Dynamic rotation of the protruding domain enhances the infectivity of norovirus

Norovirus is the major cause of epidemic nonbacterial gastroenteritis worldwide. Lack of structural information on infection and replication mechanisms hampers the development of effective vaccines and remedies. Here, using cryo-electron microscopy, we show that the capsid structure of murine noroviruses changes in response to aqueous conditions. By twisting the flexible hinge connecting two domains, the protruding (P) domain reversibly rises off the shell (S) domain in solutions of higher pH, but rests on the S domain in solutions of lower pH. Metal ions help to stabilize the resting conformation in this process. Furthermore, in the resting conformation, the cellular receptor CD300lf is readily accessible, and thus infection efficiency is significantly enhanced. Two similar P domain conformations were also found simultaneously in the human norovirus GII.3 capsid, although the mechanism of the conformational change is not yet clear. These results provide new insights into the mechanisms of non-enveloped norovirus transmission that invades host cells, replicates, and sometimes escapes the hosts immune system, through dramatic environmental changes in the gastrointestinal tract.


Introduction
Human norovirus (HNoV) is a major cause of epidemic nonbacterial gastroenteritis worldwide [1]. HNoV often causes severe sporadic infections, especially in nurseries and nursing homes. However, there are no efficient treatments or vaccines due to lack of a robust cultivation system and model animals for studying this virus [2]. Murine norovirus (MNoV), a species of norovirus affecting mice, was identified in 2003 [3]. MNoV is a unique norovirus which propagates in cell lines [3,4], and shares genetic features with HNoV and has biochemically similar properties [5]. For these reasons, MNoV has been widely used as a model for HNoV infection.
Norovirus (NoV) is a non-enveloped, positive-stranded RNA virus belonging to the Caliciviridae family [6]. The common capsid structure of 180 copies of the VP1 capsid protein forming a T = 3 icosahedral particle is shared by all identified caliciviruses, which was first reported with HNoV GI.1 virus-like particle (VLP) produced by a baculovirus expression system in 1999 [7]. To form the capsid structure, VP1 is placed in three quasi-equivalent positions of asymmetric units designated as A, B, and C monomers. The VP1's A and B dimers (A/B dimers) are located around the icosahedral fivefold axes, while the VP1's C and C dimers (C/C dimers) are placed on the icosahedral twofold axes [8]. The VP1 protein consists of a shell (S) and protruding (P) domain. The S domain, consisting of an eight-stranded β sandwich structure, forms a contiguous icosahedral backbone to protect the central viral genome and has high amino acid sequence homology within the Caliciviridae family [9]. On the other hand, the P domain extended from the S domain via a flexible hinge loop further divided into a lower P1 subdomain and an upper P2 subdomain. With respect to the amino acid sequence, the P1 subdomain can be divided into two parts, the N-terminal P1 part (P1-1) and the C-terminal P1 part (P1-2), with the P2 subdomain interposed. The P2 subdomain is exposed on the top of the P domain and functions for virus attachment to cells [10].
The P domain in caliciviruses shows two conformations [11]. The first, called here the rising conformation, is shown in MNoV-1 [12], HNoV GII.10 [13], and rabbit hemorrhagic disease virus (RHDV) [14,15], where the P domain rises from the S domain surface, forming an outer second shell. The second, called here the resting conformation, is shown in HNoV GI.1 [7], sapovirus [9], San Miguel sea lion virus (SMSV) [16] and feline calicivirus (FCV) [17,18], where the P domain rests upon the S domain, forming an integrated shell with the S domain. The potential for dynamic structural changes in capsids has been discussed for viral infection and replication, but no direct evidence has been found thus far [11]. During the preparation of this manuscript, Snowden and colleagues also reported the resting conformation of MNoV-1 infectious particles in several samples, including heat treatment, suggesting a potential for the conformational changes in a single virus [19], although the mechanism has not yet been elucidated.
Here, using cryo-electron microscopy (cryo-EM), we show that the P domain of MNoV infectious particles reversibly rotates~70˚, in response to aqueous conditions, taking two different conformations; the rising and resting P domain conformations. The P domain extends away from the S domain surface in solutions with higher pH, and rests on the S domain surface in solutions with lower pH. Metal ions help to stabilize the resting conformation in this process. Furthermore, significant differences were found in the two P domain conformations with respect to MNoV infection of cultured cells. High-resolution cryo-EM structural analysis using MNoV-VLP revealed the structural similarity between infectious particle and VLP in MNoV and elucidates the molecular mechanism of the P domain rotation. Our findings provide new insights into the mechanisms of viral infection of the non-enveloped viruses.

