High-Resolution Cryo-EM Reveals Dynamics in the Murine Norovirus Capsid

Rather than acting as rigid symmetrical shells to protect and transmit their genomes, the capsids of non-enveloped, icosahedral viruses co-ordinate multiple, essential processes during the viral life-cycle, and undergo extensive conformational rearrangements to deliver these functions. Capturing conformational flexibility has been challenging, yet could be key in understanding and combating infections that viruses cause. Noroviruses are non-enveloped, icosahedral viruses of global importance to human health. They are a common cause of acute non-bacterial gastroenteritis, yet no vaccines or antiviral agents specific to norovirus are available. Here, we use cryo-electron microscopy to study the high-resolution solution structures of infectious, inactivated and mutant virions of murine norovirus (MNV) as a model for human noroviruses. Together with genetic studies, we show that the viral capsid is highly dynamic. While there is little change to the shell domain of the capsid, the protruding domains that radiate from this are flexible and adopt distinct states both independently and synchronously. In doing so the viral capsid is able to sample a defined range of conformational space, with implications for the maintenance of virion stability and infectivity. These data will aid in developing the first generation of effective control measures against this virus.


Introduction
result of individual P domain dimer mobility, and not the coordinated movement of P 135 domains across entire capsids. 136 137 With this observation in mind, we performed focussed 3D classification on AB-and 138 CC-type P domain dimers separately ( Figure S3a,b). This approach involves the 139 assignment of sixty symmetrically redundant orientations to each particle and 140 application of a mask to focus classification on a substructure within the virion -here, 141 an AB-or CC-type P domain dimer. This approach revealed a striking diversity in the 142 orientation and location of AB dimers (Figure 1f), but much less variation in CC dimer 143 positioning ( Figure 1g). This rationalises the lower resolution of A-and B-type P 144 domain density than C-type P domain density described above. Complete capsids 145 reconstructed from individual classes showed well-resolved density for the P domain 146 dimer that was contained within the mask during focussed classification, but lower 147 quality density for P domain dimers outside of this mask ( Figure S3c). Importantly, 148 these data are consistent with P domain dimers being mobile elements that move 149 independently of each other on the capsid surface. However, it should be noted that 150 this approach did not lead to a significant improvement in the quality of P domain 151 density, presumably because the number of particles contributing to the final 152 reconstruction was split between multiple classes, reducing the data in any individual 153 reconstruction. 154 155

An atomic model of wtMNV VP1 156
The quality of the data allowed us to build a hybrid atomic model for VP1 ( Figure S4, 157 Table S2). To construct an initial atomic model for refinement into our EM density map, 158 a homology model of an MNV VP1 S domain was generated using the Phyre2 server 16 , 159 based on the crystal structure of the Norwalk virus capsid (PDB: 1IHM) 9 . This 160 homology model was rigid-body fitted into our density map and copies were fitted into 161 each quasi-equivalent position (A, B and C) within the asymmetric unit, before each 6 was manually edited and then refined to improve fit against the experimental cryo-EM 163 density and the geometry of the model. The resolution in the S domain was sufficient 164 to allow confident building of most residues up to the flexible linker region (Figure 1h). 165 As expected, the flexible linker connecting the S and P domains was not resolved for it is less mobile than the A-or B-type conformer, and correlating with the improved 168 resolution of C-type P domains. To model the P domains, a crystal structure of an MNV 169 VP1 P domain (PDB: 6C6Q) 11   2a). We also investigated capsid stability by performing PaSTRy assays, which 206 employ two fluorescent dyes, SYTO-9 (which binds to nucleic acids) and  Orange (which binds to hydrophobic regions of proteins), to assess the stability of viral 208 capsids, independent of viral infectivity 19 (Figure 2b). While there was a 10,000-fold 209 reduction in infectivity at 61°C, PaSTRy assay data suggested that capsids remained 210 essentially intact up to ~64°C, as minimal SYTO-9 fluorescence suggests that the viral 211 RNA was not exposed to bulk solvent below this temperature. Confirming this, MNV 212 heated to 61°C (which we termed heat inactivated MNV, or hiMNV) was incubated with 213 RNase and no digestion of the RNA genome was observed ( Figure 2c). Thus, we had 214 identified a temperature at which the capsids had become irrevocably non-infectious, 215 but were not disassociated into their component parts. 216

