Arrangement of the polymerase complexes inside a nine-segmented dsRNA virus

Summary

Members of the family Reoviridae package several copies of the viral polymerase complex into their capsid to carry out replication and transcription within viral particles. Classical single-particle reconstruction encounters difficulties resolving structures like the intraparticle polymerase complex because refinement can converge to an incorrect map and because the map could depict a non-representative subset of particles or an average of heterogeneous particles. Using the nine-segmented Fako virus, we tested hypotheses for the arrangement and number of polymerase complexes within the virion by measuring how well each hypothesis describes the set of cryoEM images of individual viral particles. We find that the polymerase complex in FAKV binds at ten possible sites despite having only nine genome segments. A single asymmetric configuration describes the arrangement of these complexes in both virions and genome-free capsids. Similarities between the arrangements of Reoviridae with nine, ten, and eleven segments indicate the generalizability of this architecture.

eTOC blurb

The arrangement of polymerases inside an unusual 9-segmented dsRNA virus was determined by Kaelber et al. While cryoEM is typically used to average copies of a protein into one structure, they also compared raw images to synthetic maps to test whether subsets of virions had different arrangements.

Introduction.

Members of the family Reoviridae have dsRNA genomes of twelve (e.g. Colorado tick fever virus), eleven (e.g. rotavirus), ten (e.g. cytoplasmic polyhedrosis virus, CPV), or nine (e.g. Fako virus) segments. All dsRNA segments must be present in an incoming virion to initiate a successful infection. These (Fajardo et al., 2016) and other segmented viruses utilize cis-acting RNA sequences to ensure that most virions will contain exactly one copy of each genome segment (McDonald et al., 2016). When a virion of the family Reoviridae enters a cell, it does not disassemble to release genomic RNA; rather, the dsRNA is retained within the capsid. Transcription and replication are fully conservative. Polymerase complexes (PCs) inside the capsid extrude the nascent RNA strand through pores in the capsid surface. For Fako virus (FAKV) and its close relatives, the PC consists of an RNA-dependent RNA-polymerase (henceforth, “polymerase”) and an accessory NTPase.

In all members of the family, the PC is located near the icosahedral five-fold symmetry axis, directly beneath the innermost capsid layer (Estrozi et al., 2013)(Dryden et al., 1998)(Prasad et al., 1996)(Zhang et al., 2003). There are twelve such locations in an icosahedron, and the major capsid protein (MCP) forms a homopentamer at each of the sites. Because the pentamer is symmetric, there are five equivalent positions where the PC could bind at the pentamer. However, steric occlusion prevents more than one of these five orientations to be occupied simultaneously (Estrozi et al., 2013)(Zhang et al., 2003). Although it is sterically feasible to have up to twelve PCs per virion (a maximum of one per five-fold vertex), the actual number may be lower.

Two asymmetrically reconstructed cryoEM structures have been described for cytoplasmic polyhedrosis virus, which contains ten dsRNA segments inside a single capsid shell (Zhang et al., 2015)(Liu and Cheng, 2015). In the structure by Zhang et al., ten PCs were well-resolved, leaving two vertices empty (Zhang et al., 2015). Meanwhile, Liu & Cheng report a structure wherein strong density was observed for PCs at six of the twelve vertices while the remaining six positions had weaker density consistent with partial (two-thirds) occupancy, for a total of ten PCs (Liu and Cheng, 2015). Even though no symmetry was imposed during the initial reconstruction process, the structure had approximate D₃ symmetry. If D₃ symmetry is applied to the former structure (Zhang et al.), this recapitulates the latter structure (Liu & Cheng).

It is important to note that it is, in theory, possible for two reconstructions from the same noise-free data to yield different structures due to the intrinsic limitations of iterative projection-matching (Sigworth, 2016)(Punjani et al., 2017). First, such an algorithm will converge in a finite number of iterations, but not necessarily to the best possible solution. This is particularly common in pseudosymmetric structures. It is therefore possible that the map of Liu & Cheng is consistent with the underlying data but not the correct structure of the virion. Second, where multiple structures are present in a dataset and a subset of particles are selected, it is unclear how one can assess the degree to which the subset is structurally representative of the whole dataset. It is therefore alternatively possible that the map of Zhang et al. is only one of several discrete structures co-existing that, when averaged together (or selected differently), would generate the map of Liu & Cheng. While experimenters can take some steps to mitigate the chances of convergence to a suboptimal structure such as using improved global search algorithms (Punjani et al., 2017), refining multiple times with several initial models, and obtaining the highest-possible signal-to-noise ratio in the underlying experiment, new methods (such as those described here) will be useful to this field to be employed in difficult cases.

Both manuscripts describing the PC arrangement in CPV concluded that ten PCs are present in each virion (Zhang et al., 2015)(Liu and Cheng, 2015). It is possible that the number of PCs is determined by the number of dsRNA segments. This would be required if, as has been previously speculated, each genome segment is specifically associated with one polymerase throughout the life of the particle (Zhang et al., 2003)(Liu and Cheng, 2015). Resolving the PC arrangement by cryoEM in additional

Reoviridae with nine, eleven, and twelve genome segments will address whether the number of PCs is determined by the number of dsRNA segments, and uncover the spatial arrangement of PC packaging and RNA replication.

