Access to this full-text is provided by PLOS.
Content available from PLOS Computational Biology
This content is subject to copyright.
RESEARCH ARTICLE
Extended ensemble simulations of a SARS-
CoV-2 nsp1–5’-UTR complex
Shun SakurabaID
1
*, Qilin Xie
2
, Kota KasaharaID
3
, Junichi IwakiriID
4
, Hidetoshi KonoID
1
1Institute for Quantum Life Science, National Institutes for Quantum Science and Technology, Kizugawa,
Japan, 2Graduate School of Life Sciences, Ritsumeikan University, Kusatsu, Japan, 3College of Life
Sciences, Ritsumeikan University, Kusatsu, Japan, 4Graduate School of Frontier Sciences, The University of
Tokyo, Kashiwa, Japan
*sakuraba.shun@qst.go.jp
Abstract
Nonstructural protein 1 (nsp1) of severe acute respiratory syndrome coronavirus 2 (SARS-
CoV-2) is a 180-residue protein that blocks translation of host mRNAs in SARS-CoV-2-
infected cells. Although it is known that SARS-CoV-2’s own RNA evades nsp1’s host trans-
lation shutoff, the molecular mechanism underlying the evasion was poorly understood. We
performed an extended ensemble molecular dynamics simulation to investigate the mecha-
nism of the viral RNA evasion. Simulation results suggested that the stem loop structure of
the SARS-CoV-2 RNA 5’-untranslated region (SL1) binds to both nsp1’s N-terminal globular
region and intrinsically disordered region. The consistency of the results was assessed by
modeling nsp1-40Sribosome structure based on reported nsp1 experiments, including the
X-ray crystallographic structure analysis, the cryo-EM electron density map, and cross-link-
ing experiments. The SL1 binding region predicted from the simulation was open to the sol-
vent, yet the ribosome could interact with SL1. Cluster analysis of the binding mode and
detailed analysis of the binding poses suggest residues Arg124, Lys47, Arg43, and Asn126
may be involved in the SL1 recognition mechanism, consistent with the existing mutational
analysis.
Author summary
The pandemic of COVID-19 is still rampant all over the world as of 2021 June. SARS-
CoV-2 (severe acute respiratory syndrome coronavirus 2), the causative pathogen of
COVID-19, encodes a protein called nsp1 (nonstructural protein 1), which modulates and
hijacks the ribosome of the infected host cells. With nsp1, infected human cells selectively
translate SARS-CoV-2’s RNA, which increases the virus reproduction efficiency while
evading the host immunity. Though it has been known that nsp1 recognizes characteristic
stem-loop structure at 5’-end of SARS-CoV-2’s RNA (called SL1), the molecular mecha-
nism underlying the recognition has been poorly understood. We investigated the mecha-
nism of selective translation using the all-atom molecular dynamics simulation of nsp1-
SL1 complex. Our simulation results suggest that the binding between nsp1 and SL1 is
multi-modal. The results also imply that both the N-terminal globular part and the
PLOS COMPUTATIONAL BIOLOGY
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 1 / 21
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Sakuraba S, Xie Q, Kasahara K, Iwakiri J,
Kono H (2022) Extended ensemble simulations of
a SARS-CoV-2 nsp1–5’-UTR complex. PLoS
Comput Biol 18(1): e1009804. https://doi.org/
10.1371/journal.pcbi.1009804
Editor: Bert L. de Groot, Max Planck Institute for
Biophysical Chemistry, GERMANY
Received: June 27, 2021
Accepted: January 4, 2022
Published: January 19, 2022
Peer Review History: PLOS recognizes the
benefits of transparency in the peer review
process; therefore, we enable the publication of
all of the content of peer review and author
responses alongside final, published articles. The
editorial history of this article is available here:
https://doi.org/10.1371/journal.pcbi.1009804
Copyright: ©2022 Sakuraba et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: The data and code to
reproduce the research is included in the
Supporting information and BSMA archive (https://
bsma.pdbj.org/entry/26).
C-terminal flexible tail of nsp1 are involved in the binding. The residues involved in nsp1-
SL1 binding coincides with the known mutant analyses of SARS-CoV-1 and SARS-CoV-
2, as well as experimental evidence about nsp1-ribosome interactions.
Introduction
SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2) belongs to Betacoronaviridae,
and is the causative pathogen of COVID-19. Nonstructural protein 1 (nsp1) resides at the
beginning of SARS-CoV-2’s genome, and it is the first protein translated upon SARS-CoV-2
infection. After self-cleavage of open reading frame 1a (orf1a) by an orf1a-encoded protease
(nsp3; PLpro), nsp1 is released as a 180-residue protein. SARS-CoV-2 nsp1 is homologous to
nsp1 of SARS-CoV-1, the causative pathogen of SARS, sharing 84% sequence identity with the
SARS-CoV-1 protein. Nsp1 functions to suppress host gene expression [1–6] and induce host
mRNA cleavage, [1,2,7–9] effectively blocking translation of host mRNAs. The translation
shutoff hinders the host cell’s innate immune response including interferon-dependent signal-
ing. [1,10] Multiple groups have recently reported cryogenic electron microscopy (cryo-EM)
structures of SARS-CoV-2 nsp1–40Sribosome complexes. [11–13] The structural analysis
showed that two α-helices are formed in the C-terminal region (153–160, 166–179) of nsp1
and binds to the 40Sribosome. These helices block host translation by shutting the ribosomal
tunnel used by the mRNA. This blockade inhibits the formation of the 48Sribosome pre-initi-
ation complex, which is essential for translation initiation. [3,13] But while nsp1 shuts down
host mRNA translation, it is known that the viral RNAs are translated even in the presence of
the nsp1, and that they evade degradation. [2–4]
These mechanisms force infected cells to produce only viral proteins instead of normal
host cell proteins; indeed, in a transcriptome analysis, 65% of total RNA reads from Vero cells
infected with SARS-CoV-2 were mapped to the viral genome. [14] It has also been shown that
nsp1 recognizes the 5’-untranslated region (5’-UTR) of the viral RNA [4,6,12] and selectively
enables translation of RNAs that have a specific sequence. The first stem loop in the 5’-UTR [4,
6,15] has been shown to be necessary for translation initiation in the presence of nsp1. Specifi-
cally, with SARS-CoV-1, [4] bases 1–36 of the 5’-UTR enable translation of viral RNA; with
SARS-CoV-2, bases 1–33 [15] or 1–40 [6] of the 5’-UTR of SARS-CoV-2 enable translation.
However, the precise molecular mechanism remains poorly understood.
In the present research, therefore, our aim was to accumulate information about the molec-
ular mechanism by which SARS-CoV-2 RNA evades nsp1. As a first step, we focused on how
and where SARS-CoV-2 5’-UTR binds to nsp1, and tackled the problem from computational
simulations. We modeled and simulated a complex comprised of SARS-CoV-2 nsp1 and the
SARS-CoV-2 5’-UTR’s first stem loop using extended ensemble molecular simulations. The
simulations suggested the importance of the nsp1’s C-terminal disordered region as well as
that of the globular region. The binding preference of the 5’-UTR onto nsp1 was assessed, and
its consistency to the current ribosome-nsp1 model was investigated to further confirm the
simulation results.
Materials and methods
Overview
We constructed a complex of nsp1 and 5’-UTR of SARS-CoV-2 RNA and performed simula-
tions to investigate the mechanism behind the self-evasion of the nsp1’s translation shutoff.
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 2 / 21
Funding: SS was supported by a Grant-in-Aid for
Early-Career Scientists from the Japan Society for
the Promotion of Science (JSPS; https://www.jsps.
go.jp/english/), Japan (JP16K17778), by Grants-in-
Aid for Scientific Research (A) from the JSPS
(JP16H02484 and JP21H04912), and by a Grant-
in-Aid for Scientific Research on Innovative Areas
from the Ministry of Education, Culture, Sports,
Science and Technology (MEXT; https://www.
mext.go.jp/en/; JP19H05410). KK was supported
by a Grant-in-Aid for Scientific Research (C) from
the JSPS (JP20K12069). JI was supported by a
Grant-in-Aid for Scientific Research (C) from the
JSPS (JP20K12041). HK was supported by by
Platform Project for Supporting Drug Discovery
and Life Science Research (Basis for Supporting
Innovative Drug Discovery and Life Science
Research (BINDS)) from AMED under Grant
Number JP21am0101106, Agency for Medical
Research and Development (AMED; https://www.
amed.go.jp/en/), Japan. The funders had no role in
study design, data collection and analysis, decision
to publish, or preparation of the manuscript.
Competing interests: The authors have declared
that no competing interests exist.