217
The cryo-EM reconstruction of hiMNV reveals an increase in P domain mobility 218 To understand the structural changes that occurred during thermal stress, we 219 determined the structure of hiMNV by cryo-EM at 2.9 Å resolution. This resolution is 220 significantly higher than for wtMNV (2.9 Å vs 3.1 Å; Figure S1b  To identify mutation(s) present in the thermostable MNV population, the structural 248 protein-encoding region of the MNV genome (ORF2 and ORF3) was amplified by  PCR and sequenced at the consensus level. No mutation was seen in ORF3 250 (encoding VP2), but a single mutation was found in ORF2 that leads to a single amino 251 acid substitution in VP1, L412Q. Consistent with our hypothesis, this mutation was 252 located in the VP1 P domain, on the hinge loop connecting the P1 and P2 subdomains 253 (for reference, see Figure 1i). To further characterise the effect of the L412Q 254 substitution, we reconstituted the mutation in an infectious clone of MNV and used it 255 to recover 'heat-stable' (hs)MNV particles. Like MNV52, hsMNV remained infectious 256 after incubation at temperatures that rendered wtMNV non-infectious ( Figure S5). 257 Given that hsMNV had an amino acid substitution in the P domain of VP1, we also 258 looked for changes in antigenicity by ELISA that may be indicative of a conformational temperature that wtMNV could not tolerate) (Figure 4d). 262 263

The cryo-EM reconstruction of hsMNV shows 'twisted' AB-type P domains 264
To gain a structural insight into the mechanism of stabilisation of hsMNV, we 265 determined the structure of hsMNV to a resolution of 3.1 Å ( Figure S1c, S2c). 266 Interestingly, while CC-type P domain dimers appear virtually identical to wtMNV, AB-267 type P domain dimers showed a subtle difference in their orientation (Figure 5a,b, 268 Movie S2). To explore this change, we performed rigid-body fitting of the atomic 269 coordinates for wtMNV VP1 into the hsMNV map. This was followed by refinement 270 against the hsMNV map with secondary structure restraints enabled. In the wtMNV 271 map an interface is formed between A-type and C-type P domains ( Figure 5c), but for 272 hsMNV, this interface has been disrupted ( Figure 5d,e). While the C-type P domain 273 did not show any significant movement, the AB-type P domain dimer has tilted 274 upwards, angling away from the S domains, and rotated in an anti-clockwise direction 275 (Movie S3). As such the mutated residue now points away from the interface. In 276 agreement with this, PDBePISA 22 analysis of VP1 fitted into the wtMNV map suggests 277 that C-type VP1 L412 is a buried residue and contributes to an interface with A-type 278 VP1. When fitted into the hsMNV map, no interface is detected between A-type and 279 C-type VP1. 280

281
In summary, our data suggest that the mutant virion is sampling a subset of the 282 conformations that can be explored by the wild-type virion. Although not affecting 283 infectivity, this is likely to impact other aspects of the viral lifecycle. 284

286
The structural data presented above shows the dynamic flexibility of a murine 287 norovirus virion, by comparing three high-resolution structures of a single viral species. 288 We showed that the infectious norovirus virion is a highly flexible macromolecular 289 machine that is capable of sampling a range of conformational space, whilst 290 maintaining its integrity and functionality. However, it is clear from the increased 291 dynamics of heat-inactivated virions that too much P domain mobility is detrimental to 292 infectivity and fitness. The virus has therefore evolved to control these dynamics and 293 establish a balance between too much and too little P domain mobility. 294 295 MNV is a genogroup V norovirus so is closely related genetically and structurally to 296 human noroviruses (genogroups I, II and IV), and the well-established cell culture and 297 reverse genetic systems for MNV allowed us to specifically probe the functional 298 significance of interactions revealed by our structural data 23,24 . Firstly, we report the 299 high-resolution solution structure of an infectious norovirus. Our cryo-EM 300 reconstruction of wtMNV is strikingly different to the previous, 8.0 Å resolution cryo-301 EM reconstruction of wtMNV, with large changes in the positioning and orientation of 302 P domain dimers relative to the S domains 6 . This gross conformational change in P 303 domains was consistent across all three of our independent reconstructions 304 (discussed below) and importantly, the orientation and positioning of P domains in our 305 wtMNV reconstruction are in line with those of most human norovirus VLPs 5,9 . 306 307 Currently, the reasons for the differences between our reconstruction and that in 308 there were small differences in virus purification protocols, and we cannot rule out that 318 the two reconstructions were on viruses with subtly different VP1 primary sequences 319 (e.g. arising from mutation during passage), despite both starting with a CW1 strain of 320 MNV-1 7 . It is also possible that differences in buffer composition are responsible for 321 the alternative conformations. Given the changing ionic composition of the 322 environment that the virus is exposed to during endocytosis into a target cell 26 , this 323 latter point may be of biological importance. Using the crystal structure of an MNV P generated a receptor docking model based on the MNV reconstruction reported here 326 ( Figure S6). Interestingly, a clash between trimers of CD300lf in the previous docking 327 model 11 is resolved in this model, and there is space for higher receptor occupancy. 328

