in inflammation through modification of the sialyl-Lewisx (CD15s)
antigen on leukocytes.8, 9 Although a variety of studies have been conducted using biological
models where hNEUs have been disrupted by genetic methods,4, 6 the availability of competitive
inhibitors for these enzymes should provide important tools to reveal the role of individual
isoenzymes in disease and could form the basis of new therapeutic strategies.10
Inhibitors of influenza NEU form the basis of clinically available antiviral therapeutics;11
however, these inhibitors may not be generally active against NEUs from different species. Most
NEU inhibitors are based on 2-deoxy-2,3-dehydro-N-acetylneuraminic acid (DANA, 2) as a
parent structure. Zanamivir 3 and oseltamivir 4, analogs of DANA originally designed to target
influenza A neuraminidase, are also active against NanA from Streptococcus pneumoniae, with
IC50 values reported in the low micromolar range.12 An analysis of multiple NanA sequences
suggested that oseltamivir has broad activity against bacterial NEUs.13 Cross species viral–
bacterial NEU inhibition could play a role in the severity of influenza infections, and inhibitors
with activity against NEUs from both species could provide clinical benefit.14 If viral NEU
3

(vNEU) inhibitors have broad specificity for bacterial species, one might expect the same is true
for hNEUs. However, studies that have examined the activity of oseltamivir and zanamivir
against hNEUs have generally observed substantially weaker potency against these enzymes,
typically observing high micromolar to low millimolar IC50 values. Hata et al. found that
oseltamivir had low millimolar inhibition of NEU2, as did zanamivir for NEU1, NEU2, and
NEU4. In contrast, zanamivir had low micromolar activity against the NEU3 isoenzyme.15
Chavas et al. observed mid-to-high micromolar activity for zanamivir and peramivir (5) with
NEU2, but oseltamivir showed very low activity.16 Analogs of oseltamivir tested against NEU3
and NEU4 showed generally low activity and suggested an important role for the C7–C9
glycerol side-chain of Neu5Ac in enzyme recognition.17 Despite low in vitro inhibitory potency
for oseltamivir with NEU1, several reports have investigated its use to block signaling pathways
related to cancer, diabetes, and fibrosis linked to this isoenzyme.18-20 We note that subtle changes
in formulation of oseltamivir can show significant variations in activity in cell-based assays.21 It
has been previously suggested that mutations in hNEUs could make individuals more susceptible
to adverse effects of antiviral medications.22 The de-sialylation of platelets is a major mechanism
of cell clearance in transfusions, and hNEUs also appear to be involved in senescence and
autoimmune mechanisms of platelet clearance.23, 24 Inhibitors of NEU have been investigated for
enhancement of platelet storage by targeting NEU from contaminating bacteria.25
While inhibitors optimized to act as antivirals show weaker activity against hNEUs, the
design of inhibitors that specifically target these enzymes has identified important features of
their binding sites.10 The development of inhibitors tailored to the human NEUs has been
impeded by limited structural data on the enzymes. Of the four hNEUs, only NEU2 has been
crystallized – both in its apo form and in complex with several inhibitors, including DANA,
4

zanamivir, and peramivir.16, 26 While the substrate Neu5Ac is found in a 2C5 conformation in
solution,27 inhibitors 2–4 are all six-membered rings with half-chair conformations that
presumably mimic the transition state of catalyzed Neu5Ac cleavage.28 Magesh et al. identified
that C9-amide derivatives of DANA were active against NEU1.29 Our group has reported
inhibitors with C9- or C4,C9-modifications are able to specifically target NEU4 and NEU3
enzymes with nanomolar potency and excellent selectivity.30, 31
To expand our understanding of differences between the active sites of hNEUs and their
viral counterparts, we set out to examine the activity of known viral inhibitors against a panel of
human NEUs. We first developed homology models for NEU1, NEU3, and NEU4, and used
these, along with the crystal structures of NEU2, as starting points for molecular dynamics (MD)
simulations to provide insight into the selectivity of 3 toward NEU2 and NEU3. Furthermore, we
measured the IC50 and inhibitory constants (Ki) of compounds 1–5 against individual hNEU to
provide experimental context for our models. The results of the simulations, and their relation to
the design of new selective inhibitors of the four NEU isoenzymes, is described.
Figure 1. Compounds evaluated in this study.
5

2
Materials and methods
Compounds 1 and 5 were purchased from Dextra Laboratories Ltd. and Ontario
Chemicals Inc., respectively, and used without further purification. Compounds 2 and 3 were
synthesized as described previously.31 Oseltamivir phosphate was purchased from Sigma-
Aldrich, and the ethyl ester was hydrolyzed with NaOH to generate the carboxylate form 4,
which was used for assays.32 See Supporting Information for details of the protocol.
2.1 Km determinations
A sample of the enzyme (5 μL; 0.25 mU for NEU1, 0.5 mU for NEU2 – NEU4) was
incubated in a 384-well plate with reaction buffer (10 µL; 0.1 M sodium acetate buffer) at the
enzyme’s optimum pH, and 4MU-NANA (5 µL; 0-400 µM). The rate of product formation was
monitored using fluorescence (λex. = 315 nm λem. = 450 nm) every 2 minutes for 30 minutes
(NEU1, NEU3, and NEU4) or every 20 seconds for 5 minutes (NEU2). Michaelis-Menten
kinetics were determined from duplicate measurements using nonlinear regression in GraphPad
Prism 7.0.
2.2 Inhibition assays
Human NEU2, NEU3, and NEU4 enzymes were expressed as MBP-fusion proteins in E.
coli and purified as previously reported.17, 33 NEU1 was overexpressed as a (His)6 fusion protein
in HEK293 cells and used as a crude preparation from cell lysate.34 All assays were conducted in
0.1 M sodium acetate buffer at optimum pH for each enzyme (pH 4.5 for NEU1, NEU3, and
NEU4; pH 5.5 for NEU2).35 To get comparable IC50 values among the four isoenzymes, similar
activity of each enzyme was used in the assay based on 4MU-NANA (4’-methylumbelliferyl α-
6