Nsp1 has an intrinsically disordered region (IDR) and is considered to bind to the RNA. How-
ever, it has generally been considered that RNA-protein complexes are difficult to simulate
because structures tend to be trapped around the initial configurations in a reasonable simula-
tion time, due strong charge-charge interactions between the RNA and the positively charged
protein residues. To ease the problem, we performed an extended ensemble simulation. In
extended ensemble simulations, modified energy functions are used to sample various possible
structures of complexes. The effect of modified energy functions can be statistically removed
in the post-process phase (with a procedure called reweighting), thereby enabling us to obtain
structures of the nsp1-RNA complex at the given temperature in a comparably shorter simula-
tion time than the conventional molecular dynamics (MD) simulations. After performing the
simulation, we analyzed the trajectory to investigate which residues in nsp1 are contacting
RNA and how the structure is formed.
Simulation setup
Nsp1 is a partially disordered 180-residue protein, in which the structures of residues 12–127
and 14–125 have been solved by X-ray crystallography in SARS-CoV-1 and SARS-CoV-2,
respectively. The structures of other residues (1–11, 128–180) are unknown, and residues 130–
180 are thought to be an IDR. [16,17] We constructed the SARS-CoV-2 nsp1 structure using
homology modeling based on the SARS-CoV-1 nsp1 conformation (Protein Data Bank (PDB)
ID: 2HSX [16]). Modeling was performed using MODELLER. [18] We noted that SARS-CoV-
1 nsp1 and SARS-CoV-2 nsp1 are aligned without gaps. The structure of the IDR was con-
structed so as to form an extended structure. For nsp1, we used the AMBER ff14SB force field
[19–22] in the subsequent simulations.
The initial structure of the RNA stem was constructed using RNAcomposer. [23,24] Bases
numbered 1–35 from the SARS-CoV-2 reference genome (NCBI reference sequence ID
NC_045512.2) [25] were used in the present research. This sequence corresponds to the first
stem loop of the SARS-CoV-2 RNA 5’-UTR. Hereafter, we will call this RNA “SL1.” SL1 was
capped by 7-methyl guanosine triphosphate (m7G-ppp-). The first base (A1) after the cap was
methylated at the 2’-O position to reflect the viral capped RNA. Charges and bonded force
field parameters for these modified bases were respectively prepared using the restrained elec-
trostatic potential (RESP) method [26] and analogy to existing parameters. For SL1, we used a
combination of AMBER99 + bsc0 + χOL3. [19,20,27,28] To maintain the structural stability
of the stem loop, we employed distance restraints between the G-C bases. Specifically, between
residues G7–C33, G8–C32, C15–G24 and C16–G23, distance restraints were applied such that
the distances between the N1, O6 and N2 atoms of guanosine and the N3, N4 and O2 atoms
of cytidine, did not exceed 4.0 Å. Between these atoms, flat-bottom potentials were applied,
where each potential was zero when the distance between two atoms was less than 4.0 Å, and a
harmonic restraint with a spring constant of 1 kJ mol
−1
Å
−2
was applied when it exceeds 4.0 Å.
We used acpype [29] to convert the AMBER force field files generated by AmberTools [30]
into GROMACS. Parameter files are presented in S1 File.
The nsp1 and SL1 models were then merged and, using TIP3P [31] water model with Joung-
Cheatham monovalent ion parameters [32] (73,468 water molecules, 253 K
+
ions, 209 Cl
−
ions),
were solvated in 150 mM KCl solution. The initial structure is presented in Fig 1A. A periodic
boundary condition using a rhombic dodecahedron unit cell was used with a size of ca. 140 Å
along the X-axis. Note that we started the simulation from the unbound state; that is, nsp1 and
SL1 were not directly in contact with each other. The total number of atoms in the system was
224,798. After preparing the system we also prepared the system without SL1 by removing it
from the nsp1-SL1 initial structure (the total number of atoms was 223,598).
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 3 / 21
Although it is possible to perform a MD simulation of an nsp1-SL1 complex, due to the
excessive charges on both molecules, the model tends to be trapped around the initial configu-
ration of the complex in conventional MD simulations. Authors have previously shown that
the sampling for nucleic acid–protein systems can be effectively solved by extended ensemble
simulations. [33–35] In this work, we used replica exchange with solute tempering (REST) ver-
sion 2 to sample various configurations of SL1 and the nsp1 IDR. [36] In REST2, the simula-
tions are performed so that specific residues (called a “hot” region) have weaker interactions
with others than the conventional MD simulations. This modification to the potential function
prevents the simulation to be trapped around the initial configuration. We set both the disor-
dered region (nsp1 1–11 and 128–180) and the entire SL1 as the “hot” region of the REST2
Fig 1. Structures of nsp1. (A) Initial structure before starting the simulation. (B) Structureof the complex at 50 ns in the 0th replica (i.e., the simulation with the
unscaled potential). (C) Structures from superimposition of 20 representative snapshots of the nsp1-SL1 complex. Snapshots were obtained from a weighted random
sampling. Different snapshots from SL1 are colored differently. (D) Nsp1 segmentation used in the analysis: (i) residues 1 to 18, green; (ii) residues 31 to 50, cyan; (iii)
residues 74 to 90, magenta; (iv) residues 121 to 146, orange; (v) residues 147 to 180, blue.
https://doi.org/10.1371/journal.pcbi.1009804.g001
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 4 / 21
simulation. Therefore, even though some base pairs of SL1 were restrained, the interactions
between nsp1 and SL1 as well as SL1 and solvent were scaled in the REST2 simulation, allow-
ing broader configurations to be sampled. Note that in addition to the charge scaling for nsp1
and SL1, we also scaled the charges of counter-ions to prevent unneutralized system charge in
the Ewald summation. The total number of replicas used in the simulation was 192. The rep-
lica numbered 0 corresponds to the simulation with the unscaled potential. In the final replica
(numbered 191), nonbonded potentials between “hot”-“hot” groups were scaled by 0.25.
Exchange ratios were 53–78% across all replicas. To prevent numerical errors originating from
the loss of significant digits, we used a double-precision version of GROMACS as the simula-
tion software. [37] We also modified GROMACS to enable the replica exchange simulation
with an arbitrary Hamiltonian. [38] The patch representing modifications is supplied in
S2 File.
The simulation was performed for 50 ns (thus, 50 ns×192 = 9.6 μs in total), and the first 25
ns were discarded as the equilibration time. The simulation was performed with NVT and the
temperature was set to 300 K. The temperature was controlled using the velocity rescaling
method. [39] The timestep was set to 2 fs, and hydrogens attached to heavy atoms were con-
strained with LINCS. [40] Similar to the nsp1-SL1 complex, we also performed the simulation
of nsp1 only (wihtout SL1) with exactly the same condition for 50 ns (another 9.6 μs simulation
in total). Simulation input files and trajectories for the 8 lowest numbered replicas are depos-
ited at Biological Structure Model Archive (BSMA; entry ID 26) https://bsma.pdbj.org/entry/
26. Full trajectories for all replicas used in this research are available upon request.
In addition to these extended ensemble simulations, we performed 8 MD simulations of
500 ns length each, starting from the initial configuration of the nsp1-SL1 complex to see the
difference between conventional simulations and extended ensemble simulations. First 50 ns
chunks of the simulations were removed from the data as the equilibration time, and remain-
ing 450 ns simulation results were used in the subsequent analyses.
Simulation analysis
As the simulations were performed with modified potential functions, we performed the
reweighting procedure to subtract the effect of modified potential functions. We used the mul-
tistate Bennett acceptance ratio (MBAR) method [41,42] to calculate statistical weights to the
structures in trajectories; in essence, structures that are difficult to obtain without potential
modification have smaller weight values. With that method, we obtained a weighted ensemble
corresponding to the canonical ensemble (trajectory with a weight assigned on each frame)
from multiple simulations performed with different potentials. Only eight replicas correspond-
ing to the eight lowest replica indices (i.e., the one with the unscaled potential function and
seven replicas with the potentials closest to the unscaled potential) were used in the MBAR
analysis. The weighted ensemble of the trajectory was used in the subsequent analyses. Visuali-
zation was performed using VMD [43] and pymol [44]. The secondary structure of nsp1 was
analyzed using the definition of DSSP [45] with mdtraj. [46]
For both nsp1 simulations with and without SL1, we analyzed the intra-residue contact
within nsp1. We defined the contact between nsp1 residues by having at least one inter-atomic
distance between heavy atoms less than or equal to 4 Å, or having a residue number difference
no larger than 2. From the simulation ensemble, we calculated the ratio of the contact between
residues by taking the weighted average.
The relative orientation of the SL1 and nsp1 IDR was analyzed using principal axes (the eas-
iest axis for the rotation) of two groups. For SL1, phosphate atoms in the stem loop region (res-
idues 7–33) were used for the principal axis calculation; the sign of the principal axis vector
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 5 / 21
was chosen to match the direction along C19 to G7’s phosphate atoms. For nsp1, C-terminal
end of globular region and N-terminal side of IDR (residues 121–146) were used, and the sign
was chosen such that the direction matches that from residue 121 Cαatom to residue 146 Cα
atom. The angle between two axes was used to analyze the orientational preference between
the two.