In order to shed light on the spatial arrangement of PCs in Reoviridae, we chose to use FAKV as an experimental system for three main reasons. FAKV has nine genome segments, an uncommon number for this family, so its PC arrangement can be contrasted with viruses having 10–12 segments. Second, due to reductive evolution FAKV contains fewer virion polypeptides than any other member of its family (Auguste et al., 2015), an advantage for cryoEM of the PCs because scattering from the icosahedrally-symmetric proteins can obscure the PCs. Third, a purified preparation of FAKV contains a roughly 50/50 mixture of full (RNA-containing) virions and empty capsids with little or no RNA genomic material (Figure 1A). The empty capsids are analogous to “top component” in preparations of mammalian orthoreovirus type 3 Dearing (Smith et al., 1969), and of several other viruses in this family (Fang et al., 2008)(Zhang et al., 1999). Studies of “top component” in other species have revealed that top component is produced during normal infections (Chen et al., 2011) and that the polypeptide composition of top component particles is identical to that of infectious particles. In FAKV, The ratio of genome-filled to genome-free particles is unchanged by varying EDTA or cationic concentrations, suggesting cationic concentrations were not directly responsible for genome ejection (Auguste et al., 2015). Here, we describe the asymmetric structures of both full virions and empty particles of FAKV, and further exploit the absence of genomic RNA to make accurate occupancy measurements in the empty particles.

Figure 1.

Cryoelectron microscopy of FAKV. (A) Representative, contrast-inverted micrograph of FAKV. Full virions (indicated with a yellow arrow), empty (genome-free) particles (indicated with a cyan arrow), and free RNA (indicated with a magenta arrow) can be seen. (B) Icosahedral reconstruction of full virions of FAKV. One asymmetric unit is colored to show the individual subunits: yellow for MCP-A, cyan for MCP-B, purple for turret, and red for clamp.

In this study, we uncovered the arrangement of PCs in FAKV by asymmetric reconstruction. We then constructed a series of structural hypotheses that could describe the dataset of FAKV particles and tested which of those hypotheses matched data the best, which represents an alternative avenue for obtaining structural knowledge as compared to 3D reconstruction of averaged particles. With this method we showed that FAKV particles are best described by a unique arrangement of 10 PCs (as opposed to a mixed population). This arrangement shared with other viruses of this family.

Results.

Cryoelectron microscopy

FAKV preparations consisted of a mixture of virions (containing genome) and empty particles that contain little or no genomic material (Figure 1A). Boxed particles were sorted by hand into “empty” and “full” types, which were reconstructed independently of each other. The gold-standard resolution of the reconstructed maps, with icosahedral symmetry imposed, were 3.9Å for full virions (EMD-7944)(Figure 1B) and 4.5Å for empty particles (EMD-7945). We intend to describe elsewhere the particular details of the icosahedral map, as they are not pertinent to the PC arrangement we describe herein. Inside the virion, blurred density corresponding to the viral RNA-dependent RNA polymerase and NTPase proteins is visible in the icosahedral reconstructions of both empty and full particles. These are not well resolved in that map as the PC does not obey icosahedral symmetry, existing in less than 60 copies per capsid.

Structure of the polymerase complex

An asymmetric reconstruction of FAKV was performed by traditional symmetry-breaking (Supplementary Figure 1)(Kaelber et al., 2017). First, the dataset was refined with icosahedral symmetry enforced to determine particle positions and other parameters to high precision (Supplementary Figure 1A). To go from an icosahedral model to a C₁ (asymmetric) model, each particle was permitted to rotate to one of the 60 icosahedrally-related orientations only, with all other parameters being fixed (Supplementary Figure 1B) (Liu et al., 2010)(Jiang et al., 2006a). Symmetry-breaking was performed by iterative refinement of full (RNA-containing) particles, starting with a random model. Data converged to a structure with RNA visible and eight PCs well resolved (deposited as EMD-7954). This structure (EMD-7954) had three mutually perpendicular two-fold axes of symmetry, which is to say D₂ symmetry. We then enforced D₂ symmetry to increase the resolution, and consistently aligning particles were selected and reconstructed with a gold-standard FSC=0.143 at 6Å for empty particles (deposited as EMD-7948)(Figure 2A,B) and 5.5Å for full virions (deposited as EMD-7949)(Figure 2C–D)). In both the unsymmetrized and symmetrized maps, in addition to the eight well-resolved PCs (which presumably have a roughly D₂ organization), density for a PC could be seen at low threshold at the remaining four vertices (Figure 2A–D). These PC densities occupied two (of the possible five) orientations (Figure 2A, starred, and Figure 2B, green boxes) at these four vertices. Since only one PC can bind per vertex because of steric hindrance, detecting two PCs per pentamer in the structure means the PC is binding in one orientation in some particles and another orientation in other particles. As discussed below, FAKV is only approximately D₂-symmetric, and using new techniques we later obtained a fully asymmetric map of empty particles; however, this D₂-symmetrized map of full virions provided the highest resolution of the internal structure of the PC.

Figure 2.