D-N-acetylneuraminic acid) activity. Assays were performed using protocols reported
previously.35
Inhibitors over a concentration range of 3-fold serial dilutions were incubated with
enzyme at 0 °C for 15 min. 4MU-NANA was then added to the mixture, bringing the final
concentration of 4MU-NANA to 50 μM and the total volume of the reaction mixture to 20 μL.
After incubation at 37 °C for 30 min, the reaction was quenched with 100 μL of 0.2 M sodium
glycine buffer (pH 10.2). The reaction mixture was transferred to 386-well plate and the enzyme
activity was determined by measuring fluorescence (λ = 365 nm; λ = 445 nm) using a plate
ex
em
reader (Molecular Devices, Sunnyvale CA). Assays were performed with duplicates for each
point and IC50 was obtained by plotting the data with GraphPad Prism 7.0. For curves that
showed less than a 50% decrease in signal, fits were conducted using the maximum inhibition
values found for DANA.
2.3 Ki determinations
Enzymes were incubated with serial concentrations of inhibitors at 0 °C for 15 min and
serial concentrations of 4MU-NANA were then added. The reaction mixture was transferred to a
384-well plate immediately and the rate of product formation was obtained by measuring
fluorescence (λ = 315 nm; λ = 450 nm) every 1 min for 30 min. The data were processed with
ex
em
GraphPad Prism 7.0 for Ki determination.
2.4 Computational methodology
2.4.1 Enzyme models
Molecular dynamics (MD) simulations were run for compounds 1–5 bound to the four
hNEU isoenzymes. We used the crystal structure of NEU2 bound to 2 (PDB ID: 1VCU)26, 36 as
7

the initial structure for NEU2 and added the missing non-terminal residues G227, E228, S284,
G295, P286, and G287 using Modeller in Chimera.37-40 For NEU3 and NEU4, we used our
previously reported homology models.2, 31, 33, 41 Additionally, the model for NEU3 was modified
to eliminate unreasonable conformations using Modeller in Chimera. We built a homology
model for NEU1 also using Modeller in Chimera and the alignment reported by Magesh, et al.29,
42 The alignments for the four isoenzymes are shown in Figure 2. Residues 1–53 in NEU1 and
290–367 in NEU4 have no homology to NEU2 and were removed before running MD
simulations.
2.4.2 Inhibitor models
The starting coordinates for the inhibitors were obtained for 1 from 4NCS,43, 44 for 2 from
1VCU,26, 36 for 3 from 2F0Z, 16, 45 for 4 from 2QWK,46, 47 and for 5 from 2F10.16, 48 We kept the
active site waters from 4NCS for simulations with 1, from 1VCU for simulations with 2, from
2F0Z for simulations with 3, from 2QWK for simulations with 4, and from 2F10 for simulations
with 5. We aligned each of the enzymes NEU1, NEU3, and NEU4 (MatchMaker in Chimera38-40
separately with NEU2 to obtain the initial positions for 1–5 bound to the homology models. All
compounds were simulated with the protonation states at physiological pH (i.e. as carboxylate,
ammonium, and/or guanidinium ions). See Supporting Information for starting structures of MD
simulations in PDB format,49, 50 along with the files necessary for running inhibitors 2–5 in
AMBER.
)
8

Figure 2. Alignment of the four hNEU isoenzymes.
9

2.4.3 Simulations
All simulations were run in AMBER 1551 using pmemd.cuda (GPU acceleration) on
Nvidia GeForce GTX 980 GPUs. The ff14SB force field52 was used for the proteins,
GLYCAM06 was used for 1,53 and the general AMBER force field (GAFF)54 was used for 2–5.
Hydrogens were added to 2–4 using the program Avogadro,55, 56 then partial charges for 2–4
were assigned using AM1 with bond charge correction (AM1-BCC) model57 in the antechamber
module of AmberTools15.51 We used our previously reported parameters and charges for 5.58
The enzyme–inhibitor complexes were neutralized with the addition of Na+ ions or Cl– ions as
necessary, and Joung-Cheatham parameters were used for the ions.59 All complexes were
solvated in a box of TIP3P water60 with 10 Å between the solute and the edges of the box in all
three dimensions. For all systems, the water was first minimized using 100 steps of steepest
descent, followed by 4900 steps of conjugate gradient. Then the entire system was minimized
with 100 steps of steepest descent, followed by 4900 steps of conjugate gradient. The systems
were further equilibrated by heating from 5 K to 300 K over 50 ps, followed by cooling back to 5
K over an additional 50 ps. After the annealing step, the systems were again heated from 5 K to
300 K over 100 ps, then allowed to run at 300 K for 100 ps before the production simulations
were started. Production was run for 100 ns in each system. The timestep was 2 fs, bonds to
hydrogen were constrained with the SHAKE61 algorithm, and the cutoff for non-bonded
interactions was 8.0 Å. The temperature was maintained with the Berendsen thermostat62 (ntt =
1) with velocities rescaled every 1 ps. The simulations were analyzed using the cpptraj module
of AmberTools15.51, 63
10