Clustering. After we obtained multiple poses of the nsp1-SL1 complex from the simula-
tion, we classified structures into clusters. Typically, measures such as the root-mean-square
deviation (RMSD) are used to distinguish different structures; however, for the IDR, the
RMSD is not informative because the structures are more diverse, and also because the RMSD
is extremely sensitive to motions far from the center of mass. We thus used the contact infor-
mation between nsp1 and SL1 residues to analyze the simulation results. Here, inter-residue
contacts were detected with the criterion that the inter-atomic distance between the Cαof an
amino acid residue and C4’ of a nucleotide residue was less than or equal to 12 Å.
On the basis of the inter-residue contact information, the binding modes of the nsp1–SL1
complex observed in the ensemble were evaluated by applying the clustering method. The
inter-residue contact information in each snapshot was represented as a contact map consist-
ing of a 180 ×36 binary matrix. The distance between two snapshots was then calculated as the
Euclidian distance of binary vectors with 180 ×36 = 6480 elements. We applied the DBSCAN
method [47] to classify the binding modes. We arbitrarily determined two parameters, eps
and minPts, for the DBSCAN method to obtain a reasonable number of clusters each of
which had distinct binding modes (the validity of parameters is also discussed in Fig A in S1
Text). Note that the DBSCAN generates clusters each of which has more than minPts mem-
bers based on the similarity threshold eps. The clusters with fewer than minPts members
(including singletons) were treated as outliers. We used eps = 6 and minPts = 200 in this
research.
After clusters were obtained, we applied two other criteria to characterize interactions
between nsp1 and SL1 in each cluster. (i) Hydrogen bonds were detected with the criteria that
the hydrogen-acceptor distance was less than 2.5 Åand the donor–hydrogen–acceptor angle
was greater than 120 degrees. (ii) Salt-bridges were detected with the criterion that the distance
between a phosphorous atom in the RNA backbone and the distal nitrogen atom of Arg or Lys
was less than 4.0 Å.
We also extracted 10 representative structures corresponding to each cluster. These repre-
sentative structures are also deposited to the BSMA archive. We assessed the stability of these
representative structures in clusters 1 and 2 (clusters having the two largest populations) by
running a simulation from representative structures. Four structures were sampled from clus-
ter 1 and 2 each by randomly resampling structures with weight factors obtained from the
reweighting. Then, 500 ns conventional MD simulations from these 4 ×2 structures were per-
formed with new random initial velocities assigned. Resulting trajectories were converted to
the binary matrix by the same procedure we used in the clustering, then the distances from the
centers of clusters were calculated to assess the stability.
Modeling nsp1–40Sribosome complex
To compare the binding poses obtained from the simulation with the recent experimental
results, we modeled the complex structure of nsp1 and 40Sribosome based on the density map
from the cryo-EM and the cross-linking experiment. We first modeled an nsp1–40S ribosome
complex by the density fitting approach. It has been reported that, in the ribosome–nsp1 com-
plex cryo-EM density map (Electron Microscopy Data Bank ID: EMD-11276), where a chunk
of electron density was observed near the C-terminal structures of nsp1, which is considered
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 6 / 21
to be the N-terminal globular region of nsp1. [11] We fitted the SARS-CoV-2 nsp1 N-terminal
domain structure (PDB ID: 7K3N) into the density map using the structure of 40S ribosome–
nsp1 C-terminal helices complex (PDB ID: 6ZLW) to find appropriate candidates of nsp1 N-
terminal region. We used UCSF Chimera [48] to fit the density map. Six models with the cor-
relation coefficient greater than 0.80 were found, and were used for further analysis.
It has been reported that nsp1 and ribosomal protein S3 could form cross-links with tar-
geted in situ cross-linking mass spectrometry. [49] Two inter-residue crosslinks between nsp1
K120–S3 K62 and nsp1 K141–S3 K108 were reported, where the lysine residue in nsp1 in the
latter pair was mapped to the IDR. We measured the distance between Cαatoms at nsp1 K120
and S3 K62 of 6 nsp1–40Sribosome candidate structures. We selected candidate 2 as the
model because the distance between cross-linked residues met the criterion (<25 Å) and the
number of collisions between Cαatoms was the lowest (Table A in S1 Text). For the conve-
nience of the readers we deposited the model structure of nsp1 bound to the ribosome to
BSMA.
Results and discussion
Convergence of the extended ensemble simulation
We first monitored the convergence of the ensemble using the secondary structure distribu-
tion and the stability of the hydrogen bonds between nsp1 and SL1 (Text A and Figs B and C
in S1 Text). The hydrogen bond and secondary structure statistics reached a plateau at *30
ns. However, as expected from the relatively short simulation length and large number of repli-
cas, the replica states were not well mixed. The replica state indices of each continuous trajec-
tory were limited in a narrow range, demonstrating that the sampling is still insufficient (Fig D
in S1 Text). Our simulation trajectories henceforth should be recognized as a set of meta-stable
structures without acheiving the total convergence to the canonical ensemble. Nevertheless,
the cluster analysis of conventional MD results starting from the initial structure indicates that
the structures from REST2 extended ensemble simulations resulted in a totally different struc-
ture obtained in the conventional MD (Fig E in S1 Text). Furthermore, we observed that
major structure clusters obtained from the REST2 simulation were stable with the conven-
tional MD (we will discuss in “Clustering analysis of the binding poses”). The limitations of
the present calculation will be discussed in “Limitations of this study”.
The IDR partially forms secondary structure and binds to SL1
Although we did not restrain the RNA-nsp1 distance in the simulation and started the simula-
tion with the two molecules apart, they formed a complex within the simulation. Fig 1B shows
a representative snapshot of the complex at the end of the simulation. The RNA stem binds to
the C-terminal disordered region. However, as shown in Fig 1C, when the N-terminal domain
of nsp1 was superimposed, the RNA structures did not have a specific conformation. This
implies that there was no distinct, rigid structure mediating nsp1-RNA binding.
We next investigated the secondary structure of the nsp1 region simulated with SL1 (Fig 2).
Although we started the simulation from an extended configuration, the C-terminal region at
residues 153–179 partially formed two α-helices, which is consistent with the fact that the C-
terminal region forms two helices (residues 153–160, 166–179) and shuts down translation by
capping the pore that mRNA goes through in the cryo-EM structural analysis. The result also
indicates that the cap structure may be formed before nsp1 binds to the ribosome, reflecting a
pre-existing equilibrium, although the ratio of the helix-forming structures is only up to 50%.
In addition to these known helices, residues 140–150 also weakly formed a mixture of α-helix
and 3–10 helix. Residues at other regions (1–11, 128–139) remained disordered. We also
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 7 / 21
Fig 2. Secondary structure distribution of nsp1. Probabilities were calculated using the reweighting of the last 25 ns
simulation trajectories.
https://doi.org/10.1371/journal.pcbi.1009804.g002
Table 1. Hydrogen bonds observed between SL1 and nsp1.
Nsp1 residue Main/side SL1 base BB/base %
Arg124 Side U18 Backbone 26.0
Lys47 Side C16 Backbone 23.0
Arg43 Side U17 Backbone 19.6
Asn126 Side U17 Backbone 18.7
Gly127 Main U18 Backbone 18.2
Asn126 Side C20 Base 17.4
Ser135 Main C20 Base 14.8
Arg124 Main U17 Base 14.4
Asn126 Side C20 Backbone 13.4
Ser40 Side U17 Backbone 13.1
Asn126 Side C16 Backbone 13.0
Asp75 Main U18 Base 12.7
Asn126 Side U18 Backbone 12.3
Ala131 Main C19 Base 12.2
Ser135 Side C16 Sugar 12.2
Lys47 Side C20 Backbone 12.0
Tyr136 Main C20 Base 11.9
Ser135 Side C20 Base 11.6
His134 Main C19 Base 10.8
Asp75 Side U18 Base 10.4
https://doi.org/10.1371/journal.pcbi.1009804.t001
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 8 / 21
investigated the structure of nsp1 without SL1. There were no substantial change in the sec-
ondary structures except slightly lower α-helix formation ratio at residues 153–160 (Fig F in
S1 Text).
SL1’s hairpin region binds to the nsp1 IDR
Inter-residue contact probabilities between nsp1 and SL1 in the canonical ensemble are sum-
marized in Table 1 and Figs 3and 4. Based on the distribution of the interactions, we catego-
rized the binding interface of nsp1 into five regions (Fig 1D and Table A in S1 Text): (i) the N-
terminus (residues 1–18), (ii) the α1 helix (residues 31–50), (iii) the disordered loop between
β3 and β4 (residues 74–90), (iv) C-terminal end of the globular region and the N-terminal side
of the IDR (residues 121–146), and (v) the C-terminal side of the IDR (residues 147–180).