Symmetry-breaking reconstruction of FAKV with D₂ symmetry applied. (A) After masking away the capsid density, PCs can be seen in the 6Å reconstruction of empty capsids. An icosahedron is displayed in color as a visual guide; the four equatorial vertices contain lower-intensity partially occupied positions. The mutually perpendicular axes of D₂ symmetry are shown as arrows. The partial-occupancy, equatorial PC sites are starred. (B) After masking away the capsid, the D₂-symmetrized reconstruction of empty particles is flattened onto a plane by unwrapping first in longitude and then in latitude. Density is viewed from the outside and colored arbitrarily by height. The partialoccupancy, equatorial PC sites are highlighted with green boxes. (C) The interior of full virions filtered to 10Å is shown as in panel A. Arrows are omitted for clarity but the orientation is identical. (D) The reconstruction of full virions is unwrapped onto a plane as in panel B. We identify as dsRNA the tubular densities present in full virions but not empty particles as dsRNA. (E) The D₂-symmetrized map and icosahedral map of empty virions were filtered to 10Å resolution, masked to focus on the capsid region, and subtracted to form a difference map showing areas of higher density in the D₂ map (dark blue) or the icosahedral map (dark green). The capsid is overlaid in faint red for perspective and viewed down the fivefold axis of symmetry. The isosurface threshold used to display differences is less than half the threshold necessary to enclose all atoms in the pre-subtraction maps (F) Architecture of the PC as seen in the 5.5Å reconstruction of full virions. Shown are the RNA-dependent RNA polymerase (blue), the accessory NTPase (green), and the laterally-bound RNA (gray). The laterally-bound RNA is not observed in empty particles. (G) Zooming in on the template entry channel, we fit deposited atomic coordinates of transcribing (yellow) or quiescent (red) CPV (Zhang et al., 2015) and display the bracelet domain in a ribbon representation.

Though the overall map resolution is measured at 5.5Å, there is considerable variation in local resolution. For example, though the RNA major groove is a lowerresolution feature than α-helices, the major grooves are not apparent in the RNA helices of this map while helices are clear in the polymerase protein. This is likely because the dsRNA is not as well ordered as the PC. As in dsDNA viruses, nonspecific interactions with the phosphate backbone probably dominate interaction between each RNA double helix and neighboring molecules—a globally ordered asymmetric genome (as in ssRNA Leviviridae) is unlikely (Kaelber et al., 2017).

At the site of polymerase attachment there are deviations from icosahedral symmetry in the MCP, but no asymmetries are observed in the distal, β-rich domain of MCP that forms the border between decamers or at any other site among the capsid proteins (Figure 2E). Because the local asymmetry induced by binding of a PC to an MCP decamer does not propagate to neighboring decamers, the global arrangement of PCs in the capsid is likely orchestrated by molecule(s) other than the capsid polypeptides.

Secondary structural elements (but not side chains) can be differentiated in the PC within the 5.5 Å map. The overall structure of the polymerase (Figure 2F) is similar to prototypical members of the family (Tao et al., 2002). We purified virions in the presence of EDTA and without cofactors such as nucleotides and S-adenosylmethionine, so these virions are transcriptionally inactive. Unlike its closest homolog (Li et al., 2017), the polymerase of FAKV does not refold its bracelet domain for autoinhibition of polymerization (Figure 2G).

Seven tubular densities skirt the surface of the polymerase in the full virion. Because they are present in the full virion (Figure 2D) but absent from the empty particle (Figure 2B), these must correspond to viral RNA. Of particular interest to the discussion below is a well-resolved RNA we term the “laterally-bound RNA” that follows a positively-charged groove running perpendicular to the substrate exit channel (Figure 2F). This double-helix has been observed in CPV and interacts with R997 (sometimes referred to in the paper as R979) of that virus (Zhang et al., 2015). While this particular residue is not conserved, the charged grove is present, with contributions from R811, R980, R985, K1163, and most especially R1035 (H1031 is positioned to be a member of the charged groove as well; the local charge environment of this histidine is not known).

Arrangement of the polymerase complexes

While the PC was well resolved in the D₂-symmetrized map (Figure 2A–D), two map features show that the map does not represent a unique, global structure of PCs within the virion. First, the intensity of PCs varied abnormally from position to position (Figure 3A). If PCs are consistently present or absent at each position, the intensity distribution should be bimodal, whereas if PCs are always present/absent at certain positions but choose randomly between another subset of positions (Liu and Cheng, 2015), we expect a trimodal distribution. In contrast, we could not discern any pattern in the actual distribution of intensities (below we will describe a fully asymmetric structure that does exhibit such a bimodal pattern [Figure 3B]). Second, a low-occupancy “ghost” was seen at low thresholds at four (D₂-related) of the high-occupancy positions (Figure 2A–D, 3A). In addition, we were innately skeptical of any solution with a high degree of symmetry (such as D₂) because a symmetric configuration often represents a local minimum in structure refinement by projection-matching when the underlying complex is pseudosymmetric (Liao et al., 2014)(Cong et al., 2012)(Guo et al., 2013). We therefore sought a new approach to determine the PC arrangement.

Figure 3.

The arrangement of PCs inside FAKV. (A) The occupancies of the 60 potential PC sites in the standard symmetry-relaxation reconstruction are not bimodal. Scores from both half-maps are shown in blue and red, respectively. (B) In the final map obtained by the decoy method, the occupancies of the 60 potential PC sites are bimodal. There are ten occupied and 50 unoccupied sites, and occupancies are higher in the putatively occupied sites compared to panel A. Scores from both half-maps are shown in blue and red, respectively. (C) Ten decoys were constructed by placing PCs randomly at 8 of 60 possible sites. For each of the ten, the fraction of particles aligning better to that decoy than any other decoy is shown with a gray bar. The PCs alone of each decoy are displayed overlaying the corresponding bar, with an icosahedron for reference. One of the ten random decoys reflected the underlying data better than the other decoys did. (D) After masking away the capsid density, ten PCs can be seen in the reconstruction from empty particles aligned to the final-round decoy. An icosahedron is displayed in color as a visual guide; two icosahedral vertices contain no PC.