3
Results and discussion
3.1 Activity of viral inhibitors against hNEU
Previous reports have investigated the activity of viral inhibitors with hNEU isoenzymes.
In most cases, the 4MU-NANA substrate is used to allow for rapid analysis using fluorescence
spectroscopy. The activity of the purified enzymes for this substrate have been reported
elsewhere;15 and we sought to measure the activity of our recombinant enzymes before
investigating the activity of inhibitors using 4MU-NANA as a substrate (Table 1). Of the four
neuraminidases, NEU4 had the lowest measured Km of 17 ± 2 μM. We observed a Km of 48 ± 9
μM for NEU3, in agreement with previous data.33, 64 These results are in contrast to studies using
overexpressed sources of hNEU isoenzymes; which observed much higher Km values for NEU1
and NEU2.15 We note that previous work with our recombinant enzyme preparations have found
good agreement between their activity and enzymes from eukaryotic cells.31, 65 It is interesting to
note that the activity of NEU4 for the 4MU-NANA substrate has been consistently found to be
higher than the other isoenzymes, suggesting some substrate preference for the 4-
methylumbulliferyl aglycone.2, 15 The NEU2, NEU3, and NEU4 isoenzymes are reported to
modify glycolipid substrates to varying degrees, in contrast to NEU1 which generally prefers
glycoproteins.66, 67 Our data suggest that 4MU-NANA shows similar Km values for all four
isoenzymes, which should allow for comparison of inhibition assay data using this substrate.
Table 1. Km values of 4MU-NANA against hNEUs
Enzyme
NEU1
NEU2
NEU3
Km (μM) a
56 ± 18
119 ± 21
48 ± 9
NEU4
17 ± 2
a
Error shown is standard error calculated from nonlinear regression.
11

Testing of compounds 1–5 for inhibitor activity against hNEU was performed with 4MU-
NANA as the substrate to determine IC50 (Table 2) and Ki values (Table 3). We observed that
NEU1 had undetectable inhibition from all compounds tested, with the exception of 2, which had
an IC50 of 49 ± 8 µM (Ki of 12 ± 1 µM). Compound 2 was the only compound which had activity
against all four isoenzymes, ranging from 8–50 µM.31 The monosaccharide, Neu5Ac 1, had
undetectable inhibition of all four isoenzymes; as did oseltamivir 4. Peramivir 5 only had
detectable inhibition against NEU2 (IC50 of 70 ± 7 µM; Ki of 56 ± 13 µM) and was inactive
against all other isoenzymes. Previous reports have observed higher Ki values for peramivir 5
with NEU2.16 Zanamivir 3 was the most potent compound tested and showed activity against
NEU2, NEU3, and NEU4 consistent with previous reports.15, 16, 31 Zanamivir had sub-
micromolar activity against NEU3 (Ki 0.62 ± 0.09 µM) but only had low micromolar activity
against NEU2 and NEU4. These inhibition data suggest that the only viral NEU inhibitor tested
with notable activity against hNEU isoenzymes was zanamivir. We have previously identified 4-
guanidino analogs of DANA as selective inhibitors of NEU3 when combined with modifications
at the C9 position.31 Additionally, these data are consistent with previous investigations of
oseltamivir analogs with hNEU, which found that replacement of the amino group of 4 with a
guanidine was insufficient to improve activity of these compounds against NEU3 and NEU4
isoenzymes.17 In earlier analysis of the conformation of peramivir in solution and in viral
enzyme active sites, we observed large changes in ring conformation.58 Based on our results
here, we sought to re-evaluate the binding of viral compounds to models of the hNEU
isoenzymes using molecular modeling.
12

Table 2. IC50 (μM) valuesa for 1–5 against hNEUs
Compound
1 - Neu5Ac
2 - DANA
3 - Zanamivir
4 - Oseltamivir
NEU1
>500
49 ± 8
>500
>500
>500
NEU2
>500
37 ± 6
7.8 ± 2.0
>500
NEU3
>500
7.7 ± 0.8
4.0 ± 0.6
>500
NEU4
>500
8.3 ± 1.0
47 ± 6
>500
5 - Peramivir
70 ± 7
>500
>500
a
Error shown is standard error calculated from nonlinear regression.Table 3. Ki
(μM) valuesa for inhibitors 1–5 against hNEUs
Compound
1 - Neu5Ac
2 - DANA
3 - Zanamivir
4 - Oseltamivir
NEU1
NA
12 ± 1
NA
NEU2
NA
25 ± 4
5.7 ± 1.5
NA
NEU3
NA
1.6 ± 0.3
0.62 ± 0.09
NA
NEU4
NA
5.8 ± 0.6
26 ± 4
NA
NA
5 - Peramivir
NA
56 ± 13
NA
NA
a
Error shown is standard error calculated from nonlinear regression.
3.2 MD simulations of hNEUs with viral inhibitors
There is limited structural data for the hNEU enzymes, with crystal structures only
having been reported for NEU2,16, 26 though the structure of NEU3 has been investigated using
STD NMR.41 Molecular modeling has been previously used to develop homology models of
NEU1-4.33, 35, 42, 68, 69 The proposed key residues for each enzyme, based on sequence alignment
(Figure 2) and site-directed mutagenesis,33, 70 are listed in Table 4. We docked each of the
inhibitors 1–5 in the active sites of the homology models and conducted molecular dynamics
(MD) simulations for 100 ns for each system.
13