329
The independent mobility of individual P domain dimers on the viral capsid surface is 330 striking. While it is suspected that individual norovirus species can adopt multiple 331 different gross morphologies 15 , here we provide direct structural evidence for 332 alternative morphologies within a single norovirus species, and within individual 333 norovirus particles. Using focussed 3D classification, we identified a remarkable 334 diversity in P domain dimer positioning, which was not co-ordinated over the capsid 335 surface, and was not equal between different quasi-conformers, with CC-type dimers 336 less mobile than AB-type dimers. This contrasts with data obtained for another 337 calicivirus, Tulane virus, where CC-type dimers were shown to be more mobile 27 . 338 Surprisingly, one class (out of ten) for each of the AB-and CC-type dimer 339 classifications had an inverted orientation in the z-axis ( Figure S3a,b). However, for 340 the remaining classes, no significant improvement in P domain density was observed 341 -likely because any improvement from removing heterogeneity in the reconstruction 342 is countered by a decreased number of particles contributing to the final reconstruction. 343 344 By determining the structure of the wtMNV capsid using cryo-EM, we were able to 345 identify unambiguous differences between quasi-equivalent subunits of VP1. The N-346 terminal regions of VP1 were particularly variable -in A-type VP1, the N-terminal arm 347 protrudes deeper into the capsid to interact with other A-type subunits around the 348 icosahedral five-fold axis. B-type VP1 N-terminal regions run towards the icosahedral 349 three-fold axis, and C-type VP1 N-terminal regions are not resolved in our map, 350 suggesting increased flexibility. We note that this difference between quasi-equivalent 351 subunits could provide a mechanism to guide the positioning of other components of 352 the virion, such as VP2. Based on work with feline calicivirus, it has been proposed 353 that VP2 may bind the capsid interior at a single icosahedral three-fold axis 28 . While it 354 has been shown through mutational analysis that VP1 N-termini are not required for 355 co-precipitation with VP2 13 , the ability of VP1 N-termini to organise differently may 356 provide an interface for 'recognition' of the three-fold axis (formed by B-and C-type 357 VP1 subunits); i.e., while not required for VP2 packaging, one may speculate that differences between N-terminal regions of VP1 could guide the positioning of VP2 359 within the capsid interior. Alternatively, these differences may guide the positioning of 360 VPg or the RNA genome, which may in turn guide VP2. 361

362
To support our hypothesis that viral P domains are mobile elements that can adopt 363 multiple conformations, we endeavoured to induce conformational changes reflective 364 of those that may occur during infection by thermally stressing MNV, leading to the 365 generation of non-infectious virus with intact capsids (hiMNV). While cryo-EM revealed 366 no morphological differences, P domain density was weaker than for wtMNV. Weaker 367 density may arise from 'partial occupancy' (i.e. inclusion of particles that have lost 368 some or all P domains in the final reconstruction), or more likely from increased P 369 domain flexibility/mobility. While we saw small particles reminiscent of capsid proteins 370 in the background of raw micrographs from the hiMNV data set ( Figure S1b), we did 371 not observe any P domain-lacking capsids in raw micrographs and failed to pull out 372 classes lacking P domains in asymmetric 3D classification, so believe the latter 373 explanation to be the most plausible. It is not immediately clear why increased P 374 domain mobility would render the virus non-infectious -perhaps the range of 375 conformations sampled by the P domains is changed, such that a conformation 376 required for infection is no longer accessible. However, we cannot rule out the 377 possibility of other subtle structural changes within the unresolved regions of P 378 domains that contribute to virus inactivation. These potential changes may disrupt the 379 interaction between the virus and its receptor, or could cause a change in the 380 amount/composition of metal ions coordinated by VP1, which are important for viral 381 infectivity and capsid stability 8,11 . We also cannot discount the possibility that structural 382 changes to other capsid components/contents (such as VP2 or VPg which are not 383 resolved in our icosahedrally-averaged reconstructions) may contribute to the loss of 384 infectivity, though we believe that changes to VP1 are ultimately responsible. 385