These five regions interacted primarily with bases around C20 of the RNA fragment, which
composes the stem loop. The most important region for recognition of SL1 was region (iv), the
N-terminal side of the IDR. The probability of contacts between any residue in this region and
SL1 was 97.4%. In particular, contact between Asn126 and U18 was observed in 84.1% of the
canonical ensemble. The most frequently observed hydrogen bond in the canonical ensemble
was Arg124–U18, the probability of which was 26.0% (Table 1). The second most important
interface region was region (ii), α1 helix, which has two basic residues (Arg43 and Lys47), that
frequently formed salt-bridges with the backbone of SL1. At least one salt-bridge in this region
was included in 69.8% of the canonical ensemble. The third most important was region (iii),
consisting of the loop between β3 and β4; 63.2% of the canonical ensemble included at least
one contact in this region. Asp75 sometimes formed hydrogen bonds with the bases of SL1.
Fig 3. Contact probabilities between nsp1 and RNA. Residue-wise, all-against-all contact probability in the canonical ensemble. The
color at each grid point indicates the statistical weight of the contact betweenthe corresponding pair of residues (color scale is shown at
the right of the panel). The points filled by white indicate no detectable probability of contacts. The line plots at the top and right of the
contact map depict the contact probability for each residue, regardless of its counterpart.
https://doi.org/10.1371/journal.pcbi.1009804.g003
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 9 / 21
Regions (i) and (v) tended not to form hydrogen bonds or salt-bridges, but frequently con-
tacted residues in these regions; the probability for interactions with regions (i) and (v) were
72.1% and 59.2%, respectively.
As an overall shape, the nsp1 surface consists of positive and negative electrostatic surface
patches separated by a neutral region (Fig 5A). [50] The α1 helix in region (i) forms the inter-
face between these two patches; one side of the helix contains basic residues (Arg43 and
Lys47), and the other side contains some hydrophobic residues (Val38, Leu39, Ala42, and
Leu46). The positive side of the α1 helix assumes a mound-like shape with a positively charged
cliff (Fig 5B). The bottom of the valley formed by the N-terminus and β3-β4 loop, or regions
(i) and (iii), respectively, also contains positive electrostatic potentials. The positively charged
cliff and valley attract and fit to the negatively charged backbone of SL1. Eventually the IDRs
in region (iv) and (v) grab SL1.
Although the binding site for SL1 on nsp1 can be characterized as an interface consisting of
regions (i) through (v), SL1 did not assume a stable conformation, even when it was bound to
these regions. Diverse binding modes were observed in the canonical ensemble. Although SL1
nearly always interacted with residues in the region (iv), its conformation was diverse and fluc-
tuated greatly. In addition, the nsp1 IDR was also highly flexible.
Nsp1’s globular region and IDR do not stably interact with each other
Next, we investigated the intra-residue contacts within nsp1 with and without SL1. Fig 6
shows the contact map between nsp1 residues. The result shows that nsp1’s globular region
and IDR did not have stable contacts regardless of the presence of the SL1. We further ana-
lyzed the difference between two contact maps to investigate the specific changes in the struc-
tures (Fig 6 right). Overall, the difference in contacts was small, and thus nsp1 alone may not
experience significant structural changes with and without SL1, which is consistent to the sec-
ondary structure analysis. The largest difference in the contact ratio appeared between residues
Glu65 and Tyr68, which are located in the loop between α2 helix and β3 helix. However, the
ratio of the contacts between the loop on residues 64–68 and SL1 was low in the inter-contact
Fig 4. Graphical representation of the hydrogen bond interactions between SL1 and nsp1. Bases of U17 to C20
(colored blue) are recognized by the hydrogen bonds.
https://doi.org/10.1371/journal.pcbi.1009804.g004
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 10 / 21
analysis (Fig 3), suggesting that the change in the loop structure is caused indirectly. Because
there are also contact ratio changes at Gly30–Glu65 and Gly30–Gln66, and Gly30 is located
next to α1 helix, it is possible that the contact of α1 to SL1 shifted α1 and led to Glu65–Tyr68
contact difference.
Fig 5. Binding surface of nsp1. (A) Surface electrostatic potential of the nsp1 and (B) annotated surface structure of
the nsp1 recognition sites for SL1. In (A), units are in k
B
T/e, where k
B
is the Boltzmann factor, Tis the temperature of
the system (= 300 K), and eis the unit charge of a proton. Color coding in (B) corresponds to the region defined in Fig
1D.
https://doi.org/10.1371/journal.pcbi.1009804.g005
Fig 6. Interactions within nsp1. Contact probabilities between nsp1 residues were color-coded. Regions corresponding to residues 14–125, which are visible in the X-
ray crystallographic structure (PDB ID 2HSX), are shown as arrows on the right and the top of each image. (Left, middle) the probabilities for nsp1 with and without
SL1. (Right) the difference of the two (with SL1 minus without SL1). Residue pairs referenced in the main text are annotated by black and red wedges (pointing residue
pairs 65–68 and 30–65, respectively).
https://doi.org/10.1371/journal.pcbi.1009804.g006
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 11 / 21
Nsp1’s IDR may be aligned with SL1
Although there were no stable contacts between nsp1 globular region and IDR, the radius
of gyration R
g
of the nsp1 without SL1 was smaller compared to that without SL1 (Fig 7A),
suggesting that nsp1 alone is more compact compared to nsp1 with SL1. This result raises
another question: why nsp1 is elongated under the presence of SL1 while having no interaction
between nsp1 globular and IDR regions? We hypothesized that nsp1 is extended alongside the
stable stem loop structure. We analyzed the angle between SL1 and nsp1 region (iv) (Fig 7B).
The result indicated that the angle between the two was more likely to be <90 deg, i.e., two
axes were weakly aligned. As a result, the radius of the gyration of nsp1 region (iv) with SL1
was also larger than that without SL1 (Fig 7C).
SL1’s binding position in 40Sribosome-nsp1 complex
We further investigated the consistency between the known structure and the SL1’s binding
preference. For that purpose, we constructed the model of the 40Sribosome-nsp1 complex.
Fig 8A shows the overall structure of the 40Sribosome-nsp1 complex and Fig 8B presents the
closeup view around nsp1. The “valley” of nsp1 was close to the nsp1–S3 binding interface,
albeit open to the solvent. Thus, SL1 has enough space for binding even in the presence of the
40Sribosome. These results suggest that SL1 may form the trimer complex with the ribosome
and the nsp1.
We note that in addition to the reported interactions between nsp1 and S3, the C-terminal
disordered region of ribosomal protein S10 is also in proximity to nsp1 and the putative bind-
ing site of SL1 in the complex structure. The result suggests that nsp1 and/or SL1 may have
interactions with the disordered C-terminal tail of S10.
Clustering analysis of the binding poses
The diversity of the binding modes was further investigated using cluster analysis based on the
contact map for each snapshot (see Materials and methods). We determined the clustering
threshold using the criterion that any cluster has at least one inter-residue contacts with more
than 80% in each cluster. As a result, the binding modes could be categorized into 14 clusters
Fig 7. Nsp1 IDR and SL1 are partially aligned. (A) The radius of the gyration of nsp1 Cαatoms with or without SL1. Vertical lines represent the average. (B) The
distribution of angle between SL1 and nsp1 residues 121–146 (see Materials and methods for the definition). The theoretical angle distribution of two random vectors are
also presented for comparison. (C) The radius of the gyration of nsp1 C-terminalIDR. For (A)-(C), the ordinate represent the probability density, i.e. the ordinate are
scaled so that the area under the curve is exactly 1.
https://doi.org/10.1371/journal.pcbi.1009804.g007
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 12 / 21
and outliers, which had 34.2% of the statistical weight in the canonical ensemble. In even the
most major cluster, the statistical weight was only 15.5%; those for the second, third, and
fourth clusters were 9.9%, 7.4%, and 5.0%, respectively. Each cluster had a unique tendency to
use a set of binding regions (Text C and Fig G in S1 Text). We also analyzed the differences in
surface areas of the interacting interfaces in the ordered and disordered regions of nsp1
among the 14 clusters (Fig H in S1 Text). The distribution shows the unique characteristics of
each cluster. These results indicate that SL1 binds to nsp1 by multimodal binding modes.
The representative structure of cluster 1, which had the largest population among all clus-
ters, is presented in Fig 9 and Table C in S1 Text. Nsp1 recognized SL1 via regions (ii), (iii)
and (iv). In the region (ii), the basic residues in H2 formed the Arg43–C17 and Lys47–U16
salt-bridges. Region (iii) recognized SL1 via the Asp75–U18 hydrogen bond. Residues Arg124
through Gly137 in region (iv) attached to SL1 via the Arg124–U17, Ala131–C19, and Ser135–
C16 hydrogen bonds; Tyr136 stacks between C21 and G23 instead of A22, which was flipped
out. Representative structures of clusters 2 and 3 are also presented in the supporting material
(Text C, Figs I and J, and Tables C and D in S1 Text).