Dynamic rotation of the protruding domain in MNoV controls viral infection
For our structural studies, infectious particles of MNoV type 1 (MNoV-1) were produced by a reverse genetics system [20], propagated in RAW264.7 cells, and stored in DMEM (Dulbecco's Modified Eagle Medium). Viral particles were subjected to analysis by single-particle cryo-EM using a 200kV transmission electron microscope (TEM), and a 3D model was reconstructed at 5.3 Å resolution, in which the P domain was stabilized with an outer P2 subdomain-based interaction and rested on the S-domain (Fig 1A and 1C, S1A-S1D Fig). Interestingly, our results differed from the previous report of the 8 Å cryo-EM structure of the MNoV-1 suspended in PBS [14], in which the P domains rotated~70˚clockwise were mutually stabilized by interactions based on the inner P1 subdomain, and extended away from the S domain. To investigate these structural inconsistencies, we suspended the infectious particles in various aqueous solutions and examined them by cryo-EM. Finally, we were able to reproduce the reported MNoV-1 capsid structure at 7.3 Å resolution when the viral particles were suspended in a PBS(-) solution containing 20 mM EDTA at pH 8.0 (PBS(-)+EDTA (pH 8.0)) ( Fig 1B, 1D and 1E, S1E-S1H Fig). The previously reported structure [14] of MNoV-1 was found to appear when pH was above 7 and metal ions were removed from solution by chelation with EDTA ( Fig 1E). Furthermore, the dynamic rotation of the P domain of MNoV-1 infectious particles occurred reversibly in response to aqueous conditions, where the rotation from the rising to resting conformation was simply triggered at a pH lower than 7 ( Fig 1E).
Next, we sought to elucidate the physiological functions of the two P domain conformations in terms of infectivity of the viral particles. When the MNoV-1 infectious particles pretreated with PBS(-)+EDTA (pH 8) were infected to RAW264.7 cells cultured in DMEM (pH 7.2-7.4) (see Materials and Methods), they showed significantly more propagation delay (~3 hours) than those pretreated with DMEM ( Fig 1F). To investigate the reason for the propagation delay, we first compared the virus attachment on the host cell surface at 30 minutes after infection using quantitative real-time RT-PCR (qRT-PCR) (S2 Fig). Binding of the MNoV-1 particles pretreated with PBS(-)+EDTA (pH 8) to RAW264.7 cells, however, showed no significant difference from those pretreated with DMEM (Left bars in S2A Fig). RAW264.7 cells are known to express several molecules involved in MNoV adsorption [21]. In contrast, HEK293T cells genetically expressing CD300lf (HEK293T/CD300lf cells) can adsorb MNoV particles on the cell surface via direct interaction with the CD300lf [22,23]. Thus, HEK293T/CD300lf cells were used to evaluate MNoV adsorption depending on the two P domain conformations.

Resting and rising P domain conformations in HNoV GII.3 VLP
We also observed the two conformations of the P domain in the HNoV GII.3 (TCH04-577 strain) VLPs by cryo-EM (S3 Fig). Single particle analysis showed that the two P domain conformations were similar to the resting and rising P domain conformations of MNoV-1 ( Fig  1G-1J). In the HNoV GII.3 VLP suspended in DMEM, 16% of the total showed the rising P domain conformation, and the rest showed the resting P domain conformation. The ratio did not change even in PBS(-)+EDTA (pH 8.0).
The structure of the resting and rising P domain conformation was compared in more detail between MNoV-1 and HNoV GII.3 (S4 Fig). In the two structures, the angle of rotation between the resting and rising P domains was coincident (~70˚clockwise), and the interactions between P domain dimers were similarly observed at the P2 levels in the resting conformation and at the P1 levels in the rising conformation, respectively. However, the heights of the P domain are significantly different in the rising P domain conformations, where the P domain of HNoV GII.3 is~4 Å lower than MNoV-1. The length of the hinge loop between the S and P domains are the same in these NoVs (S5 Fig). It suggests that the holding manner in the hinge loop is different between MNoV and HNoV.

Transformation process of the dynamic rotation of P domain in MNoV-1
We identified using cryo-EM that the rising P domain conformation gradually changed to the resting conformation in 2 to 8 hours, when MNoV-1 particles pretreated with PBS (-)+EDTA (pH 8) were suspended in DMEM (pH 7.2-7.4) (Fig 2A). Interestingly, particles mixed with two conformations were also observed between 2 and 6 hours ( Fig 2B). The mixed P domain structure within a single particle suggests that even if the conversion of individual P domains may be rapid, it would take more time to modify the overall conformation of the capsid, where the P domains interact and are connected together like an entangled net, in which the P . The maps are low-pass filtered to 8 Å resolution to highlight the domain structure. The left and right panels are the isosurface representation and the center section, respectively. Black arrowheads indicate gaps between the S and P domains that appear in each of the resting and rising P domain conformations. (C and D) Left panels are the enlarged views of the boxes in A and B. Right panels are views from the direction of the dotted arrows shown in the left panel. The P domain of MNoV-1 suspended in DMEM rests on the S domain (black arrowheads in A and C) and interacts with the adjacent P domains at the outer P2 subdomain level (black arrows in C), called the resting P domain conformation. By contrast, the P domain of MNoV-1 suspended in PBS(-)+EDTA (pH 8) rises off the S domain (black arrowheads in B and D) and interacts with the adjacent P domains at the inner P1 subdomain level (black arrows in D), called the rising P domain conformation. In the rising P domain conformation, all P domain dimers rotate~70˚clockwise, compared to the resting P domain conformation, and the interaction sites are changed from the P2 to the P1 level.