In a symmetry-breaking reconstruction, useful signal comes only from the portions of the structure that differ between icosahedrally-related orientations and are consistent throughout the population. Other signal is detrimental, but over the course of iterations these unwanted signals may dominate. For instance, the mass of the capsid is large compared to the PC, so in naïve implementations of iterative asymmetric reconstruction we often generate incorrect structures with more intense capsid density on one side and less intense on the other: “symmetry-breaking artifacts” (Ludtke et al., 1999). At this point in the experiment, we had obtained experimentally-determined density maps for the PC and for the capsid shell, which are the only components in empty particles. We combined the map of the capsid and several copies of the map of the PC in various orientations to create a series of synthetic maps, which we term “decoys.” Among these decoys, we intend that the best decoy will “lure” the particles; the particles will match the best synthetic map better than the particles will match the competing synthetic maps. It is conceivable to enumerate every combination of PC orientations and score them by comparing to raw data; due to the computational burden (the number of combinations is in the quadrillions, though the vast majority are degenerate) we used a random subset as a starting point. This approach (Supplementary Figure 2) is not biased by the presence of random variations in the level of capsid signal or the presence of additional densities that are randomly distributed in the capsid. This is because references for projection are not generated directly from the underlying particle data but by recombining constant maps in a quantized set of possible configurations. Consequently, unlike other techniques, noise does not propagate through iterations.

We first generated ten decoys containing eight PCs each. A random number generator determined in which of the 60 admissible positions each PC would be placed, subject to the constraint that no two PCs could have a steric clash (i.e. max one PC per five-fold vertex). To test the agreement between the experimental data and the various decoys, every particle was rotated to the sixty icosahedrally-related orientations and the best cross-correlation coefficient (ccc) was saved. We evaluated each random decoy by computing the fraction of particles that matched it better than any other decoy (Figure 3C). The proportions differed significantly (p<0.001) from parity. One random decoy clearly outperformed the other random decoys (Figure 3C). Given that the number of possible PC arrangements exceeds the number of random decoys tested by many orders of magnitude, we did not interpret this to mean that the outperforming decoy corresponds to the true arrangement. However, we could begin aligning to this random starting point and bootstrap to a final structure. We aligned each particle to that decoy and reconstructed a 3D map. The resulting structure was not identical to the input random decoy; some PCs were in different locations but nevertheless well-resolved. This shows that the decoy was sufficient to provide a unique orientation for particles, which allowed recovery of some information not present in the input model. We generated a second-round decoy using the resulting structure to select a subset of the 60 possible PC locations, aligned each particle to that decoy, and reconstructed a 3D map. We repeated this process until the decoys did not change. The resulting map (deposited as EMD7941) had a PC arrangement like that of CPV in that the ten highest-occupancy sites in this map corresponded to the ten PC positions in CPV.

To initialize the reconstruction using the “decoy” method, we had constructed ten random decoys and selected the one that best matched the boxed particle data. To examine the sensitivity to initial conditions, we performed the iterative reconstruction again starting from the second-best of the random decoys, or from the worst of the random decoys. In both cases, convergence took three iterations as opposed to two iterations when starting from the best decoy. However, the final structure was identical in all cases. This indicates that initial model bias is unlikely to have influenced our result in this instance. The use of decoys was key to obtaining the final structure: if after any iteration we use the raw map as the reference map for subsequent projection-matching instead of using a decoy based on that map as a reference, then a local minimum is quickly achieved in the refinement. This behavior has been previously observed by other groups; one way it has been dealt with is by heavy filtering and re-refinement to remove accumulation of undesirable features (Zhang et al., 2015). The methods required to obtain an asymmetric reconstruction depend on several factors including data quality and the relative contributions of the asymmetric and symmetric components (Dai et al., 2016), so we cannot make a general statement as to the necessity or sufficiency of these techniques in other cases.

To create a final map for analysis we aligned all particles to a decoy with the PC arrangement to which each of the three runs had converged. We reconstructed a map from the particles (Figure 3D). Applying D₂ symmetry to this map recapitulated the D₂symmetric structure we previously obtained (Figure 2A), even unto differences in occupancy from site to site and “ghosts.” This is consistent with our claim that the D₂symmetric structure produced by iterative refinement from random initial models was stuck in a local minimum produced by pseudosymmetry. Our final (asymmetric) decoy matched the data better than a D₂-symmetric decoy, a decoy based on the arrangement of mud crab reovirus (Matthew Baker, personal communication), an icosahedrallysymmetrized decoy, and all random decoys tested.

FAKV PCs bind at ten sites

We measured the occupancy at the 60 icosahedrally-related possible PC sites by masking a region unique to each site (Figure 3A–B). First, we created a mask that encloses the entire PC. The mask was rotated to each of the five symmetry-related positions about the fivefold axis of symmetry and any voxel that could also be found in one of these rotated copies was removed from the mask. In this way, a mask was produced that contains density from only one PC orientation without contribution from icosahedrally-related orientations. The final mask was rotated to each of the sixty icosahedrally-related positions and applied to the map. The mean intensity within the mask was recorded for each position. A histogram of the intensity of the 60 sites is clearly bimodal (Figure 3B) with ten sites of approximately equal, high occupancy and fifty unoccupied sites; no low-intensity “ghosts” are present. Maps reconstructed along the path to the true asymmetric structure of FAKV also showed evidence for ten PC sites. First, the D₂-symmetric structure obtained from random initial models had eight well-resolved PCs and four PCs at an occupancy of roughly 0.25 each. Second, the map reconstructed from the best random decoy contained ten PCs.