Table 4. Key active site residues for hNEU homology models
Catalytic
tyrosine
General
acid/base
Enzyme
NEU1
NEU2
NEU3
Arginine triad
Other active site residues
H220, T222, E95, R97,
D103, L139 a
Y370
Y334
Y370
Y419
R78, R341, R280
R21, R304, R237
R25, R340, R245
R23, R389, R242
E264
E218
E225
E222
Q270, Y179, Y181, E111,
E39, R41, D46, N86
H277, Y179, Y181, E113,
E43, R45, D50, N88
W274, Y177, Y179, E111,
E41, R43, D48, N86
NEU4
a
In NEU1, there are no clear homologs to Q270 or E111 (NEU2).
We found that our model of NEU1 was unreliable and likely does not provide significant
insight into the active site of this isoenzyme. In simulations with NEU1, only 3 and 5 remained
close to the active site of the enzyme; however, the distance between the carboxylate and the
nucleophilic tyrosine-OH increased in both cases. In simulations of 3, this intermolecular
distance increased from 5.1 Å after equilibration to 10.9 Å at the end of the simulation, with a
maximum distance of 14.4 Å. For 5, there was a smaller increase from 4.5 Å after equilibration
to 7.6 Å at the end of the simulation, with a maximum distance of 10.4 Å. The guanidinium
group acts as an anchor during the simulations with 3 and 5, keeping them in contact with NEU1
for a larger portion of the simulation. The remaining compounds left the active site of NEU1
within 42–58 ns of the simulation. These results led us to conclude that the NEU1 model cannot
be used reliably to identify specific protein–inhibitor interactions without further refinement. We
observed that all compounds evaluated remained bound to NEU2, NEU3, and NEU4 during the
simulations and further details of those simulations are provided below.
3.2.1. Ring conformations
In previous MD simulations with trisaccharide substrates binding to hNEUs, we observed
that the Neu5Ac ring remained in 2C5 conformation in the active site of NEU2 and NEU3.2, 71
14

However, modifications at the C9 position of the residue could drive the conformation to B2,5.
We decided to analyze the ring conformations of free Neu5Ac in these simulations. In the
simulation with NEU1, the average conformation of 1 was 2C5 (see Table 5, and Supporting
Information), which is the expected solution conformation since the ligand leaves the active site
of NEU1 within 44 ns of the simulation. The ring conformation of 1 during simulations with
NEU2, NEU3, and NEU4, however, was not 2C5. Rather, the range of ring conformations varied
depending on the isoenzyme. In simulations with NEU2, the ring conformations of 1 spanned a
range from 4,OB to 3SO, with the average at 6S2. With NEU3, the ring conformations of 1 ranged
between 4S2 and B4,O for 80% of the simulation and was in the solution conformation, 2C5, for
the remainder of the simulation. For NEU4, the range of ring conformations of 1 was the
smallest observed, 4,OB to 3,6B, with the average between 4S2 and B2,5. Presumably, the hydrogen
bonds between the carboxylate and the arginine triad compensate for the higher energies of the
boat and skew conformers.
Table 5. Average ring conformations of 1–5 during the MD simulations
Compound
1 - Neu5Ac
2 - DANA
3 - Zanamivir
4 - Oseltamivir
NEU1
2C5
NEU2
6S2
NEU3
3,6B a
6H5
NEU4
4S2/B2,5
6H5
6H5
6H5
6H5
6H5
6H5
6H5
4H5
4H5
4H5
E3
4H5
5 - Peramivir
2T3/E3
4T3
4E
a
80% of the population is OS3–B2,5 conformation, while 20% is in the 2C5 conformation. See
supporting information for more details.
The ring conformation of Neu5Ac in the active site of NEU enzymes has often been
observed in boat or skew conformations. A co-crystal structure of the vNEU observed 4S2 and
B2,5.46 Additionally, Tvaroška and coworkers have run MD simulations of Neu5Ac in solution
and bound to the vNEU N1.72, 73 While the 2C5 conformation was the minimum energy
conformation in solution, the 4,OB/OS3 conformation was only 2.4 ± 0.4 kcal mol-1 higher in
15

energy. In simulations in the N1 neuraminidase, differing ring conformations for Neu5Ac were
observed depending on which ligand was bound. For free Neu5Ac, the major population was
2C5, although there was a minor population at 6S2. Taken together, the MD simulations described
here, as well as previous work,2, 72, 73 demonstrated that sialic acid ring conformation depends
both on environment and substitution at the anomeric and N5 positions.
The ring conformation of 2–4 remained the same in all MD simulations, regardless of
isoenzyme evaluated (Table 4, and Supporting Information). All these compounds contain an
endocyclic alkene, which maintains a half-chair conformation when bound to the enzyme or free
in solution (as seen in simulations with NEU1). However, as we have previously observed, the
conformation of the cyclopentane ring in 5 varies over a range of 180° on the pseudorotational
wheel, depending on its environment (Figure 3 and Supporting Information).58 For simulations
with NEU1 and NEU3, the conformation of 5 was close to the solution conformation, 2T3/E3 and
E3, respectively. With NEU2, the average conformation (4T3) is between solution conformation
and that in the crystal structure (2F10);48 while with NEU4, the average conformation of 4E is
quite similar to that in the crystal structure of NEU2. The flexibility of the ring of 5 allows its
functional groups to make different contacts with NEU2, NEU3, and NEU4 as the other
compounds (vide infra).
16