PLOS PATHOGENS
Dynamic rotation of norovirus protruding domain domain connections stabilize a given conformation causing hysteresis. In our observation, the transformations between the resting and rising P domain conformations were accelerated in higher or lower pH conditions, where they were completed in less than 5 minutes at the fastest, respectively. Consequently, the delay time of the virus propagation pretreated with PBS(-) +EDTA (pH 8) possibly corresponds to the time of the overall P domain conformational change of the virus particles in the cell culture media (pH 7.2-7.4). These results suggest that the rising conformation of the P domain prevents the initial viral binding to the host cell surface via CD300lf, and it takes time to recover infectivity by changing the rising conformation to the resting conformation in the whole capsid. Hence, MNoV-1 can control viral infectivity by the dynamic rotation of the P domain.

Dynamic rotation of the P domain in MNoV-S7
The similar dynamic rotation of the P domain was also observed with MNoV type S7 (MNoV-S7) (S6 Fig). MNoV-S7 is a norovirus isolated from mouse stools in Japan in 2007 [24] and shows similar pathogenicity to MNoV-1. The amino acid sequence of MNoV-S7 VP1 protein shows 97% sequence identity with MNoV-1, with the nonhomologous 18 amino acids concentrated in the P domain (S5 Fig). Originally, we used this strain for structural study because we were already able to produce both the infectious particle and VLP [20]. However, since the P domain structure of the particle acquired was different from the previously reported cryo-EM map of MNoV-1 [14], MNoV-1 was used to confirm the structure in our sample preparation method. Interestingly, in the case of MNoV-S7, the rising P domain conformation only occurred in the solution of PBS(-)+EDTA at pH 8 or higher (S6K Fig), which is slightly higher than that of MNoV-1 (pH 7 or higher (Fig 1E)). The reversible rotation from Particles mixed with two conformations were observed (second row panels). The mixed P domain structure within a single particle suggests that even if the conversion of individual P domains may be rapid, it would take time to modify the overall conformation of the capsid in this aqueous condition, where the P domains are connected together like an entangled net. https://doi.org/10.1371/journal.ppat.1008619.g002

PLOS PATHOGENS
Dynamic rotation of norovirus protruding domain the rising to resting conformation was also triggered at a pH lower than 8 (S6K Fig). Viral propagation was delayed for the viruses pretreated with PBS(-)+EDTA (pH 8) solution compared to the viruses pretreated with DMEM, but the difference was smaller than that of MNoV-1 (Fig 1F and S6L Fig). The results indicate that the MNoV-S7 particles more readily convert conformation to the resting state than the MNoV-1 particles in the infection medium of DMEM (pH 7.2-7.4). However, binding of the MNoV-S7 particles pretreated with PBS(-) +EDTA (pH 8) to both RAW264.7 and HEK293T/CD300lf cells showed a significant difference from those pretreated with DMEM (S7A Fig). For early genome replication in RAW264.7 cells from 30 minutes to 12 hours, the virus genome of the MNoV-S7 pretreated with PBS(-) +EDTA (pH 8) increased only 7.4 fold, while that of the MNoV-S7 pretreated with DMEM increased 17.2 fold (S7B Fig). The binding mechanism of MNoV-S7 to host cells may be slightly different from that of MNoV-1. HNoV particles have often been shown to be unstable at alkaline pH, while sensitivity has been shown to vary significantly between genotypes and to be affected by slight sequence variations [25]. Differences in pH sensitivity between the two MNoV strains may also follow a similar mechanism.

Capsid structure of MNoV-S7 VLP
To investigate molecular mechanism of the dynamic rotation of the P domain in the MNoV capsid, we determined the atomic structure of the entire MNoV capsid by single-particle cryo-EM using a 300kV TEM. VLPs of MNoV were used for this structural analysis, because the higher resolution capsid models are easier to obtain from VLPs than infectious particles containing nucleotides [26], and VLPs satisfy the biosafety level requirements of the 300kV EM room. VLPs of MNoV-1 and MNoV-S7 were individually prepared using a modified baculovirus expression system [27,28] and stored in DMEM. The VLPs of MNoV-1 formed multiple types of icosahedral particles of 40-50 nm (S8 Fig), whereas those of MNoV-S7 formed a uniform icosahedral particle in the size of~40 nm (S9A and S9B Figs). Therefore, the VLP of MNoV-S7 was used for the high resolution cryo-EM analysis, and the 3D model was determined at 3.5 Å resolution (Fig 3 and S9 Fig). Local resolutions were in the range of 3.3 to 3.7 Å in the capsid ( Fig 3B). The highest resolution of 3.3 Å was observed in the S domain, and lower resolutions were mainly located in the peripheral regions of the P domain.
Atomic models of the MNoV-S7 VLP capsid were built for the quasi-equivalent A, B, and C monomers of VP1, respectively (S1 Table). The complete Cα backbone structure of the MNoV-S7 VLP is shown in Fig 3C. Other features of the models (e.g., S and P domains, Cα backbone of the VP1 C monomer) are presented in Fig  A comparison of the atomic model of the MNoV-S7 VLP with the reported crystallographic models of the MNoV-1 P domain (PDB ID: 3LQ6 and 3LQE), which were expectedly analyzed in the aqueous condition of the rising P domain conformation [29], shows several interesting results. First, our cryo-EM model of the P domain in MNoV-S7, which was analyzed in the aqueous condition of the resting P domain conformation, had a higher degree of structural similarity to 3LQE than 3LQ6 (S10 Fig [30]. The facts suggest that an allosteric effect in the P domain [31] may exist during the dynamic P domain rotation and the receptor binding, though we have not had the detail structure of P domain in these conformations yet.