Despite these three lines of evidence (occupancy value in the final map, occupancy distribution in the final map, and visualization of intermediate maps), because FAKV has only nine dsRNA segments, we wondered if it might have a unique structure with nine of twelve possible PC locations occupied. We created ten more decoys, each missing a different one of the ten PCs from our final map (Figure 4A; the PCs from each decoy are displayed along the x-axis). When particles are aligned to these decoys and reconstructed, the PC missing from each decoy should reappear in the output map if the PC is genuinely present (except that one decoy has perfect three-fold symmetry, so the PC should reappear as a low-intensity “ghost” at each of three symmetry-related sites). If there are only nine PCs in FAKV, one of the PCs should not reappear. All ten PCs reappeared in the correct place in this experiment (in the decoy with threefold symmetry, the tenth PC was spread over three symmetry-related sites).

Figure 4.

Comparison of how well the individual viral particles imaged match various decoys representing possible PC configurations. (A) Ten decoys of nine PCs each were constructed by removing one PC from the final structure shown in Figure 3D. For each of the ten, the fraction of particles aligning better to that decoy than any other decoy is shown with a gray bar, while a dashed line indicates the null hypothesis of parity amongst all the decoys. The PCs alone of each decoy are displayed over the corresponding bar and are colored: orange and red for locally three-fold symmetric PCs in a tip-to-tail configuration, purple for the three PCs whose RNA entry channels point towards the red PCs, and blue for the PC whose RNA entry channel points toward an orange PC. (B) Four decoys were constructed representing all possible nondegenerate configurations (of ten PCs each) that, when symmetrized, give 2/3 occupancy at the six equatorial vertices. The expected frequency of these decoys under a random distribution of PCs among possible sites is shown in light gray, while the fraction of particles actually matching each decoy better than the other three is shown in dark gray. The PCs alone from each decoy are displayed under the corresponding bar and colored as in (A). Note that the leftmost decoy is equivalent to the structure shown in Figure 3D.

Furthermore, if there exists a unique configuration of nine PCs in FAKV, one of these decoys should match the raw data better than the others. Nine of the ten decoys were indistinguishable at the level of agreement with the particle data and the threefold-symmetrical one was slightly worse (Figure 4A). This is the expected outcome if there are ten equal-occupancy sites (all other factors being equal, in the presence of noise it is predicted that fewer particles would align to the decoy with an additional symmetry axis). As this method is able to distinguish between decoys differing only by a single PC, the best explanation for the equivalence of these nine decoys is that all ten PC locations we found have equal occupancy.

These data do not preclude the possibility that in FAKV each of the ten PC sites has an equal 1/10^th chance of being unoccupied in any given particle. In an attempt to address this possibility, we performed multi-reference 3D classification to sort the particles over all ten of the aforementioned decoys, each of which is missing one of the ten PCs from our final structure. For each of the ten classes, we reconstructed only the particles that were assigned to that class. For each class, all ten PCs reappeared in the predicted location, with the exception of the threefold-symmetric decoy as expected. In the case of noise-free data and perfect decoy models, this would rule out the possibility that most FAKV particles contain nine PCs with sites selected from the ten PC sites we have established. For real-world data, it is still possible that enough particles were misclassified in each class as to yield this result.

Additionally, we created sixteen decoys by adding two more PCs to fill the two empty vertices in our final ten-PC decoy. There are 25 possible decoys (5 orientation choices to the power of two PCs) but we excluded the single orientation from each PC that would bring the structure closer to pseudosymmetry. When the particles were aligned to any of these decoys and reconstructed, the resulting map contained ten PCs as the two PCs added to the decoy were absent in the resulting map.

There is a unique arrangement, not a mixed population, of the ten PC sites

Comparing the two published asymmetric structures of CPV raised the question of whether there exists a unique arrangement of ten PCs or several similar arrangements that, when averaged together, create a structure with D₃ symmetry (vide supra). Of the <di>64=15</di> possible combinations of PCs that, when symmetrized, would give rise to the published D₃ structure, four are nondegenerate arrangements (Figure 4B; the four nondegenerate arrangements are displayed along the x-axis). If each of the 15 possible combinations were equally likely, as from purely stochastic incorporation of PCs over the available sites, the ratio of the four nondegenerate arrangement would be 2:1:1:1 (Figure 4B, light gray bars), with the arrangement of our final map twice as common as the other arrangements. We created a decoy representing each of these ten-PC arrangements. We aligned FAKV empty particles to each. The observed proportions differed significantly (χ² p<0.001) from the expected proportions under a stochastic arrangement. By comparison, in the dsRNA phage φ6 of the family Cystoviridae PC components are distributed randomly over potential sites (Nemecek et al., 2010). Among the four FAKV decoys, one outperformed the others; this decoy had the PC arrangement of our final map, or in other words the PC arrangement of the asymmetric published CPV structure (Figure 4B, dark gray bars). Also, we applied D₃ symmetry to this decoy: the unsymmetrized version matched the data better than the symmetrized version (not shown). While it is possible that there are minor subpopulations that differ in PC arrangement, there exists a unique structure that accounts for the PCs in a typical virion.

Discussion.