The stability of the obtained structures in the cluster was assessed with the conventional
MD simulation. Starting from 8 structures of clusters 1 and 2 (4 structures each), we per-
formed 500 ns MD simulations (4 μs in total) and analyzed whether the structure stably main-
tains the configuration found in the simulation. Fig K in S1 Text presents the nsp1-SL1
Fig 8. Mapping of the binding surface over the nsp1–40Sribosome complex. (A) Overview of the nsp1–40Sribosome complex
structure modeled from cryo-EM structure and its electron density map combined with the cross-linking mass spectrometry. (B) Close-
up view of the structures around nsp1. The N-terminal and C-terminal parts of nsp1 are colored blue and cyan. A hairpin of rRNA
(residue number 531 to 550) is colored orange, and the ribosomal proteins S3 and S10 are colored yellow and lime green, respectively. C-
terminal region of S10 after residue 97 (purple sphere) is considered to be the disordered region and is not visible in the structure.
Region corresponding to the “valley” of nsp1 binding surface is presented as a red transparent circle.
https://doi.org/10.1371/journal.pcbi.1009804.g008
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 13 / 21
contact map distances between the trajectory of conventional MD simulations and the cluster
center. All the four trajectories started from the most populated cluster (cluster 1) kept their
conformations during 500 ns simulations. The simulations started from the second most pop-
ulated cluster (cluster 2) were less stable. Two trajectories showed conformational changes
around 200 ns while the other two kept their conformations. Therefore, the structures found
in the cluster analysis, especially cluster 1, are considered stable for a reasonable time span.
Relation to other experimental results
It has been reported that the Arg124Ala–Lys125Ala double nsp1 mutant lacks the ability to
recognize viral RNA. [3,51] This can be explained by the results of our simulation, which
showed that sidechain of Arg124 strongly interacts with the phosphate backbone of U18
(Table 1 and Figs 4and 9). An Arg124Ala mutation would eliminate the ionic interaction
between the sidechain and the backbone, and nsp1 would lose its ability to recognize viral
RNA. Additionally, Arg124 and Lys125 are not contacting to the ribosome in the model struc-
ture, which is consistent to the fact that the UV cross-linking to 18S RNA was unaffected by
these two mutations. [15] On the other hand, recently reported Arg99Ala mutation to nsp1,
which also lacks the ability to recognize viral RNA, [51] did not match important hydrogen
bonds we found in the top clusters. This may be attributed to the insufficient sampling around
Arg99 (it is not included in the REST2 region) and/or lack of important binding partners in
the system, e.g. the ribosome.
The circular dichroism spectrum of the SARS-CoV-2 nsp1 C-terminal region (residues
130–180) [17] in solution had only a single peak at 198 nm and did not show ellipticities at 208
nm and 222 nm. This indicates that the nsp1 C-terminal region did not form α-helices or β-
sheets and was disordered. Similarly, in the analysis of NMR [52] spectra, nsp1 N- and C-ter-
mini are predicted to be fully unstructuted, but the predicted order parameters were different
among residues. Notably, relatively low order parameters were observed for residues 165–180
(corresponds to the α-helix region that shuts the ribosome), which is inconsistent with our
simulation result. Although in our simulation we found that nsp1 partially forms the α-helix
in the IDR, our simulation also showed that the percentage of the helix in the IDR was low
(<60%) and the structure was unstable, which may explain the difference from the experimen-
tal results; without SL1 the propensity was lowered further (Fig 2, and Fig F in S1 Text). Note
Fig 9. Interactions between nsp1 and SL1 observed in cluster 1. (A) Pairwise contact probability in cluster 1. See the legend to Fig 3. (B) Representative
snapshot of the cluster 1. The interface regions (i) through (v) are shown as green, cyan, magenta, red, and blue ribbons. Bases 16–26 of SL1 are shown in
orange.
https://doi.org/10.1371/journal.pcbi.1009804.g009
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 14 / 21
that these experiments were conducted without SL1. The propensity of the structure formation
may be affected in the presence of the highly charged molecules like RNA. Further study will
be needed before a conclusion can be drawn.
In X-ray crystallographic analysis of SARS-CoV-2 nsp1 N-terminal region, [50]β5-strand
of residues 95–97 only exits in SARS-CoV-2 nsp1 and not in SARS-CoV, despite the sequences
at residues 95–97 were unchanged. It was thus considered as a characteristic difference
between the two, although the site was near the crystal contact. In NMR, however, β5-strand
was not observed, [52] which was also corroborated by the order parameter and NOE analysis.
Our simulation data supports the latter, where residues 95–97 did not form β-ladder as shown
in Fig 2.
Whether SARS-CoV-2 nsp1 and SL1 bind without the ribosome is controversial. It has
been reported that nsp1 and bases 7–33 of SARS-CoV-2 bind with a binding constant of
0.18 μM [53], but it has also been reported that a gel shift does not occur with the 5’-UTR of
SARS-CoV-2 at concentrations up to 20 μM when tRNAs was used to exclude the non-specific
binding. [6] The present simulation results indicate that the binding mode observed herein
did not have a specific, defined structure. Typically, with such binding modes, the binding is
expected to be weak. Therefore, these simulation results do not contradict with the results
from either of the aforementioned experiments.
Mutations to SL1 bases 14–25, which disrupt the Watson-Crick pairs of the stem loop,
reportedly cause translation to be shut off. [6] That observation is consistent with our finding
that the hairpin structure of bases 18–22 in SL1 is recognized by nsp1. Hydrogen-bond inter-
action analysis showed that the RNA phosphate backbone is mainly recognized within the
C15-C20 region (Table 1 and Fig 4). Moreover, our finding is consistent with the fact that the
sequence of the hairpin region (corresponding to U18-C21 in our simulation) is not well con-
served among SARS-CoV-2 mutational variants, whereas that of the stem is well conserved.
[54] Our simulation shows that the interaction between nsp1 and the SL1 backbone is stronger
than that between nsp1 and the SL1 sidechains (Table 1), which highlights the importance of
the backbone interaction.
Limitations of this study
Our simulations were performed based on several assumptions. Here, we list the limitations of
the present study.
First, as we explained in “Convergence of the extended ensemble simulation”, even though
the current simulation uses the extended ensemble method, it is difficult to achieve full conver-
gence. Sampling RNA structures are generally considered difficult even with the small system
size, [55–59] and so does sampling the protein-RNA interaction. Given the length of IDR and
the size of RNAs, the convergence of the simulation may be beyond the capability of the cur-
rent computational resources. Current simulation results should thus be considered to achieve
only partial convergence at best, i.e., current structures may not be the fully determined most
stable structure under the current simulation force field, nor may it encounter enough transi-
tions to obtain unbiased samples. [60] Therefore, in this research, we avoided the quantitative
discussion of the energetics, which require complete convergence of the simulation; further-
more, the structures obtained in this research should be treated with caution.
Our simulations were performed without the ribosome. This was mainly because the simu-
lation started before the structure of the nsp1-ribosome complex as well as the cross-linking
experiment results were deposited. Furthermore, with the 40Sribosome, ribosomal proteins
S3 and S10 as well as rRNA hairpins at around residue 540 may interact with nsp1 or the SL1
as presented in Fig 8, which makes a proper sampling of the configurations difficult. With the
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 15 / 21
40Sribosome, the environment around nsp1 may be altered and so be the interaction between
the RNA and nsp1.
To maintain the stability of the hairpin loop structure, we performed the simulation with
restraints on the G-C pairs in the 5’-UTR. These restraints may have hindered RNA forming
structures other than the initial hairpin structure. However, in the secondary structure predic-
tion using CentroidFold [61] and the reference sequence, these base pairs were predicted to
exist in more than 92% of the ensemble. Furthermore, a recent study [59] showed that, even
with a rigorous extended ensemble simulation, the hairpin structure remained intact. Given
these results, the drawback of structural restraints to SL1 is expected to be minimal.
Finally, as is always the case with a simulation study, the mismatch between the simulation
force field and the real world leaves a non-negligible gap. For the simulation of IDR,
AMBER14SB used in this research may favor the folded state. [62] To overcome this problem,
several force fields specialized for IDR simulations have been proposed. [63,64] However,
IDR-oriented force fields are not suitable to simulate ordered regions in general, and are not
always better than conventional force fields even in IDR simulations. [65,66] In this study we
used AMBER14SB for proteins to balance the stability of both globular and disordered regions.
The result may depend on the force field used, e.g., the high propensity of the folded state on
the C-terminal region may be attributed to the property of the force field. Not only force fields
for proteins, but the choices for RNA force fields should also be considered, as each force field
has different characteristics upon reproducing RNA structures as well as protein-RNA com-
plexes [57,58,67]. The simulations with multiple different force fields will be almost necessary
to avoid drawing conclusions biased by a specific force-field. In addition to the force field
issues, some residues may have alternative protonation states upon binding to RNA (e.g., histi-
dine protonation state), which should be investigated further.