Figure 3. A. The ring atoms of peramivir 5 are numbered the same as our previous study.58 B.
Average ring conformation of 5 in MD simulations with NEU1 (tan), NEU2 (grey), NEU3
(green), NEU4 (blue). The conformation of 5 from the crystal structure with NEU2 is also shown
(magenta, PDB ID: 2F10), as well as the average conformation from solution (black).58
3.2.2. Key points of contact between inhibitors and hNEUs
As discussed above, the homology model for NEU1 is inadequate, as such we have
limited the discussion here to NEU2, NEU3, and NEU4. Average structures from simulations of
compounds 1–5 in the active sites of NEU2, NEU3, and NEU4 shown in Figure 4–Figure 6.
Ten conformers for each inhibitor relative to the average structure of each hNEU can be found in
Supporting Information.
17

Figure 4. Average structures from MD simulations for Neu5Ac 1 with NEU2 (A.), NEU3 (B.),
and NEU4 (C.) 1 is shown in purple, and side chains for key residues are shown, along with the
surface representation of the enzymes.
18

Figure 5. Average structures from MD simulations for DANA 2 and zanamivir 3. The inhibitors
are shown in magenta, and side chains for key residues are shown, along with the surface
representation of the enzymes. A. NEU2 with 2, B. NEU2 with 3, C. NEU3 with 2, D. NEU3
with 3, E. NEU4 with 2, F. NEU4 with 3.
19

Figure 6. Average structures from MD simulations for oseltamivir 4 and peramivir 5. The
inhibitors are shown in magenta, and side chains for key residues are shown, along with the
surface representation of the enzymes. A. NEU2 with 4, B. NEU2 with 5, C. NEU3 with 4, D.
NEU3 with 5, E. NEU4 with 4, F. NEU4 with 5.
20

Compounds 1–5 all contain a carboxylate that forms highly populated (65%–97%)
hydrogen bonds to the arginine triad in the active sites of NEU2, NEU3, and NEU4. For most of
the simulations, this hydrogen bond is to the central arginine residue – R304 for NEU2, R340 for
NEU3, and R389 for NEU4. One exception is the simulation with 3 and NEU3, where 3 is tilted
in the active site, exposing the carboxylate to solvent (Figure 5D). Consequently, the most
populated hydrogen bond between the carboxylate and NEU3 is to Y370 for only 5% of the
simulation. Additionally, 5 makes hydrogen bond contacts to a different arginine of the triad –
R21 for NEU2 (54%), R245 for NEU3 (81%), and R23 (42%). These hydrogen bonds are also
not as populated as those to the carboxylate in the simulations for 1–4. This difference in
hydrogen bond occupancy may be a result of increased ring flexibility in 5.
Lectins and glycosidases often have shallow active sites, with the hydroxyl groups of
carbohydrates displacing ordered water molecules upon binding.74 With that in mind, our models
retained ordered water molecules observed in the starting crystallography structures that each
model was based on (see Sec 2.4.2 for details). In the crystal structure of NEU2 with 3-fluoro-
Neu5Ac (4CNS),43, 44 there are two water molecules in the active site – one in the C4 pocket
(HOH 686) and one in the NAc pocket (HOH 515). These solvent molecules remain bound for
84 ns of the simulation of 1 with NEU2, but they leave the active site by 10 ns in the simulation
with NEU3 and between 30–48 ns in the simulation with NEU4. Similarly, there is a water
molecule in the C4-pocket of the crystal structure of NEU2 with 2 (HOH 395 A, 1VCU).26, 36
That water molecule remains in the active sites of both NEU2 and NEU3 over the entire
simulation with 2. However, the equivalent water in NEU4 leaves the active site within 68 ns.
The C4-pocket of the viral N9 neuraminidase in complex with 4 also contains two water
molecules (HOH 1343S and 1349S, 2QWK).46, 47 These water molecules leave the active sites of
21

NEU2 by 15 ns, NEU3 by 38 ns, and NEU4 by 21 ns. The guanidinium groups of 3 and 5 take
the place of the C4 active-site water in the crystal structure of NEU2 with 2 (compare 1VCU
with 2F0Z and 2F10),26, 36, 45, 48 and none of the ordered waters from the crystal structures remain
in the active sites during any of the simulations with NEU2, NEU3, or NEU4.
The active site waters form water bridges between 1 and NEU2, NEU3, and NEU4
during the MD simulations; specifically, with 1, E39, and N86 for NEU2 (73%) and with 1, E41,
and N88 and 1 for NEU4 (34%). In NEU3, the corresponding residues were not involved in a
water bridge; however, there is a water bridge between 1 and E113 in NEU3 that is populated for
52% of the simulation. This solvent bridge in NEU4 (between 1 and E111) is also seen for 63%
of the simulation. In simulations with 2, the active site waters form solvent bridges among 2,
E39, and N86 in NEU2 (89%); 2, E43, and N88 in NEU3 (67%); and 2, E41, and N86 in NEU4
(81%). The C4-hydroxyl in 1 and 2 is replaced by an ammonium in 4. Hydrogen bonds to
solvent dominate this site, but water bridges were less occupied than for 1 and 2 (0%–43%),
indicating that the water was less organized around the ammonium group.
While 2 shows inhibitory activity for all four hNEU, 3 is approximately 10-fold selective
for NEU2 and NEU3 as compared to NEU4. The guanidinium of 3 takes the place of an active
site water in the crystal structure of NEU2 (2F0Z),45 and there are hydrogen bonds from the
guanidinium to E39 in NEU2 (62%), to E43 in NEU3 (45%), and to E41 in NEU4 (52%), all
residues that were involved in water bridges in simulations with 2. However, the asparagine
residues of those solvent bridges do not engage the guanidinium of 3. Thus, while the
guanidinium of 3 occupies a similar space as the active site water in simulations with 2, the
contacts between 3 and the enzymes are not the same as those in the water bridge with 2 and the
enzymes. The guanidinium group of 5 forms similar hydrogen bonds in NEU2 and NEU4 – to
22