Molecular interactions between VP1 proteins
To understand the capsid structure of the resting P domain conformation, which shows higher levels of infectivity, we investigated the molecular interactions between the P domain dimers of the MNoV-S7 VLP suspended in DMEM. The cryo-EM map of the MNoV-S7 VLP in DMEM revealed several hydrophobic and polar interactions between the neighboring A/B and C/C dimers at the P2 level (yellow in Fig 4). We propose the following chemical interactions between residues on the P domain surface: Asn409, Gly411, Leu412, and Pro415 located on the βF''-βB' loop (S5 Fig As expected from sequence homology (S5 Fig), the structure of the S domain of MNoVs was similar to the HNoV GI.1 VLP (PDB ID: 1IHM) [7] except for some local distortions. Our cryo-EM structural study showed several elaborate interactions between adjacent N-terminals of the S domains, by successfully modeling the residues from Gln19 in the A monomer, Ala16 in the B monomer, and Val30 in the C monomer (S11A Fig). The flexible N-terminals extended from A, B, and C monomers formed various interactions with each other between adjacent S domains at the twofold, threefold, pseudo-threefold, and fivefold axes, respectively (S11B-S11F Fig). A new feature found in the structure of MNoV is that the N-terminal arm (NTA) in the A monomer extends to the adjacent A monomer, forming a complex of the fivefold axis (S11E Fig). On the other hand, the NTA in the B monomer extends to the adjacent C monomer, forming a complex of the threefold axis (S11F Fig), though the NTA in C monomer is more disordered. Eventually, these NTAs contribute to form the stable icosahedral inner shell.
As shown in S11G-S11H Fig, the previously reported NTAs of caliciviruses are structurally divided into two groups. One group, belonging to norovirus and lagovirus shown in HNoV GI.1 [7] and RHDV [15], is that the NTA runs along the bottom edge of its S domain and interacts with the adjacent S domains (blue and cyan in S11G and S11H Fig). The other group, belonging to vesivirus shown in SMSV [16] and FCV [17] is that the NTA runs along the bottom edge of the adjacent S domain and interacts with the adjacent S domains (yellow and green in S11G and S11H Fig). NTA in MNoV has a similar conformation as the first group (red in S11G and S11H Fig), further clarifying the function of the NTA in the A monomer. Consequently, the rigid capsid shell of MNoV was maintained by these outer P domain interactions and inner S domain crosslinks.

Molecular mechanism of the reversible rotation of the P domain dimers
We investigated the capsid structure of MNoV in the rising P domain conformation to understand the molecular mechanism of the dynamic rotation of the P domain. The atomic model

PLOS PATHOGENS
Dynamic rotation of norovirus protruding domain An animated comparison of the P domain conformations and reversible structural dynamics of the MNoV capsid under different aqueous conditions is shown in S1 Movie. The P domains in DMEM rested on the S domains and were stabilized by interaction between the adjacent P domains at the P2 level (Fig 5A, 5C and 5E), while the P domains in PBS(-)+EDTA (pH 8) rose up from the S domain and were stabilized by interaction between the adjacent P domains at the P1 level (Fig 5B, 5D and 5F). The flexibility of the hinge connecting the P and S domains (S14A Fig) is important for dynamic twisting of the P domain, which moves the P domain up and down by~13 Å. In addition, as the P domain rose, the A/B dimer was bent slightly toward the fivefold axis (S15 Fig).