A hypothesis-testing approach to single-particle cryoEM datasets

The most common use of cryoEM data is to iteratively align and average particles to obtain a density map. Data can be classified into multiple subsets and obtain a number of maps, perhaps representing multiple states or conformations—a very active area of methods development (Kaelber et al., 2017). However, projection-matching to initial models does not guarantee convergence to a global minimum structure for any class or for the distribution of particles over classes (Sigworth, 2016). In practice, classes of interest can often be recovered even without prior knowledge, but the assignment of any individual particle to a given state is dependent on initial conditions. We attribute the differing structures deposited for asymmetric CPV particles to a known, intrinsic difficulty with iterative projection-matching methods: that in some cases, more than one self-consistent, converged, and validated structure is possible.

Here, we constructed synthetic “decoy” models differing only in the arrangement of PCs to explicitly evaluate whether individual particles reflected one model better than another: “structure-testing” rather than “structure-generating.” When all possibilities can be enumerated (Figure 4), the discrimination on a per-particle basis need not be large to produce a robust difference at the level of the dataset.

Previously it has been shown that a single PC can be detected within a virion of this family with 71% sensitivity even in the presence of the genomic RNA (Rickgauer et al., 2017). Detection was highly dependent on signal-to-noise ratio. We make use of genome-free particles, decreasing the “noise” coming from the strong scattering of the RNA. The standard symmetry-breaking technique (iterative symmetry-relaxation projection matching, Supplementary Figure 1) yielded the same PC arrangement (Figure 2A–D) in empty particles and genome-containing virions (even the low-occupancy “ghosts”), giving us confidence that the PC arrangement we determined for empty particles should reflect the arrangement in virions. All components of empty particles other than the PC are icosahedrally symmetric and we determined their structure to atomic resolution; we account for this density and hold it constant in our references. To simplify the calculations, in this study we restricted head-to-head comparisons to decoys with the exact same number of PCs (and thus the same total mass, spatial frequency distribution, et cetera). With appropriate normalizations, spectral prewhitening (Rickgauer et al., 2017), and/or an alternate scoring function, this could be extended to a wider range of problems.

Polymerase anchoring

We determined the arrangement of PCs within FAKV, a nine-segmented member of the family Reoviridae (deposited as EMD-7941). Each PC lies under one of the twelve fivefold capsid vertices. These vertices are far enough apart that PCs at neighboring vertices cannot touch one another, no matter what the orientation is. Yet, a global arrangement of PCs exists (Figure 3D). This arrangement is not patterned by the capsid alone because the capsid does not propagate any conformational changes from one PC site to another; it is icosahedrally symmetric in the area between PC sites. The global arrangement of PCs may be patterned by viral RNA and/or nonstructural proteins.

Nonstructural proteins are absent from virions, and empty particles also lack RNA inside the capsid. The PC arrangement of empty particles is nevertheless identical to that of RNA-containing virions. In this respect the Reoviridae and sister dsRNA family Cystoviridae differ: the RdRp’s of bacteriophage φ6 is bound underneath the threefold axis of the P1 shell before dsRNA packaging (Ilca et al., 2015) but leave these sites and may relocalize to a central cavity within the spooled genome (Ilca et al., 2019). We conclude that the global arrangement of PCs in FAKV is stable even after the patterning molecule(s) are removed, probably due to tight binding between the polymerase and MCP. It is possible that RNA is involved in patterning the PC arrangement, as we have not excluded the possibility that empty particles contained RNA at some point in the past and then lost it. Whether or not the RNA was required to first form the arrangement, we can conclude that RNA is not necessary to maintain it.

Common features in the PC arrangement

We note two patterns when considering the arrangement of a PC with respect to its neighbors. The laterally-bound RNA (Figure 2F) always faces another laterally-bound RNA or an empty vertex. The NTPase side of the PC, which protrudes furthest from fivefold axis, is never oriented toward a neighboring NTPase. No fewer than three dsRNA helices are positioned between neighboring PCs, but if two PCs were oriented with NTPases facing each other there would not be room for three dsRNA helices. These two patterns also hold for PCs in mud crab reovirus (Matthew Baker, personal communication). Perhaps they minimize RNA bending energy or the potential for clashes, and may play a role in the global arrangement of PCs. The nine-segmented FAKV shares with ten-segmented cytoplasmic polyhedrosis virus (Zhang et al., 2015)(Liu and Cheng, 2015) and eleven-segmented grass carp reovirus (Ding et al., 2018) a common, pseudo-D₃ supramolecular architecture of PCs. There is agreement between all experimenters on the two antipodeal rings of three PCs, and on the orientation of PCs at the medial positions (their NTPases point toward the rings). FAKV and grass carp reovirus architectures differ only in number of medial PCs. FAKV, CPV, and grass carb reovirus are all members of the subfamily Spinareovirinae, the turreted reoviruses. Future studies should assess whether these rules hold true for the non-turreted reoviruses, subfamily Sedoreovirinae. Although it includes several important pathogens, no asymmetric structures are available for members of this subfamily.