E39 in NEU2 for 86% of the simulation and to E41 in NEU4 for 99% of the simulation.
However, for NEU3, the guanidinium of 5 forms highly populated hydrogen bonds to a different
residue, E225 (87%).
Compounds 1–3 contain the glycerol side chain found in the native substrates for hNEUs,
and we have observed that changes to this side chain affect selectivity and potency of
inhibitors.35 In our MD simulations, hydrogen bonds to solvent dominate these sites with 1,
except for the simulation with NEU4, in which an intramolecular hydrogen bond between HO7
and O1 was populated for 77% of the simulation (Figure 4C). In simulations with 2 and 3, HO7
participates in the most populated hydrogen bonds to the enzymes (38%–80%) – to E111 in
NEU2, to E113 in NEU3, and to E111 in NEU4. The other potential hydrogen bonding sites on
the glycerol side chain of 2 and 3 are dominated by solvent, apart from the simulation with 3 and
NEU4. In that simulation, HO9 forms a hydrogen bond to E222 for 70% of the simulation.
The pentyl side chain of 4 and 5 is known to fit into a hydrophobic pocket in the viral
enzymes;75, 76 however, it makes contact to mainly polar residues in the simulations between 4
and NEU2, NEU3, and NEU4 (Figure 6). In NEU2, the carbon atoms of the pentyl group are
closest to (4.2–5.3 Å) the phenolic oxygen of Y181 for 90–98% of the simulation. In NEU3, two
of the carbon atoms on the pentyl side chain form a hydrophobic contact to the side chain of
P198, but the remaining atoms are close (4.7–5.3 Å) to one of the guanidinium nitrogens of R245
for 68%–100% of the simulation. The simulation with NEU4 is like that of NEU2, the carbon
atoms of the pentyl group are closest to (4.3–5.0 Å) the phenolic oxygen of Y179 for 85–99% of
the simulation. Many contacts between the pentyl group of 5 are also to polar sites on the
enzymes. For NEU2, two carbons in the pentyl group make hydrophobic contact to the γ-carbon
of T156, while the remaining carbons in the pentyl group are closest to the phenolic oxygen of
23

Y179 (4.6–5.6 Å) for the entire simulation. For NEU3, three carbons in the pentyl group make
hydrophobic contact to the β-carbon of E113; the remaining carbons in the pentyl group are
closest to the phenolic oxygens of Y179 and Y181 (4.2–4.5 Å) for the entire simulation. For
NEU4, two carbons in the pentyl group make hydrophobic contact to the γ-carbon of E111; the
remaining carbons in the pentyl group are closest to oxygen atoms in Y419 or E222 (4.5–5.3 Å)
for the entire simulation.
4
Conclusions
Our experimental determinations of the activity of antiviral compounds 1–5 confirm that
these inhibitors have generally poor activity against hNEU.15 Among these, only zanamivir 3
shows activity in the low micromolar range (primarily for NEU2 and NEU3). We used molecular
dynamics and homology modeling of hNEU enzymes to identify features which help explain the
low activity of these compounds. Significant features that differ between the viral enzymes and
hNEUs include the binding pockets for the glycerol sidechain and the C4 substituent.16
Furthermore, our results reinforce that the activity of inhibitors designed for the vNEU enzymes
against the family of hNEUs cannot be assumed to be identical. In fact, these compounds are
generally low in activity against hNEUs, and specifically-designed inhibitors are likely to
provide better research and therapeutic avenues.10
MD simulations identified the glycerol sidechain and C4 binding pockets of hNEU
enzymes as recognition elements that were not adequately engaged by vNEU inhibitors. Chavas
et al. have previously identified substantial differences in the recognition of the C7–C9 sidechain
of Neu5Ac between NEU2 and vNEU.16 The structure of NEU2 identified E111, Y179, and
Y181 residues as coordinating the glycerol sidechain of zanamivir.16 Furthermore, structures
with peramivir bound in the active site of NEU2 showed substantial re-arrangement of these
24