Discussion
Here, we reported that reversible dynamic rotation of the P domain results in two conformations of the MNoV capsid. The two conformations of P domain were also identified in HNoV GII.3 VLP at first time. So far, only one of two conformations has been reported in each HNoV VLP (resting conformation: GI.1 (pH4.8), GI.7, GII.2; rising conformation: GII.4, GII.10) and other caliciviruses [11,13,32]. In the case of MNoV, the main trigger for the conformational change was pH of the solution: a higher pH changed the conformation of the P domain to the rising state, while a lower pH changed it to the resting state. Interestingly, the pH thresholds differed by genotype: MNoV-1 showed pH 7 ( Fig 1E) and MNoV-S7 showed pH 8 (S6K Fig). However, it was not clear what determines the threshold for interconversion. Some different amino acids localized between the P and S domains may function by altering

PLOS PATHOGENS
Dynamic rotation of norovirus protruding domain the surface charge, while metal ions bound to the region between the P and S domains, may also help to stabilize the resting conformation. Therefore, in addition to higher pH, EDTA is necessary to remove metal ions from solution, permitting the conversion of the capsid protein to the rising state. As a candidate, we found a potential interaction via a metal ion between two carboxylates (Glu223 and Asp231) observed in the folded, flexible hinge of the resting conformation (S14 Fig). This suggests that the flexible hinge between the P and S domains could be folded by binding metal ions in DMEM and unfolded by releasing the metal ion in PBS(-) +EDTA (pH 8). In addition, several potential interaction sites connecting the S and P domains are also identified in the resting state map (e.g. Glu66 in S domain-Ser519 in P domain, Glu176 in S domain-Tyr227 in P domain, Asp174 in S domain-Gln469 in P domain) (S5 Fig). These residues may function to stabilize the resting conformation via metal ions in DMEM, which includes many inorganic salts such as CaCl 2 , (Fe(NO 3 ) 3 �9H 2 O, MgSO 4 , KCl, NaHCO 3 , NaCl, NaH 2 PO 4 -H 2 O [33]. At present, the metal ions that contribute to domain binding have not been identified, so that the dynamic P domain rotation cannot be controlled by changing their concentrations. To fully understand this mechanism, it is necessary to identify the localization of all metal ions that directly couple the P and S domains must be determined in the future.
Our observations suggest that the capsid of HNoV can also adopt two reversible P-domain conformations, though the dynamic rotation is currently identified only in the MNoV particles. Further investigations are needed to determine the factors that control the orientation of the P domain in HNoV. So far, the resting and rising conformations were independently observed in several HNoV VLPs [32]. However, this report is the first case to present two conformations simultaneously in the same particle in a reversible state. Koromyslova et al. reported the structural alterations in presence of citrate in HNoV GII.10 VLP [34], which strongly resemble the micrographs of the rising P domain in S1E and S6G Figs. We, however, believe the mechanism works differently in the expanded VLP in citrate, because we have confirmed the GII.10 VLP originally represented the rising conformation in the aqueous condition and the resting conformation has not yet been identified [13].
We observed that MNoV-1 particles showed slow transformation between the resting and rising P domain conformations in near neutral pH solution. In HNoV GII.4, post-translational modification of Asp373 by a deamidation was reported to attenuate histo-blood group antigen (HBGA) binding with an estimated half-life of a few days [35]. The slow reaction is also strongly linked to pH. However, it seems to be a different mechanism from the P-domain rotation in MNoV, as no compatible residues of Asp373 were found in the sequences (S5 Fig), in addition to the fact that glycan binding is not essential for infection of MNoV. The slow regulation step to control infectivity may be common between MNoV and HNoV and essential for infection in the dramatic environmental changes present in the gastrointestinal tract, though the approaches are different.
Our infection and adhesion experiments of the virus particles to the host cells clearly suggested that the viruses having the resting P domain conformation attach to and infect host cells more easily than the viruses having the rising P domain conformation. To structurally confirm the difference in the affinity for the host cell surface between the resting and rising P domain conformations, the crystallographic model of the P domain and soluble domain of CD300lf (sCD300lf) (PDB ID: 5OR7) was fitted to the cryo-EM maps of MNoV-1 in the different P domain conformations. As shown in S16 Fig, the sCD300lf binding site on the P domain is formed inside a triangle between one C/C dimer and two A/B dimers on the viral surface. In the resting P domain conformation, the triangle is distorted and the sCD300lf molecules are more flexibly able to approach the binding site (S16A Fig). On the other hand, in the rising P domain conformation, the triangle is squeezed, limiting the space for the sCD300lf molecules