Number of polymerases: implications for packaging

Efficient packaging of RNA segments is critical to maintaining virion infectivity. Although recent experiments have revealed several aspects of packaging in Reoviridae, the overall mechanism in Reoviridae is not yet clear. RNA segments vary in size and sequence but are packaged in equimolar amounts (McDonald et al., 2016) as ssRNA. The ssRNA are thought to pre-form a supramolecular complex (Fajardo et al., 2016; Sung and Roy, 2014) that is incorporated into the capsid, where the complementary strand is synthesized (Hay and Joklik, 1971)(Long and McDonald, 2017). Virion assembly regulates polymerase function: the polymerase is autoinhibited until bound to the underside of the capsid (Long and McDonald, 2017). At least for rotavirus, an RNA segment is associated with one polymerase at a time during transcription (Periz et al., 2013). The polymerase of human picobirnavirus, a segmented dsRNA virus from another family, is not encapsidated in the absence of RNA (Collier et al., 2016), though the same cannot be said for all Reoviridae. In one straightforward model, Reoviridae RNA segments each carry a single PC into the assembling virion and remain associated with that PC for the lifetime of the virion. That the number of RNA segments equals the number of PC binding sites in CPV and in the 11-segmented grass carp reovirus (Ding et al., 2018) was taken as additional support for this hypothesis (Zhang et al., 2015). FAKV is closely related to CPV and shares its PC arrangement. FAKV has nine segments in the genome, but the PCs bind at ten sites. In this work, we could not determine with confidence whether these ten sites had 100% or 90% occupancy. If further studies confirm that there are exactly ten PCs in Fako virus or another Dinovernavirus, this would indicate that the number of genome segments is not the only factor specifying the number of PCs per virion in Reoviridae. Beyond number, the arrangement of the PC binding sites is nonrandom and is conserved with respect to other family members. In light of the common architectural motifs among dsRNA Reoviridae with varying numbers of genome segments, we conclude factors beyond the number of segments contribute to directed encapsidation of the polymerase complex. Among the possible encapsidation determinants are: viroplasm proteins such as homologs of rotavirus NSP5 or NSP2 (Jiang et al., 2006b), specific genomic RNA sequences or structures (which may be present zero, one, or multiple times in a given segment), or the steric constraints of copackaging RNA and PCs in a virion. Applying the methods presented here to mutated strains, virions assembled under perturbed conditions, or assembly intermediates should clarify the identities and roles of determinants of PC arrangement in virions of this family.

STAR Methods

Lead Contact and Materials Availability

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jason Kaelber (jason.kaelber@rutgers.edu). This study did not generate new unique reagents.

Experimental Model Details

Fako virus

FAKV type strain CSW-77 was propagated on C6/36 Aedes albopictus cells as previously described (Auguste et al., 2015). We recently sequenced strain CSW-77 (Auguste et al., 2015) and the nucleotide sequences of the nine dsRNA segments are available as NC_025485.1 through NC_025493.1.

C6/36 cells

C6/36 Aedes albopictus cells were originally obtained from ATCC; CRL-1660. Sequencing of C6/36 cells infected with FAKV (Auguste et al., 2015) confirmed the species of origin and did not reveal the presence of any contaminating virus besides FAKV. Cells were propagated in DMEM media (Gibco) containing 10% FBS, penicillin (100 U/ml), streptomycin (100 μg/ml), 1% non-essential amino acids, and 1% tryptose phosphate broth, and after infection, cells were maintained in DMEM media with 5% FBS, penicillin (100 U/ml), streptomycin (100 μg/ml), 1% non-essential amino acids, and 1% tryptose phosphate broth. Cells were maintained at 28°C with 5% CO₂.

Method Details

Virus purification and vitrification.

C6/36 cells infected with FAKV were freeze-thawed thrice and clarified by centrifugation. The supernatant and pellet were processed separately as both contained virus. From the supernatant, virus was precipitated overnight with 7% PEG and 2.3% w/v NaCl. Meanwhile the pellet was resuspended and centrifuged at 80,000×g for 60 minutes. The pellet was resuspended in TEN buffer (10 mM Tris-HCl, 1 mM EDTA, 100 mM NaCl, pH 7.8) and emulsified with Freon by vortexing. After separating the emulsion by centrifugation, virus was pelleted from the aqueous phase by centrifugation at 80,000×g for 60 minutes. Pellets from overnight PEG-precipitation and Freon extraction were resuspended in TEN buffer and combined for sucrose centrifugation, carried out as described previously (Auguste et al., 2015). In pilot experiments the amount of virus precipitated from the culture supernatant was comparable to the amount liberated from the cell pellet by Freon extraction. Grids were prepared as previously described (Auguste et al., 2015) by blotting in a Vitrobot Mark IV.

Cryoelectron microscopy.

Grids were imaged using a JEM3200FSC cryoelectron microscope. Micrographs were recorded mostly on a DE-20 camera at 1.071Å/pix. Some micrographs were recorded on a K2 summit camera at 1.284Å/pix. These were resampled to 1.071Å/pix by Fourier padding and combined with the preponderance of the data. Motion correction and damage filtering were performed on a per-particle basis (Bammes et al., 2013). Around 2,400 micrographs had visible CTF oscillations in the high-resolution regime and were used for further processing. 10,588 full particles and 9,772 empty particles were boxed manually from these micrographs. With icosahedral symmetry enforced, the position, orientation, defocus, and astigmatism of each particle was determined by iterative refinement with JSPR (Guo and Jiang, 2014), a program built on the extensible EMAN2 background (Tang et al., 2007). Even though particles were acquired on two different detectors, per-micrograph scale refinement (Yu et al., 2016) did not improve the resolution as measured by gold-standard FSC, so refined scales were not used.

Asymmetric refinement.