Conclusion
Future research and conclusions
The present simulation was performed with only nsp1 and SL1. Arguably, simulation of a
complex consisting of the 40Sribosome, nsp1 and SL1 will be an important step toward further
understanding the details of the mechanism underlying the evasion of nsp1 by viral RNA. Our
results suggest that the nsp1-SL1 complex without ribosome has multimodal binding struc-
tures. The addition of the 40Sribosome to the system may restrict the structure to a smaller
number of possible binding poses and possibly tighter binding poses may be obtained, while
the convergence of the simulation may be mitigated. However, as shown in Fig 8B, in addition
to the contacts between nsp1 and S3 and rRNA around residue 540, the C-terminal IDR of S10
may also interfere with nsp1, which may make sampling proper configurations more difficult.
Additionally, recent researches suggest possible caveats and remedies in the REST2 protocol;
[68,69] the combination of methodological advances and more refined models may enable us
to sample structures such that the stability of the complex can be discussed quantitatively. Fur-
ther researches will be necessary in this direction.
In addition to a simulation study, mutational analysis of nsp1 will be informative. In addi-
tion to the already known mutation at Arg124, current simulation results predict Lys47,
Arg43, and Asn126 are important to nsp1-SL1 bninding. Mutation analyses of these residues
will help us to understand the molecular mechanism of nsp1.
Finally, the development of inhibitors of nsp1-stem loop binding, is highly anticipated in
the current pandemic. Although the present results imply that a specific binding structure
might not exist, important residues in nsp1 and bases in SL1 were detected. Blocking or mim-
icking the binding of these residues/bases, could potentially nullify the function of nsp1.
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 16 / 21
In conclusion, using MD simulation, we investigated the binding and molecular mecha-
nism of SARS-CoV-2 nsp1 and the 5’-UTR stem loop of SARS-CoV-2 RNA. The results sug-
gest that the 5’-UTR stem loop of SARS-CoV-2 has the preference of binding onto regions
spanned from α1 helix to the disordered region. Upon the binding, the disordered region may
extend along the stem loop. The interaction analysis further suggested that the hairpin loop
structure of the 5’-UTR stem loop binds to the N-terminal domain and the intrinsically disor-
dered region of nsp1. Combined with the modeling, in the presence of the ribosome, the 5’-
UTR stem loop may bind to the interface of nsp1 and ribosomal protein S3, and ribosomal
protein S10 may also be involved in recognition of the 5’-UTR stem loop. Multiple binding
poses of nsp1 and the stem loop were obtained, and the largest cluster of the binding poses
included interactions that can explain the results of the cryo-EM, the cross-linking experi-
ments, and the previous mutational analyses.
Supporting information
S1 Text. Supporting information document. Text A: Convergence of the simulations. Text B:
Characterstics of clusters. Text C: Details of clusters 2 and 3. Fig A: Survey for the clustering
parameters. Fig B: Convergence of the secondary structure distribution. Fig C: Convergence
of the hydrogen bond forming ratio. Fig D: Timecourse of the replica indices. Fig E: Distances
to clusters in canonical simulations starting from the initial configuration. Fig F: Secondary
structure distribution of nsp1 without SL1. Fig G: Representative structure of each cluster. Fig
H: Surface area of interaction intefaces of nsp1. Fig I: Interactions between nsp1 and SL1 in
cluster 2. Fig J: Interactions between nsp1 and SL1 in cluster 3. Fig K: Distances to clusters in
canonical simulations starting from cluster 1 structures. Table A: Characteristics of the nsp1–
40S ribosome complex models. Table B: Characteristics of SL1 binding regions of nsp1.
Table C: Characteristics of each conformational cluster.
(PDF)
S1 File. RNA force field file used in this work.
(ZIP)
S2 File. Patches applied to GROMACS 2016 used in this work.
(ZIP)
Acknowledgments
We thank Dr. Atsushi Matsumoto for his technical assistance. Simulations were performed on
supercomputers at Research Center for Computational Science, Okazaki, and Academic Cen-
ter for Computing and Media Studies, Kyoto University.
Author Contributions
Conceptualization: Shun Sakuraba, Hidetoshi Kono.
Formal analysis: Shun Sakuraba, Qilin Xie, Kota Kasahara, Junichi Iwakiri.
Funding acquisition: Shun Sakuraba, Kota Kasahara, Junichi Iwakiri, Hidetoshi Kono.
Methodology: Shun Sakuraba, Kota Kasahara, Junichi Iwakiri.
Project administration: Shun Sakuraba.
Resources: Hidetoshi Kono.
Software: Shun Sakuraba.
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 17 / 21
Supervision: Kota Kasahara, Hidetoshi Kono.
Visualization: Shun Sakuraba, Kota Kasahara.
Writing – original draft: Shun Sakuraba, Qilin Xie, Kota Kasahara.
Writing – review & editing: Shun Sakuraba, Kota Kasahara, Junichi Iwakiri, Hidetoshi Kono.
References
1. Narayanan K, Huang C, Lokugamage K, Kamitani W, Ikegami T, Tseng CTK, et al. Severe Acute Respi-
ratory Syndrome Coronavirus nsp1 Suppresses Host Gene Expression, Including That of Type I Inter-
feron, in Infected Cells. Journal of Virology. 2008; 82(9):4471–4479. https://doi.org/10.1128/JVI.02472-
07 PMID: 18305050
2. Kamitani W, Huang C, Narayanan K, Lokugamage KG, Makino S. A two-pronged strategy to suppress
host protein synthesis by SARS coronavirus Nsp1 protein. Nature Structural & Molecular Biology. 2009;
16(11):1134–1140. https://doi.org/10.1038/nsmb.1680 PMID: 19838190
3. Lokugamage KG, Narayanan K, Huang C, Makino S. Severe Acute Respiratory Syndrome Coronavirus
Protein nsp1 Is a Novel Eukaryotic Translation Inhibitor That Represses Multiple Steps of Translation
Initiation. Journal of Virology. 2012; 86(24):13598–13608. https://doi.org/10.1128/JVI.01958-12 PMID:
23035226
4. Tanaka T, Kamitani W, DeDiego ML, Enjuanes L, Matsuura Y. Severe Acute Respiratory Syndrome
Coronavirus nsp1 Facilitates Efficient Propagation in Cells through a Specific Translational Shutoff of
Host mRNA. Journal of Virology. 2012; 86(20):11128–11137. https://doi.org/10.1128/JVI.01700-12
PMID: 22855488
5. Narayanan K, Ramirez SI, Lokugamage KG, Makino S. Coronavirus nonstructural protein 1: Common
and distinct functions in the regulation of host and viral gene expression. Virus Research. 2015;
202:89–100. https://doi.org/10.1016/j.virusres.2014.11.019 PMID: 25432065
6. Tidu A, Janvier A, Schaeffer L, Sosnowski P, Kuhn L, Hammann P, et al. The viral protein NSP1 acts as
a ribosome gatekeeper for shutting down host translation and fostering SARS-CoV-2 translation. RNA.
2020. https://doi.org/10.1261/rna.078121.120 PMID: 33268501
7. Kamitani W, Narayanan K, Huang C, Lokugamage K, Ikegami T, Ito N, et al. Severe acute respiratory
syndrome coronavirus nsp1 protein suppresses host gene expression by promoting host mRNA degra-
dation. Proceedings of the National Academy of Sciences. 2006; 103(34):12885–12890. https://doi.org/
10.1073/pnas.0603144103 PMID: 16912115
8. Huang C, Lokugamage KG, Rozovics JM, Narayanan K, Semler BL, Makino S. SARS Coronavirus
nsp1 Protein Induces Template-Dependent Endonucleolytic Cleavage of mRNAs: Viral mRNAs Are
Resistant to nsp1-Induced RNA Cleavage. PLoS Pathogens. 2011; 7(12):e1002433. https://doi.org/10.
1371/journal.ppat.1002433 PMID: 22174690
9. Finkel Y, Gluck A, Nachshon A, Winkler R, Fisher T, Rozman B, et al. SARS-CoV-2 uses a multi-
pronged strategy to impede host protein synthesis. Nature. 2021; 594(7862):240–245. https://doi.org/
10.1038/s41586-021-03610-3 PMID: 33979833
10. Wathelet MG, Orr M, Frieman MB, Baric RS. Severe Acute Respiratory Syndrome Coronavirus Evades
Antiviral Signaling: Role of nsp1 and Rational Design of an Attenuated Strain. Journal of Virology. 2007;
81(21):11620–11633. https://doi.org/10.1128/JVI.00702-07 PMID: 17715225
11. Thoms M, Buschauer R, Ameismeier M, Koepke L, Denk T, Hirschenberger M, et al. Structural basis for
translational shutdown and immune evasion by the Nsp1 protein of SARS-CoV-2. Science. 2020; 369
(6508):1249–1255. https://doi.org/10.1126/science.abc8665 PMID: 32680882
12. Schubert K, Karousis ED, Jomaa A, Scaiola A, Echeverria B, Gurzeler LA, et al. SARS-CoV-2 Nsp1
binds the ribosomal mRNA channel to inhibit translation. Nature Structural & Molecular Biology. 2020;
27(10):959–966. https://doi.org/10.1038/s41594-020-0511-8
13. Yuan S, Peng L, Park JJ, Hu Y, Devarkar SC, Dong MB, et al. Nonstructural Protein 1 of SARS-CoV-2
Is a Potent Pathogenicity Factor Redirecting Host Protein Synthesis Machinery toward Viral RNA.