PLOS PATHOGENS
Dynamic rotation of norovirus protruding domain to approach the binding site (S16B Fig). The previous reports also suggested that a steric hindrance in the sCD300lf molecules would occur between the P domain dimers in the rising P domain conformation, where the C-terminals of sCD300lf clashed in its multiple conformation conforming to the space formed between the P domain dimers, while it had more space between the P domain dimers in the resting conformation [11,19,30,31]. The adaptation of the sCD300lf binding model in our cryo-EM map clearly showed that the accessibility of CD300lf to the P domain is altered in the resting and rising P domain conformations in the single virus. It shows a possibility that the rotation of the P domain modulates viral infectivity by altering the accessibility of the viral epitopes to cellular receptors.
In consequence, the resting P domain conformation is more accessible to the cellular receptor CD300lf than the rising P domain conformation, causing greater viral infectivity. However, it raises a question: why does norovirus require such a reversible rising P domain conformation? One possibility is that the conformational changes are important during virus assembly and disassembly in host cells. Tobacco mosaic virus (TMV), well-studied for its structural and physiological characteristics, forms a metastable structure in the cell, where lower concentration of metal ions and higher pH are proposed as the trigger for virus disassembly [36]. A recent cryo-EM study showed the calcium ion-dependent conformational changes of TMV, indicating that chemical bonds involving calcium ions prevents disassembly of TMV [37]. Dengue and Zika viruses also show pH-dependent structural changes associated with viral replication [38,39]. In this study, we observed the MNoVs having the rising P domain conformation showed about 2 Å lower resolution (~7.2 Å) than the MNoVs having the resting P domain conformation (~5.2 Å) (S1D, S1H, S6F and S6J Figs). In addition, the particles finally contributed to the 3D map were 23~28% of the initially picked particles in the rising conformation data set, while they were 91~95% in the resting conformation data set (S2 Table), where mixed P domain conformations, departure from icosahedral symmetry, and broken particles were removed during iterative 2D and 3D map refinements. These facts suggest that MNoV with the rising P domain conformation are relatively flexible and fragile. Such structural flexibility and fragility may facilitate viral assembly and disassembly. The flexibility and fragility of the rising P domain conformation was also identified in HNoV GII.3 VLPs, where the rising P domain conformation easily collapsed in thinner vitreous ice film due to surface tension (S17C Fig), and thus the only particles with the resting P domain conformation were left in the central thinner ice area of the cryo-EM grid hole (S17A and S17B Fig). In the case of HNoV GII.3 VLPs, the particles finally contributed to the 3D map was 29% of the initially collected particles in the rising conformation data set, while they were 50% in the resting conformation data set (S2 Table). Assuming that the rising P domain conformation represents an intermediate state of the viral assembly and disassembly, norovirus changes its structure to a metastable form upon entry into host cells, and release the viral genome into the cytosol at the relatively lower metal ion concentration and higher pH environment of the cytoplasm. Conversely, norovirus preassembled in host cells may be further modified to its final stable form in the more demanding extracellular environment, such as the gastrointestinal tract, which is rich in free metal ions and lower pH. Another hypothesis has been suggested that the fragile and flexible P domains function in antibody escape or facilitate binding to target cells [11]. We have to wait for future study to provide an answer to this question.
Recently, a dynamic feature of calicivirus capsid was reported in FCV. There, the twelve copies of the minor capsid protein, VP2, form a portal-like assembly on the capsid surface after receptor engagement, suggesting that it functions as a channel for delivery of the viral genome [18]. As with FCV, no VP2 density was found in our symmetry-imposed cryo-EM map of noroviruses. One reason is that the copy number is limited and does not match the icosahedral symmetry, where the components disappear due to imposition of symmetry.
Assuming the same mechanism as FCV, VP2s in noroviruses dispersed inside the particle could accumulate and be assembled according to the structural changes due to receptor involvement. It is unknown whether MNoV has a similar portal structure consisting of VP2 at present, but a rearrangement of the capsid structure, such as the dynamic rotation of the P domain in addition to the receptor binding is believed to be a necessary step to form the portal and release the viral genome.
Interactions between P domain dimers on the capsid surface have also been observed in human noroviruses, but are more complex and diverse. The HNoV GI.1 VLP shows dimerdimer interactions at the P2 subdomain level, where the P2 subdomain of the C/C dimer links with all four adjacent A/B dimers and the P domains rest on the S domains [7]. By contrast, in HNoV GII.10 VLP the similar dimer-dimer interactions are carried out at the P1 subdomain level not at the relatively small P2 subdomains, and the P domain rises from the S domain bỹ 16 Å, like the rising P domain conformation in MNoV. These different conformations of the P domains in the two HNoV strains result from switching interaction sites at the P1 or P2 levels, which can be structurally achieved by a simple rotation of the P domain. In addition, we previously reported a unique neutralizing antibody against the HNoV GII.10 VLP, which binds to the site of occlusion between the P and S domains of the viral capsid, inhibiting binding to target cells [13]. This suggests that the antibody inhibits viral-cell attachment or/and infection by interfering with the rotation of the P domain from the rising position to the resting position. Currently, dynamic rotation of the P domain has been identified only in the infectious particles of MNoV. Further studies will reveal the full mechanism of the non-enveloped calicivirus infection.

Sample preparations of norovirus infectious particles and VLPs
Infectious particles of MNoV-S7 were produced using plasmid-based reverse genetics as previously described [22]. Those of MNoV-1 were produced in the same procedure as MNoV-S7. pMuNoV-MNoV-1 was constructed by exchanging the MNoV sequence portion from MNoV-S7 to MNoV-1 with the In-Fusion cloning system (Takara Bio Inc.), according to the manufacturer's protocol. MNoV-1 infectious cDNA was kindly provided by I. Goodfellow as an MNoV-1 cDNA plasmid [40]. Infectious viruses of MNoV-1 and MNoV-S7 were propagated with RAW264.7 cells (American Type Culture Collection). The culture supernatant was collected two days after infection. MNoV was pelleted using ultracentrifugation with a 30% sucrose cushion and purified with cesium chloride (CsCl) equilibrium ultracentrifugation. MNoV was diluted with DMEM without any additional supplements and pelleted again to remove CsCl. The purified MNoV pellet was resuspend in DMEM (Nacalai Tesque Inc.) or phosphate-buffered saline without calcium and magnesium containing 20mM EDTA at pH 8.0 (PBS(-)+EDTA (pH 8.0)) to prepare 10 7 infectious virions /μL. VLPs of MNoV-1, MNoV-S7, and HNoV GII.3 TCH04-577 strains were produced by a baculovirus expression system [27,28] and purified with CsCl equilibrium ultracentrifugation in the same procedure as infectious particles.