residues. Our MD simulations observed similar changes in the glycerol sidechain binding pocket
for NEU3 and NEU4.
Of the 10 wild-type NEU2 crystal structures16, 26, 36, 43-45, 48, 77-82 in the Protein Data
Bank,83 eight have inhibitors bound, and six of these have an equivalent active site water to the
one described above (HOH 395 A in 1VCU).26 The structures with 345 and 548 bound lack the
active site water, as do the two apo forms.26, 77, 78 The active site waters appear to be organized
by the ligand. Our MD simulations also support a role for organized water in the C4 pocket of
the active sites of NEU2 and NEU3 with DANA, which may help account for the increased
potency of zanamivir and derivatives featuring a C4 guanidinium group.31 The rigid nature of the
guanidinium effectively mimics the organized active site water observed in simulations with
DANA and NEU2 and NEU3. The C4 pocket, and the differential activity of zanamivir among
hNEUs, suggests that this pocket could be used to provide isoenzyme selectivity. Despite the low
potency of antiviral compounds tested here for hNEUs, it is worth noting that the DANA parent
scaffold is clearly capable of forming the basis of potent inhibitors. The broad specificity of
DANA across species and hNEUs suggests that the scaffold can provide essential contacts in the
active site. The DANA scaffold is insufficient alone to provide potent inhibitors, but elaboration
has been successful in identifying potent inhibitors for NEU3 and NEU4.30, 31
The MD simulations with NEU1 demonstrate that the homology model is flawed, which
is not surprising as NEU1 has the lowest similarity to NEU2 at 22%, compared with 42%
similarity between NEU2 and NEU3, and 44% similarity between NEU2 and NEU4. In addition,
assays with NEU1 are performed with enriched membrane preparations with cathepsin present,
unlike the other three hNEUs that are active in purified form. The homology model of NEU1
may require interactions with members of the complex or a membrane environment to provide
25

predictive results in simulations with 2. We also ran MD simulations with the NEU1-selective
inhibitor C9-BA29 to further test our homology model (data not shown), and that compound left
the active site within 5 ns of the simulation.
The results in the present work support the growing need for additional structural data on
other isoenzymes of the hNEU family. While reported structures of NEU2 have been
instrumental for the design of new inhibitors, homology models for the remaining isoenzymes
are incomplete and, therefore, continue to be a limitation in structure-based design of selective
inhibitors. The membrane-associated nature of NEU1, NEU3, and NEU4 is a likely explanation
for why these enzymes have not yet been characterized as thoroughly as NEU2. In the absence of
well-resolved atomic structures, biochemical, substrate, and inhibitor studies, in combination
with other spectroscopic methods, must continue to fill in the gaps in our understanding of the
differences between the hNEU isoenzymes.41, 65, 68
5
Abbreviations
4MU-NANA
DANA
GAFF
hNEU
MBP
4-methylumbelliferyl α-D-N-acetylneuraminic acid
2-deoxy-2,3-dehydro-N-acetylneuraminic acid
general AMBER force field
human neuraminidase
maltose binding protein
MD
molecular dynamics
NEU
neuraminidase
Neu5Ac
SiaT
N-acetylneuraminic acid
sialyltransferase
vNEU
viral neuraminidase
26

6
Acknowledgements:
Funding: This work was supported by the Natural Sciences and Engineering Research
Council of Canada (NSERC), the Alberta Glycomics Centre, and the Canadian Glycomics
Network.
7
References
1.
Varki A. Sialic acids in human health and disease. Trends Mol Med. 2008;14(8): 351–
360.
2.
Hunter CD, Khanna N, Richards MR, et al. Human neuraminidase isoenzymes show
variable activities for 9-O-acetyl-sialoside substrates. ACS Chem Biol. 2018;13(4): 922–
932.
3.
4.
5.
6.
7.
8.
9.
Chen X, Varki A. Advances in the Biology and Chemistry of Sialic Acids. ACS Chem
Biol. 2010;5(2): 163–176.
Miyagi T, Yamaguchi K. Mammalian sialidases: Physiological and pathological roles in
cellular functions. Glycobiology. 2012;22(7): 880–896.
Miyagi T, Wada T, Yamaguchi K, Hata K. Sialidase and malignancy: A minireview.
Glycoconjugate J. 2003;20(3): 189–198.
Pshezhetsky AV, Hinek A. Where catabolism meets signalling: neuraminidase 1 as a
modulator of cell receptors. Glycoconjugate J. 2011;28(7): 441–452.
Dridi L, Seyrantepe V, Fougerat A, et al. Positive regulation of insulin signaling by
neuraminidase 1. Diabetes. 2013;62(7): 2338–2346.
Gadhoum SZ, Sackstein R. CD15 expression in human myeloid cell differentiation is
regulated by sialidase activity. Nat Chem Biol. 2008;4(12): 751–757.
Sanders WJ, Katsumoto TR, Bertozzi CR, Rosen SD, Kiessling LL. L-
Selectin−Carbohydrate Interactions:ꢀ Relevant Modifications of the Lewis x
Trisaccharide. Biochemistry. 1996;35(47): 14862–14867.
10.
11.
Cairo CW. Inhibitors of the human neuraminidase enzymes. MedChemComm. 2014;5(8):
1067–1074.
Von Itzstein M. The war against influenza: discovery and development of sialidase
inhibitors. Nat Rev Drug Discov. 2007;6(12): 967–974.
27