Molecular Cell. 2020; 80(6):1055–1066.e6. https://doi.org/10.1016/j.molcel.2020.10.034 PMID:
33188728
14. Kim D, Lee JY, Yang JS, Kim JW, Kim VN, Chang H. The Architecture of SARS-CoV-2 Transcriptome.
Cell. 2020; 181(4):914–921.e10. https://doi.org/10.1016/j.cell.2020.04.011 PMID: 32330414
15. Banerjee AK, Blanco MR, Bruce EA, Honson DD, Chen LM, Chow A, et al. SARS-CoV-2 Disrupts Splic-
ing, Translation, and Protein Trafficking to Suppress Host Defenses. Cell. 2020; 183(5):1325–1339.
e21. https://doi.org/10.1016/j.cell.2020.10.004 PMID: 33080218
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 18 / 21
16. Almeida MS, Johnson MA, Herrmann T, Geralt M, Wu¨thrich K. Novel β-Barrel Fold in the Nuclear Mag-
netic Resonance Structure of the Replicase Nonstructural Protein 1 from the Severe Acute Respiratory
Syndrome Coronavirus. Journal of Virology. 2007; 81(7):3151–3161. https://doi.org/10.1128/JVI.
01939-06 PMID: 17202208
17. Kumar A, Kumar A, Kumar P, Garg N, Giri R. SARS-CoV-2 NSP1 C-terminal region (residues 130-180)
is an intrinsically disordered region. bioRxiv. 2020.
18. Fiser A, S
ˇali A. Modeller: Generation and Refinement of Homology-Based Protein Structure Models. In:
Methods in Enzymology. Elsevier; 2003. p. 461–491.
19. Cornell WD, Cieplak P, Bayly CI, Gould IR, Merz KM, Ferguson DM, et al. A second generation force
field for the simulation of proteins, nucleic acids, and organic molecules. Journal of the American Chem-
ical Society. 1995; 117(19):5179–5197. https://doi.org/10.1021/ja00124a002
20. Wang J, Cieplak P, Kollman PA. How well does a restrained electrostatic potential (RESP) model per-
form in calculating conformational energies of organic and biological molecules? Journal of Computa-
tional Chemistry. 2000; 21(12):1049–1074. https://doi.org/10.1002/1096-987X(200009)21:12%
3C1049::AID-JCC3%3E3.0.CO;2-F
21. Hornak V, Abel R, Okur A, Strockbine B, Roitberg A, Simmerling C. Comparison of multiple Amber
force fields and development of improved protein backbone parameters. Proteins: Structure, Function,
and Bioinformatics. 2006; 65(3):712–725. https://doi.org/10.1002/prot.21123 PMID: 16981200
22. Maier JA, Martinez C, Kasavajhala K, Wickstrom L, Hauser KE, Simmerling C. ff14SB: Improving the
Accuracy of Protein Side Chain and Backbone Parameters from ff99SB. Journal of Chemical Theory
and Computation. 2015; 11(8):3696–3713. https://doi.org/10.1021/acs.jctc.5b00255 PMID: 26574453
23. Popenda M, Szachniuk M, Antczak M, Purzycka KJ, Lukasiak P, Bartol N, et al. Automated 3D structure
composition for large RNAs. Nucleic Acids Research. 2012; 40(14):e112–e112. https://doi.org/10.1093/
nar/gks339 PMID: 22539264
24. Antczak M, Popenda M, Zok T, Sarzynska J, Ratajczak T, Tomczyk K, et al. New functionality of RNA-
Composer: application to shape the axis of miR160 precursor structure. Acta Biochimica Polonica.
2017; 63(4). https://doi.org/10.18388/abp.2016_1329
25. Wu F, Zhao S, Yu B, Chen YM, Wang W, Song ZG, et al. A new coronavirus associated with human
respiratory disease in China. Nature. 2020; 579(7798):265–269. https://doi.org/10.1038/s41586-020-
2008-3 PMID: 32015508
26. Bayly CI, Cieplak P, Cornell W, Kollman PA. A well-behaved electrostatic potential based method using
charge restraints for deriving atomic charges: the RESP model. J Phys Chem. 1993; 97(40):10269–
10280. https://doi.org/10.1021/j100142a004
27. Perez A, Marchan I, Svozil D, Sponer J, Cheatham TE, Laughton CA, et al. Refinement of the AMBER
force field for nucleic acids: improving the description of alpha/gamma conformers. Biophys J. 2007; 92
(11):3817–3829. https://doi.org/10.1529/biophysj.106.097782 PMID: 17351000
28. Zgarbova
´M, Otyepka M, S
ˇponer J, Mladek A, Banas P, Cheatham TE, et al. Refinement of the Cornell
et al. Nucleic Acids Force Field Based on Reference Quantum Chemical Calculations of Glycosidic Tor-
sion Profiles. J Chem Theory Comput. 2011; 7(9):2886–2902. https://doi.org/10.1021/ct200162x PMID:
21921995
29. da Silva AWS, Vranken WF. ACPYPE—AnteChamber PYthon Parser interfacE. BMC Research Notes.
2012; 5(1):367. https://doi.org/10.1186/1756-0500-5-367
30. Case DA, Cerutti DS, Cheatham TE III, Darden TA, Duke RE, Giese TJ, et al. AMBER 2017; 2017. Uni-
versity of California, San Francisco.
31. Jorgensen WL, Chandrasekhar J, Madura JD, Impey RW, Klein ML. Comparison of simple potential
functions for simulating liquid water. The Journal of Chemical Physics. 1983; 79(2):926–935. https://doi.
org/10.1063/1.445869
32. Joung IS, Cheatham TE. Determination of alkali and halide monovalent ion parameters for use in explic-
itly solvated biomolecular simulations. J Phys Chem B. 2008; 112(30):9020–9041. https://doi.org/10.
1021/jp8001614 PMID: 18593145
33. Ikebe J, Sakuraba S, Kono H. H3 histone tail conformation within the nucleosome and the impact of
K14 acetylation studied using enhanced sampling simulation. PLoS computational biology. 2016; 12(3):
e1004788. https://doi.org/10.1371/journal.pcbi.1004788 PMID: 26967163
34. Li Z, Kono H. Investigating the Influence of Arginine Dimethylation on Nucleosome Dynamics Using All-
Atom Simulations and Kinetic Analysis. The Journal of Physical Chemistry B. 2018; 122(42):9625–
9634. https://doi.org/10.1021/acs.jpcb.8b05067 PMID: 30256111
35. Kasahara K, Shiina M, Higo J, Ogata K, Nakamura H. Phosphorylation of an intrinsically disordered
region of Ets1 shifts a multi-modal interaction ensemble to an auto-inhibitory state. Nucleic acids
research. 2018; 46(5):2243–2251. https://doi.org/10.1093/nar/gkx1297 PMID: 29309620
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 19 / 21
36. Wang L, Friesner RA, Berne BJ. Replica Exchange with Solute Scaling: A More Efficient Version of
Replica Exchange with Solute Tempering (REST2). The Journal of Physical Chemistry B. 2011; 115
(30):9431–9438. https://doi.org/10.1021/jp204407d PMID: 21714551
37. Abraham MJ, Murtola T, Schulz R, Pa
´ll S, Smith JC, Hess B, et al. GROMACS: High performance
molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX.
2015; 1:19–25. https://doi.org/10.1016/j.softx.2015.06.001
38. Bussi G. Hamiltonian replica exchange in GROMACS: a flexible implementation. Molecular Physics.
2013; 112(3-4):379–384. https://doi.org/10.1080/00268976.2013.824126
39. Bussi G, Donadio D, Parrinello M. Canonical sampling through velocity rescaling. The Journal of Chemi-
cal Physics. 2007; 126(1):014101. https://doi.org/10.1063/1.2408420 PMID: 17212484
40. Hess B, Bekker H, Berendsen HJC, Fraaije JGEM. LINCS: A linear constraint solver for molecular simu-
lations. Journal of Computational Chemistry. 1997; 18(12):1463–1472. https://doi.org/10.1002/(SICI)
1096-987X(199709)18:12%3C1463::AID-JCC4%3E3.0.CO;2-H
41. Souaille M, Roux B. Extension to the weighted histogram analysis method: combining umbrella sam-
pling with free energy calculations. Computer Physics Communications. 2001; 135(1):40–57. https://
doi.org/10.1016/S0010-4655(00)00215-0
42. Shirts MR, Chodera JD. Statistically optimal analysis of samples from multiple equilibrium states. The
Journal of Chemical Physics. 2008; 129(12):124105. https://doi.org/10.1063/1.2978177 PMID: 19045004
43. Humphrey W, Dalke A, Schulten K. VMD—Visual Molecular Dynamics. Journal of Molecular Graphics.
1996; 14:33–38. https://doi.org/10.1016/0263-7855(96)00018-5 PMID: 8744570
44. Schro
¨dinger, LLC. The PyMOL Molecular Graphics System, Version 1.8; 2015.