Analysis of MNoV one-step propagation curve
RAW264.7 cells were cultured at a density of 10 6 cells/well in 48-well plates and infected with the purified infectious particles of MNoV-1 or MNoV-S7 at a MOI = 2 or more. After 30 min of incubation at 37˚C, the inoculum was removed, and the cells were washed three times with DMEM to remove unbound viruses, and DMEM containing 10% FBS (Thermo Fisher Scientific) was added. The plates were incubated at 37˚C for the stated time; time zero indicates the

PLOS PATHOGENS
Dynamic rotation of norovirus protruding domain time at which the medium was added. After 0, 3, 6, 9, 12, 15, 18 and 24 h of incubation, 20 μL of the culture medium was sampled and centrifuged at 10,000 g for 10 min at 4˚C to remove cells and cell debris. 15 μL of supernatant was collected and stored at -80˚C until RNA extraction. 10 μL of the supernatant was used for RNA extraction and qRT-PCR, according to published protocols [41]. These assays were performed three times independently and calculated standard deviations (SD) plotted at each point.

Evaluation of MNoV-1 Binding using FACS
MNoV-1 binding to the host cells was examined by FACS (fluorescence activated cell sorter) analysis. Briefly, 1 × 10 6 of HEK293T/CD300lf cells were incubated with purified infectious MNoV-1 particles (1x10 9 CCID50), that were suspended in DMEM-10% FBS or PBS (-) + 20mM EDTA (pH 8), for 30 min at 4˚C. After washing, the cells were incubated with anti-MNoV VP1 rabbit polyclonal antibody labeled by Dylight 488 (LNK221D488, BioRad) for 30 min at 4˚C. The solution in each step includes 3% FCS and 20 mM NaN 3 to prevent the cells from virus uptake. Afterward, the cells were washed again and analyzed by BD FACSMelody (BD Bioscience) and analyzed by using FlowJo software.

Evaluation of cell binding and early genome replication of MNoV in the host cells using qRT-PCR
RAW264.7 or HEK293T/CD300lf cells were cultured at a density of 10 5 cells/well in 96-well plates and infected with the purified infectious particles of MNoV-1 or MNoV-S7 at a MOI ≦ 10. After 30 min of incubation at 37˚C, the cells were washed twice with DMEM to remove unbound viruses. The total RNA was extracted by NucleoSpin RNA (Takara Bio Inc.) and the virus particles attached on the cell surface were estimated by qRT-PCR, according to published protocols [41]. The early genome replication in each cell incubated for 0, 2, 4, 6, 8, 10, and 12 hours post-infection. RNA was extracted using the same procedure as above and used for qRT-PCR. These assays were performed three times independently and calculated standard deviations (SD) values and plotted at each point.

Cryo-EM and image processing
Aliquots (2.5 μL) of the purified MNoV infectious particles and HNoV GII.3 VLP were placed onto R 1.2/1.3 Quantifoil grids (Quantifoil Micro Tools) coated with a thin carbon membrane that were glow-discharged using a plasma ion bombarder (PIB-10, Vacuum Device Inc.) immediately beforehand. These grids were then blotted and plunge-frozen using a Vitrobot Mark IV (Thermo Fisher Scientific) with the setting of 95% humidity and 4˚C. Vitreous ice sample grids were maintained at liquid-nitrogen temperature within a JEM2200FS electron microscope (JEOL Inc.), using a side-entry Gatan 626 cryo-specimen holder (Gatan Inc.), and were imaged using a field-emission gun operated at 200 kV and an in-column (Omega-type) energy filter operating in zero-energy-loss mode with a slit width of 20 eV. Images of the frozen hydrated norovirus particles were recorded on a direct-detector CMOS camera (DE20, Direct Electron, LP) at a nominal magnification of 40,000×, corresponding to 1.422Å per pixel on the specimen. Using a low-dose method, the total electron dose for the specimen is about 20 electrons per Å 2 for a 3-second exposure. Individual images were subjected to per-frame drift correction by a manufacturer provided script.
For MNoV-1 infectious particles suspended in DMEM and PBS(-)+EDTA (pH 8.0), 6,708 and 4,704 particles were selected from 1,046 and 2,188 images, respectively, and then extracted using RELION 2.0 [42] after determining the contrast transfer function (CTF) with CTFFIND4 [43]. Alignment and classification of extracted particles were performed, and a 3D