Full and empty particles were refined separately throughout. The icosahedrallysymmetric shell was subtracted from the particles (Supplementary Figure 1B). For each refinement iteration, the sixty icosahedrally-equivalent orientations of each particle were compared to the model to determine the best orientations. At the end of each iteration, the volume was filtered to 20Å and a mask was applied to the reconstructed volume to preserve only a spherical region under each five-fold vertex (the putative PC location). Though the aforementioned reconstructions were done without symmetry, the resulting map appeared by eye to have D₂ symmetry. The refinement procedure was repeated six times in parallel. Only particles whose orientation was consistent among the repetitions (standard deviation of orientation ≤ 1°) were kept. D₂ symmetry was then applied to the final map. In a difference map between the D₂-symmetrized and unsymmetrized maps, no discernable features were present, reinforcing the fact that asymmetric refinement had indeed converged to a D₂ solution before any symmetry was applied.

Scoring agreement to PC arrangement models on a single-molecule basis.

The best 50% of each half-set was selected based on phase residual (vs. the icosahedrally-symmetrized map) and used for asymmetric refinement. The particle position, defocus, and astigmatism were fixed throughout refinement and the orientation was allowed to vary only among icosahedrally-equivalent Euler angles (Guo and Jiang, 2014). A PC was segmented from the D₂-symmetrized map using Chimera (Pettersen et al., 2004). To initialize with random decoys, the PC was copied into eight of the sixty symmetry-related positions using EMAN2 symmetry operations (Baldwin and Penczek, 2007). The eight positions were chosen by a random number generator subject to the constraint that no two positions could have steric clash. The eight PC copies were added to the icosahedrally-symmetrized capsid shell and filtered to 10Å resolution to create a “decoy” (Supplementary Figure 2). This procedure was repeated ten times to create ten random decoys. Other decoys were constructed by selecting a defined set from among the possible sixty orientations of the PC. To generate symmetric models, the corresponding asymmetric decoys were symmetrized in real space. In all cases the agreement between each particle and each decoy was measured by scoring the best cross-correlation coefficient among the sixty possible particle orientations. Because the capsid features are icosahedrally symmetric in the model, they do not contribute to differences in cross-correlation between decoys and thus they obviate the need for capsid-subtraction of the particles.

To determine the asymmetric arrangement of PCs in FAKV, first particles were aligned to random decoys and reconstructed, then a nonrandom decoy was created based on the reconstruction. Nonrandom decoys for later rounds were generated by selecting PC positions based the reconstructed map of the previous round. The 60 possible positions were overlaid onto a given reconstructed map and those containing substantial density were recorded. Where two PC positions overlapped, i.e. there were two of five sites at a given five-fold vertex both showing density, we chose whichever was not used in the previous round’s decoy. Particles were aligned to the second decoy and a map reconstructed from which a third decoy could be created until the decoy did not change from round to round (convergence). The icosahedral capsid and PCs at each recorded position were summed in real space to generate a new decoy. The number of PCs in a decoy changed from round to round based on the observed density in the reconstructed map. For fastest convergence, the first-round decoy should have sufficient asymmetric information that a noise-free particle would align in a unique orientation. Therefore, the number of PCs placed in that should be closer to 12 (the max possible) than to 1. Among possible numbers, eight was chosen arbitrarily. For construction of the decoy based on mud crab reovirus, the mud crab reovirus map was overlaid with all 60 possible PC maps in UCSF Chimera and the matching PCs were noted by eye.

Occupancy determination.

A single PC was segmented from the empty particle asymmetric map. This PC was turned into a tight mask by applying a binary threshold. The mask was rotated to each of the five symmetry-related positions about the fivefold axis of symmetry and any voxel that could also be found in one of these rotated copies was removed from the mask. In this way a mask was produced that contains density from only one PC orientation without contribution from icosahedrally-related orientations. The final mask was rotated to each of the sixty icosahedrally-related positions and applied to the map. The mean intensity within the mask was recorded for each position.

The map density values are related to the scattering potential of whatever is within the mask by some linear transform, though that transform may be unknown and depends on the implementation of reconstruction software. We subtracted the background intensity, estimated as the mean within a spherical mask of 144 Å radius; this radius does not enclose any known protein or other density in the empty particle. The resulting intensity units should be proportional to occupancy.

Quantification and Statistical Analysis

To evaluate whether particles matched equally well to all random decoys (n=10), proportions were compared to a null hypothesis by Pearson’s χ ² test.

Data and Code Availability

The cryoEM density maps were deposited in the EM Data Bank. Accessions are: EMD7941, asymmetric structure of FAKV empty particles solved using the “decoy” method; EMD-7944, icosahedrally-symmetrized map of FAKV full particles by classical projectionmatching; EMD-7945, icosahedrally-symmetrized map of FAKV empty particles by classical projection-matching; EMD-7948, D₂-symmetrized map of FAKV empty particles by symmetry-breaking, EMD-7949, D₂-symmetrized map of FAKV full particles by symmetry-breaking, EMD-7953, unsymmetrized map of FAKV empty particles by symmetry-breaking; EMD-7954, unsymmetrized map of FAKV full particles by symmetrybreaking. The coordinates of the icosahedral components of FAKV with accession code 6DJY correspond to cryoEM map EMD-7944.

Additional Resources

None

Highlights.

• RNA is not required to maintain the arrangement of polymerases inside Fako virus

• 10 polymerase complex sites are laid out in an architecture shared with other species

• Although it has 9 genome segments, there are 10 polymerase complex sites in Fako virus

• CryoEM particles can be analyzed by comparing them to synthetic maps