12.
Gut H, Xu G, Taylor GL, Walsh MA. Structural Basis for Streptococcus pneumoniae
NanA Inhibition by Influenza Antivirals Zanamivir and Oseltamivir Carboxylate. J Mol
Biol. 2011;409(4): 496–503.
13.
14.
Xu Z, Von Grafenstein S, Walther E, et al. Sequence diversity of NanA manifests in
distinct enzyme kinetics and inhibitor susceptibility. Sci Rep. 2016;6: 25169.
Walther E, Xu Z, Richter M, et al. Dual acting neuraminidase inhibitors open new
opportunities to disrupt the lethal synergism between Streptococcus pneumoniae and
influenza virus. Front Microbiol. 2016;7: 357.
15.
Hata K, Koseki K, Yamaguchi K, et al. Limited Inhibitory Effects of Oseltamivir and
Zanamivir on Human Sialidases. Antimicrob Agents Chemother. 2008;52(10): 3484–
3491.
16.
17.
Chavas LMG, Kato R, Suzuki N, et al. Complexity in Influenza Virus Targeted Drug
Design: Interaction with Human Sialidases. J Med Chem. 2010;53(7): 2998–3002.
Albohy A, Mohan S, Zheng RB, Pinto BM, Cairo CW. Inhibitor selectivity of a new class
of oseltamivir analogs against viral neuraminidase over human neuraminidase enzymes.
Bioorg Med Chem. 2011;19(9): 2817‒2822.
18.
19.
Haxho F, Haq S, Szewczuk MR. Biased G protein-coupled receptor agonism mediates
Neu1 sialidase and matrix metalloproteinase-9 crosstalk to induce transactivation of
insulin receptor signaling. Cell Signal. 2018;43: 71–84.
O'Shea LK, Abdulkhalek S, Allison S, Neufeld RJ, Szewczuk MR. Therapeutic targeting
of Neul sialidase with oseltamivir phosphate (Tamiflu (R)) disables cancer cell survival
in human pancreatic cancer with acquired chemoresistance. OncoTargets Ther. 2014;7:
117–134.
20.
21.
Karhadkar TR, Pilling D, Cox N, Gomer RH. Sialidase inhibitors attenuate pulmonary
fibrosis in a mouse model. Sci Rep. 2017;7(1): 15069.
Amith SR, Jayanth P, Franchuk S, et al. Dependence of pathogen molecule-induced Toll-
like receptor activation and cell function on Neu1 sialidase. Glycoconjugate J.
2009;26(9): 1197.
22.
23.
24.
Li C-Y, Yu Q, Ye Z-Q, et al. A nonsynonymous SNP in human cytosolic sialidase in a
small Asian population results in reduced enzyme activity: potential link with severe
adverse reactions to oseltamivir. Cell Research. 2007;17(4): 357-362.
Li J, Van Der Wal DE, Zhu G, et al. Desialylation is a mechanism of Fc-independent
platelet clearance and a therapeutic target in immune thrombocytopenia. Nat Commun.
2015;6: 7737.
Li R, Hoffmeister KM, Falet H. Glycans and the platelet life cycle. Platelets. 2016;27(6):
505–511.
28

25.
26.
27.
28.
29.
30.
Liu QP, Hoffmeister K, Sackstein R. Platelet Storage and Reduced Bacterial Proliferation
in Platelet Products Using a Sialidase Inhibitor. Google patents. US20160000064A1.
2016.
Chavas LMG, Tringali C, Fusi P, et al. Crystal Structure of the Human Cytosolic
Sialidase Neu2: Evidence for the Dynamic Nature of Substrate Recognition. J Biol Chem.
2005;280(1): 469–475.
Klepach T, Zhang W, Carmichael I, Serianni AS. 13C−1H and 13C−13C NMR J-
Couplings in 13C-Labeled N-Acetyl-neuraminic Acid: Correlations with Molecular
Structure. J Org Chem. 2008;73(12): 4376–4387.
Chong AKJ, Pegg MS, Taylor NR, vonItzstein M. Evidence for a sialosyl cation
transition‐ state complex in the reaction of sialidase from influenza virus. Eur J Biochem.
1992;207(1): 335–343.
Magesh S, Moriya S, Suzuki T, Miyagi T, Ishida H, Kiso M. Design, synthesis, and
biological evaluation of human sialidase inhibitors. Part 1: Selective inhibitors of
lysosomal sialidase (NEU1). Bioorg Med Chem Lett. 2008;18(2): 532–537.
Albohy A, Zhang Y, Smutova V, Pshezhetsky AV, Cairo CW. Identification of Selective
Nanomolar Inhibitors of the Human Neuraminidase, NEU4. ACS Med Chem Lett.
2013;4(6): 532–537.
31.
32.
Guo T, Dätwyler P, Demina E, et al. Selective Inhibitors of Human Neuraminidase 3. J
Med Chem. 2018;61(5): 1990–2008.
Albiñana CB, Machara A, Řezáčová P, Pachl P, Konvalinka J, Kožíšek M. Kinetic,
thermodynamic and structural analysis of tamiphosphor binding to neuraminidase of
H1N1 (2009) pandemic influenza. Eur J Med Chem. 2016;121: 100‒109.
33.
34.
Albohy A, Li MD, Zheng RB, Zou C, Cairo CW. Insight into substrate recognition and
catalysis by the human neuraminidase 3 (NEU3) through molecular modeling and site-
directed mutagenesis. Glycobiology. 2010;20(9): 1127–1138.
Pshezhetsky AV, Potier M. Association of N-Acetylgalactosamine-6-sulfate Sulfatase
with the Multienzyme Lysosomal Complex of β-Galactosidase, Cathepsin A, and
Neuraminidase: Possible Implication for Intralysosomal Catabolism of Keratan Sulfate. J
Biol Chem. 1996;271(45): 28359‒28365.
35.
36.
Zhang Y, Albohy A, Zou Y, Smutova V, Pshezhetsky AV, Cairo CW. Identification of
Selective Inhibitors for Human Neuraminidase Isoenzymes Using C4,C7-Modified 2-
Deoxy-2,3-didehydro-N-acetylneuraminic Acid (DANA) Analogues. J Med Chem.
2013;56(7): 2948‒2958.
Chavas LMG, Fusi P, Tringali C, et al. Structure of the human cytosolic sialidase Neu2
(PDB ID: 1VCU); 2004. http://www.rcsb.org/structure/1VCU. Accessed Jan. 18, 2018.
29