45. Kabsch W, Sander C. Dictionary of protein secondary structure: Pattern recognition of hydrogen-
bonded and geometrical features. Biopolymers. 1983; 22(12):2577–2637. https://doi.org/10.1002/bip.
360221211 PMID: 6667333
46. McGibbon RT, Beauchamp KA, Harrigan MP, Klein C, Swails JM, Herna
´ndez CX, et al. MDTraj: A Mod-
ern Open Library for the Analysis of Molecular Dynamics Trajectories. Biophysical Journal. 2015; 109
(8):1528–1532. https://doi.org/10.1016/j.bpj.2015.08.015 PMID: 26488642
47. Ester M, Kriegel HP, Sander J, Xu X, et al. A density-based algorithm for discovering clusters in large
spatial databases with noise. In: KDD. vol. 96; 1996. p. 226–231.
48. Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, Meng EC, et al. UCSF Chimera–a
visualization system for exploratory research and analysis. Journal of computational chemistry. 2004;
25:1605–1612. https://doi.org/10.1002/jcc.20084 PMID: 15264254
49. Slavin M, Zamel J, Zohar K, Eliyahu T, Braitbard M, Brielle E, et al. Targeted in situ cross-linking mass
spectrometry and integrative modeling reveal the architectures of three proteins from SARS-CoV-2.
Proceedings of the National Academy of Sciences of the United States of America. 2021; 118(34):
e2103554118. https://doi.org/10.1073/pnas.2103554118 PMID: 34373319
50. Semper C, Watanabe N, Savchenko A. Structural characterization of nonstructural protein 1 from
SARS-CoV-2. iScience. 2021; 24(1):101903. https://doi.org/10.1016/j.isci.2020.101903 PMID:
33319167
51. Mendez AS, Ly M, Gonza
´lez-Sa
´nchez AM, Hartenian E, Ingolia NT, Cate JH, et al. The N-terminal
domain of SARS-CoV-2 nsp1 plays key roles in suppression of cellular gene expression and preserva-
tion of viral gene expression. Cell Reports. 2021; 37(3):109841. https://doi.org/10.1016/j.celrep.2021.
109841 PMID: 34624207
52. Agback T, Dominguez F, Frolov I, Frolova EI, Agback P. 1H, 13C and 15N resonance assignment of
the SARS-CoV-2 full-length nsp1 protein and its mutants reveals its unique secondary structure fea-
tures in solution. bioRxiv. 2021; p. 2021.05.05.442725.
53. Vankadari N, Jeyasankar NN, Lopes WJ. Structure of the SARS-CoV-2 Nsp1/50-Untranslated Region
Complex and Implications for Potential Therapeutic Targets, a Vaccine, and Virulence. The Journal of
Physical Chemistry Letters. 2020; 11(22):9659–9668. https://doi.org/10.1021/acs.jpclett.0c02818
PMID: 33135884
54. Miao Z, Tidu A, Eriani G, Martin F. Secondary structure of the SARS-CoV-2 5’-UTR. RNA Biology.
2020; p. 1–10. https://doi.org/10.1080/15476286.2020.1814556 PMID: 32965173
55. Bergonzo C, Henriksen NM, Roe DR, Swails JM, Roitberg AE, Cheatham TE. Multidimensional Replica
Exchange Molecular Dynamics Yields a Converged Ensemble of an RNA Tetranucleotide. Journal of
Chemical Theory and Computation. 2014; 10(1):492–499. https://doi.org/10.1021/ct400862k PMID:
24453949
56. Bergonzo C, Henriksen NM, Roe DR, Cheatham TE. Highly sampled tetranucleotide and tetraloop
motifs enable evaluation of common RNA force fields. RNA. 2015; 21(9):1578–1590. https://doi.org/10.
1261/rna.051102.115 PMID: 26124199
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 20 / 21
57. Tan D, Piana S, Dirks RM, Shaw DE. RNA force field with accuracy comparable to state-of-the-art pro-
tein force fields. Proceedings of the National Academy of Sciences of the United States of America.
2018; 115(7):E1346–E1355. https://doi.org/10.1073/pnas.1713027115 PMID: 29378935
58. Ku¨hrova
´P, Mlynsky V, Zgarbova M, Krepl M, Bussi G, Best RB, et al. Improving the Performance of the
Amber RNA Force Field by Tuning the Hydrogen-Bonding Interactions. Journal of Chemical Theory and
Computation. 2019; 15(5):3288–3305. https://doi.org/10.1021/acs.jctc.8b00955 PMID: 30896943
59. Bottaro S, Bussi G, Lindorff-Larsen K. Conformational Ensembles of Non-Coding Elements in the
SARS-CoV-2 Genome from Molecular Dynamics Simulations. bioRxiv. 2020; p. 2020.12.11.421784.
60. Zuckerman DM. Equilibrium Sampling in Biomolecular Simulations. Annual Review of Biophysics.
2011; 40(1):41–62. https://doi.org/10.1146/annurev-biophys-042910-155255 PMID: 21370970
61. Sato K, Hamada M, Asai K, Mituyama T. CENTROIDFOLD: a web server for RNA secondary structure
prediction. Nucleic Acids Research. 2009; 37(Web Server):W277–W280. https://doi.org/10.1093/nar/
gkp367 PMID: 19435882
62. Song D, Luo R, Chen HF. The IDP-Specific Force Field ff14IDPSFF Improves the Conformer Sampling
of Intrinsically Disordered Proteins. Journal of Chemical Information and Modeling. 2017; 57(5):1166–
1178. https://doi.org/10.1021/acs.jcim.7b00135 PMID: 28448138
63. Kasahara K, Terazawa H, Takahashi T, Higo J. Studies on Molecular Dynamics of Intrinsically Disor-
dered Proteins and Their Fuzzy Complexes: A Mini-Review. Computational and Structural Biotechnol-
ogy Journal. 2019; 17:712–720. https://doi.org/10.1016/j.csbj.2019.06.009 PMID: 31303975
64. Mu J, Liu H, Zhang J, Luo R, Chen HF. Recent Force Field Strategies for Intrinsically Disordered Pro-
teins. Journal of Chemical Information and Modeling. 2021; 61(3):1037–1047. https://doi.org/10.1021/
acs.jcim.0c01175 PMID: 33591749
65. Rauscher S, Gapsys V, Gajda MJ, Zweckstetter M, de Groot BL, Grubmu¨ller H. Structural Ensembles
of Intrinsically Disordered Proteins Depend Strongly on Force Field: A Comparison to Experiment. Jour-
nal of Chemical Theory and Computation. 2015; 11(11):5513–5524. https://doi.org/10.1021/acs.jctc.
5b00736 PMID: 26574339
66. Robustelli P, Piana S, Shaw DE. Developing a molecular dynamics force field for both folded and disor-
dered protein states. Proceedings of the National Academy of Sciences. 2018; 115(21):E4758–E4766.
https://doi.org/10.1073/pnas.1800690115 PMID: 29735687
67. S
ˇponer J, Bussi G, Krepl M, Bana
´s
ˇP, Bottaro S, Cunha RA, et al. RNA Structural Dynamics As Cap-
tured by Molecular Simulations: A Comprehensive Overview. Chemical Reviews. 2018; 118(8):4177–
4338. https://doi.org/10.1021/acs.chemrev.7b00427 PMID: 29297679
68. Kamiya M, Sugita Y. Flexible selection of the solute region in replica exchange with solute tempering:
Application to protein-folding simulations. The Journal of Chemical Physics. 2018; 149(7):072304.
https://doi.org/10.1063/1.5016222 PMID: 30134668
69. Appadurai R, Nagesh J, Srivastava A. High resolution ensemble description of metamorphic and intrin-
sically disordered proteins using an efficient hybrid parallel tempering scheme. Nature Communica-
tions. 2021; 12(1). https://doi.org/10.1038/s41467-021-21105-7 PMID: 33574233
PLOS COMPUTATIONAL BIOLOGY
Extended ensemble simulations of a SARS-CoV-2 nsp1–5’-UTR complex
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009804 January 19, 2022 21 / 21
Available via license: CC BY 4.0
Content may be subject to copyright.