386
In support of this hypothesis, MNV passaged with a thermal selection pressure 387 acquired resistance to heat inactivation from a single point mutation in the gene 388 encoding VP1 (ORF2). Our cryo-EM reconstruction of hsMNV showed that AB-type P 389 domain dimers were 'twisted/tilted' relative to their positions in wtMNV, disrupting an 390 interface between AB-and CC-type dimers. While it is theoretically possible that this 13 altered P domain dimer positioning is unrelated to the mutation, the most plausible 392 interpretation is that the L412Q substitution (which is located within this interface) is 393 responsible. It is not obvious why the mutant morphology is more stable -one would 394 expect disruption of an interface to decrease stability. Indeed, thermostabilised foot-395 and-mouth disease virus (FMDV) capsids were generated by introducing a disulphide 396 bond to stabilise an interface between adjacent pentamers 29 , and many stabilising 397 mutations seen in poliovirus are thought to act by stabilising interfaces between 398 subunits 30 . We suspect that the 'twisted' morphology of hsMNV liberates P domain 399 dimers to enter a protective conformation upon heating, that may only become To prepare MNV samples for cryo-EM, lacey carbon 400-mesh copper grids coated 541 with a <3 nm continuous carbon film (Agar Scientific, UK) were glow-discharged in air 542 or amylamine vapour (10 mA, 30 seconds) before applying two to three 3 µl aliquots 543 of purified MNV, to improve the concentration of virus on the grid surface (as described 544 previously 39 ). Each application was followed by a 30 second incubation period at 80% 545 relative humidity (8°C), then the grid was manually blotted to remove excess fluid 546 before the next application. 30 seconds after the final application, grids were blotted 547 and vitrified in liquid nitrogen-cooled liquid ethane using a LEICA EM GP plunge 548 freezing device (Leica Microsystems). Grids were stored in liquid nitrogen prior to 549 imaging with an FEI Titan Krios transmission electron microscope (ABSL, University 550 of Leeds) at 300 kV, at a magnification of 75 000´ and a calibrated object sampling of 551 1.065 Å/pixel. A complete set of data collection parameters for each sample is provided 552 in Table S1. 553

Image processing 555
Following cryo-EM data collection, the RELION-2.1 and RELION-3.0 pipelines 40-42 556 were used for image processing. Drift correction was first performed on micrograph 557 stacks using MOTIONCOR2 43 , and the contrast transfer function for each was 558 estimated using Gctf 44 . A subset of virus particles was picked manually and subject to 559 2D classification, with resultant classes used as templates for automatic particle 560 picking 45 . Particles were classified through multiple rounds of reference-free 2D 561 classification, and particles in poor quality classes were removed after each round. An 562 initial 3D model was generated de novo 46 and used as a reference for 3D auto-563 refinement with icosahedral symmetry imposed. This reconstruction was post-564 processed to mask and correct for the B-factor of the map, before (i) taking particles 565 forward to CTF refinement and Bayesian polishing, or (ii) further 'clean-up' by 566 alignment-free 3D classification, with particles from subsequent 3D auto-refinement 567 and post-processing being used for CTF refinement and Bayesian polishing. Multiple 568 rounds of CTF refinement (with or without beamtilt refinement) and Bayesian 569 polishing 47 were performed, before final icosahedral symmetry-imposed 3D auto-570 refinement and post-processing. The nominal resolution for each map was determined 571 according to the 'gold standard' Fourier shell correlation criterion (FSC = 0.143) 48 , and 572 the local resolution estimation tool in RELION was used to generate maps filtered by 573 local resolution. 574 575 To investigate P domain mobility, a focussed 3D classification approach was employed 576 (as described previously 14,49-51 ). Briefly, each particle contributing to the final 577 icosahedral symmetry-imposed reconstruction was assigned 60 orientations 578 corresponding to its icosahedrally-related views using the relion_symmetry_expand 579 tool. SPIDER 52 was used to generate a cylindrical mask to isolate either an AB-type 580 or a CC-type P domain dimer, and the symmetry expanded particles were subjected 581 to masked 3D classification without alignment, using a regularisation parameter ('T' 582 number) of 20. Classes were inspected visually, and particles from selected classes 583 (with assigned orientation information) were used to generate full capsid 584 reconstructions without imposing symmetry, using the relion_reconstruct tool. 585 sequence corresponding to the S domain of MNV VP1 was used to build a homology 589 model with the Phyre2 server 16 , which was rigid-body fitted into each quasi-equivalent 590 position in the wtMNV density map using UCSF Chimera 53 . This preliminary model 591 was manually refined in Coot 54 , symmetrised in UCSF Chimera to generate the other 592 59 copies of the asymmetric unit that form the capsid, then subject to 'real space 593 refinement' in Phenix 55 . To improve the geometry of the coordinates and fit of the 594 model to the density map, the S domain model was iterated between manual fitting in 595 Coot and refinement in Phenix. Following this, the crystal structure of an MNV P 596 domain complexed with CD300lf (PDB: 6C6Q) 11 was also fitted into the map to occupy 597 each quasi-equivalent position of the asymmetric unit, after removing ligands/CD300lf 598 and correcting the peptide sequence. P domain coordinates were combined with the 599 refined S domain model, and subject to a single round of refinement in Phenix. For 600 each real space refinement, secondary structure restraints were imposed. Molprobity 56 601 was used to validate the model.