PLOS PATHOGENS
Dynamic rotation of norovirus protruding domain map was reconstructed in RELION 2.0 by using an initial model that generated with icosahedral symmetry by EMAN1 [44]. The 3D reconstructions were computed, and the final resolutions of the density maps were estimated to be resolutions of 5.3 and 7.3 Å, respectively, using a GS-FSC (cutoff 0.143) between two different independently generated reconstructions [45]. 3D renders of the maps were created in UCSF Chimera [46]. For MNoV-S7 infectious particles suspended in DMEM and PBS(-)+EDTA (pH 8.0), 2,049 and 2,739 images were collected, respectively, and the 3D reconstructions were calculated with 17,820 and 5,063 particles at 5.2 and 7.2 Å resolution, respectively. For HNoV GII.3 VLP containing the resting P domain and the rising P domain, 106 and 1,917 images were collected, and 3D reconstructions were calculated with 279 and 1,482 particles at 9.3 and 12.9 Å resolution, respectively. Data collection and image processing are summarized in S2 Table. For high-resolution structural analysis, cryo-EM images for MNoV-S7 VLP were acquired with a Falcon II detector at a nominal magnification of 75,000×, corresponding to 0.86 Å per pixel on a Titan Krios at 300 kV (Thermo Fisher Scientific). A low-dose method (exposures at 20 electrons per Å 2 per second) was used, and the total number of electrons accumulated on the sample was~40 electrons per Å 2 for an 2 second exposure. A GIF-quantum energy filter (Gatan Inc.) was used with a slit width of 20 eV to remove inelastically scattered electrons. Individual micrograph movies were subjected to per-frame drift correction by MotionCor2 [47]. Particles were selected from the 2,746 images and the final 3D reconstruction was computed with 41,847 particles. The resolution of the density map was estimated to be 3.5 Å using a GS-FSC criterion. Data collection and image processing are summarized in S1 Table.

Atomic model building of MNoV-S7 VP1
The P domain atomic model of MNoV-1 (PDB ID: 3LQE) and the S-domain atomic model of HNoV (PDB ID: 1IHM) were used as templates for the homology model building of the P and S domains of the MNoV-S7 VP1, respectively. Multiple-sequence alignments of VP1s of MNoV-S7, MNoV-1 and HNoV were performed using the PROMALS3D program [48]. The sequence alignments of the P and S domains were used as the input of MODELLER software [49] to generate the comparative models. The maps containing A, B, and C monomers were extracted using UCSF Chimera [46], respectively, for VP1s of MNoV-S7 and MNoV-1. The models were manually re-built in the individual maps from the homology model mentioned above using COOT [50] and refined using PHENIX [51]. The models fitted to the cryo-EM maps were evaluated with the cross-correlation coefficient between the model and the cryo-EM map respectively, using "Fit in Map" of UCSF Chimera [46]. Data collection, image processing, and model statistics are summarized in S1 Table. Supporting information S1 Table.  The fraction of MNoV-S7 particles exhibiting the rising P domain conformation in different aqueous conditions. Random sampling of 1,000 particles was carried out five times under each condition. The number of particles in each conformation was counted from the individual 2D classes. The rising P domain conformation of MNoV-S7 appeared at slightly higher pH than MNoV-1 ( Fig 1E). (L) Propagation curves of MNoV-S7 pretreated in DMEM and PBS(-)+EDTA (pH 8), respectively. The infectious particles were infected to RAW264.7 cells and measured the number of the duplicated viral RNA copies along the timeline by qRT-PCR. Error bars represent the standard deviations. The values of the data sets in DMEM and PBS(-)+EDTA (pH 8) were slightly closer and the Mann-Whitney rank test was performed at each time point as follows: 0h 0.512; 6h 0.827; 9h 1.997; 12h 0.335; 15h 0.049; 18h 0.049; 21h 0.049; 24h 0.335. Significant differences (p<0.05) between the two data set were found after 15 hours. MNoV-S7 pretreated with PBS (-)+EDTA (pH 8) produced slower propagation of the virus than that pretreated with DMEM, though the difference was small compared to that of MNoV-1 (Fig 1F).

PLOS PATHOGENS
Dynamic rotation of norovirus protruding domain sCD300lf molecules (magenta) are able to bind on the P domain inside the triangle (white dot lines) formed between one C/C dimer and two A/B dimers. In the resting P domain conformation, the triangle is distorted and the CD300lf molecules are more flexibly able to approach the binding site (A), while in the rising P domain conformation, the triangle is squeezed, limiting the space for the sCD300lf molecules to approach the binding site (B). (TIF) S17 Fig. Structural stability of  VLPs having the rising P domain conformation easily collapsed due to surface tension in thin vitreous ice, and only the stable VLPs having the resting P domain conformations were left. (C) In the ice embedded cryo-EM grid, the ice film is formed in a convex shape, where the center is thinner than the hole edge. Eventually, the unstable particles move to the edge or collapse. (TIF) S1 Movie. Molecular mechanism of dynamic rotation of the P domain dimers. (MP4)