letters

868 nature structural biology • volume 8 number 10 • october 2001

Electron cryomicroscopyand bioinformatics suggestprotein fold models for ricedwarf virusZ. Hong Zhou1,2, Matthew L. Baker2,3, Wen Jiang2,3,Matthew Dougherty3, Joanita Jakana3, Gang Dong4,Guangying Lu4 and Wah Chiu2,3

1Department of Pathology and Laboratory Medicine, University of Texas–Houston Medical School, Houston, Texas 77030, USA. 2Program inStructural and Computational Biology and Molecular Biophysics and3National Center for Macromolecular Imaging, Verna and Marrs McLeanDepartment of Biochemistry and Molecular Biology, Baylor College ofMedicine, Houston, Texas 77030, USA. 4National Laboratory of ProteinEngineering and Plant Genetic Engineering, College of Life Sciences, PekingUniversity, Beijing, 100871, China.

The three-dimensional structure of rice dwarf virus wasdetermined to 6.8 Å resolution by single particle electron cry-omicroscopy. By integrating the structural analysis withbioinformatics, the folds of the proteins in the double-shelled capsid were derived. In the outer shell protein, theuniquely orientated upper and lower domains are composedof similar secondary structure elements but have differentrelative orientations from that of bluetongue virus in thesame Reoviridae family. Differences in both sequence andstructure between these proteins may be important in defin-ing virus–host interactions. The inner shell protein adopts a

conformation similar to other members of Reoviridae, sug-gesting a common ancestor that has evolved to infect hostsranging from plants to animals. Symmetry mismatchbetween the two shells results in nonequivalent, yet specific,interactions that contribute to the stability of this largemacromolecular machine.

Rice dwarf virus (RDV) is a major pathogen of the rice plantsin Southeast Asia, where the spread of RDV can have severe eco-nomic consequences. RDV, belonging to Reoviridae, is transmit-ted by Nephotettix cincticeps leafhoppers to susceptible plantspecies, replicating in both hosts and vectors. Viruses in theReoviridae, whose hosts include plants, insects and animals,share a common mechanism of replication1,2 but vary substan-tially in their structural organization3.

RDV, a double-shelled particle containing 12 segments ofdsRNA, has a total protein mass of >26 MDa. Previous efforts instudying the RDV by electron cryomicroscopy at low resolutionhave illustrated the structural organization4–7. The outer capsidshell contains 260 trimers of P8 (46 kDa) arranged on a T = 13licosahedral lattice. The inner capsid shell is composed of 60dimers of P3 (114 kDa) arranged on a T = 1 icosahedral lattice.Although this capsid organization is the same as that of the dou-ble-shelled particles of several other members of the Reoviridae,the sequence similarity to P3 and P8 is insignificant (identity<20%)4.

Overall structureWith recent improvements in the data acquisition and process-ing of electron cryomicroscopy images8, determining the struc-ture for RDV to a significantly higher resolution has becomepossible. The integration of sequence analysis and computation-al tools9 has allowed us to identify the secondary structure ele-

a b c

Fig. 1 Structure determination of RDV. a, Representative area from amicrograph of RDV embedded in vitreous ice at –170 °C. Images weretaken at 50,000× magnification in a JEOL 4000 electron cryomicro-scope with an electron dose of 10–13 electrons Å–2. Individual parti-cles were ‘boxed’ out into an array of 300 × 300 pixels. b, Powerspectrum. The computed power spectrum was obtained by summingthe Fourier transforms of ∼ 100 particle images from the micrographshown in (a) with an underfocus of 0.6 µm. c, Resolution assessment.The effective resolution consistently improved with increasing num-bers of particles in the data set and reached 6.8 Å for the data setwith 3,261 particles. d, Shaded surface representation of the 6.8 Åresolution density map of the outer capsid shell (P8) of RDV. The massdensities corresponding to inner shell proteins and RNA were compu-tationally removed. The outer shell is composed of 132 trimers of P8organized on a T = 13l icosahedral lattice. Each asymmetric unit con-tains four 1/3 unique trimers4, including one P (red), Q (orange), R(green), S (yellow) and 1/3 of T (blue). e, Surface representation ofthe inner capsid shell computationally extracted from the 6.8 Å reso-lution map. The inner shell consists of 60 P3 dimers, each containingone A (light green) and one B (light purple) P3 subunit. The maps arecontoured at ∼ 2× σ (standard deviation) above the average density.

d e

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

nature structural biology • volume 8 number 10 • october 2001 869

ments of both P3 and P8 and, consequently, construct a struc-tural model for the double-shelled RDV capsid.

The computer power spectrum of a representative area from aclose-to-focus micrograph of RDV (Fig. 1a) containeddetectable contrast beyond 6 Å (Fig. 1b). The effective resolutionof this reconstruction was assessed to be 6.8 Å (Fig. 1c), fromwhich the structures of the outer and inner capsid shells werecomputationally segmented (Fig. 1d,e). Although the overallorganization of the double-shelled capsid is similar to that of thetranscriptionally competent particles of bluetongue virus(BTV)10 and animal reovirus11, which are circular, RDV is dis-tinctly angular.

Outer capsid and P8 modelThe outer shell of each RDV capsid is made of five uniquetrimers (P, Q, R, S and T) (Fig. 1d) of protein P8 organized ona T = 13l icosahedral lattice. Although only the T trimer hasicosahedral three-fold symmetry imposed during the recon-struction, all five trimers appeared very similar to each other.

Additionally, all of the trimers correlated well with each otherand displayed a high degree of three-fold symmetry with acorrelation coefficient >0.85. Thus, to improve the signal-to-noise ratio, the P, Q, R and S trimers were aligned and aver-aged. The averaged trimer is ∼ 70 Å in height and ∼ 70 Å indiameter, with a monomeric subunit consisting of an upperand lower domain twisted ∼ 60° with respect to each otherabout the three-fold axis (Fig. 2a). The upper domain of eachP8 subunit primarily contains thin, flat and contiguous densi-ties, whereas the lower domain is mainly composed of rod-likedensities (Fig. 2b).

Using HELIXHUNTER9, we identified and localized ninehelices in the lower domain of P8 (Fig. 2c). To identify the cor-responding amino acid sequences to these helices, we correlat-ed their relative spatial locations and lengths of helicalsegments with those predicted from multiple secondary struc-ture prediction methods (PsiPred12, PHD13 and SSPRO14).Although the locations of the nine consensus helices identifiedby secondary structure prediction in the sequence of P8

a b

c d e

Fig. 2 Fold model of the outer shell protein P8. a, Side view of an averaged trimer contoured at ∼ 4 σ with a single subunit highlighted in blue. The mol-ecular boundaries between adjacent subunits were determined by interactively examining the continuity of mass densities at a high density threshold.b, Stereo view of the P8 monomer. The wire frame represents a low threshold (4 σ), whereas the solid density represents a high threshold (6 σ), illus-trating the ability to identify structural features at this resolution. The view direction is ∼ 180° from (a) about the three-fold axis. The three layers of β-sheets are evident in the upper domain of the P8 monomer. c, P8 subunit and helices identified by HELIXHUNTER9. A computationally isolated P8 sub-unit is shown with a wire frame representation. Assigned helices from HELIXHUNTER, shown as 5 Å diameter cylinders, are superimposed on themonomer. Seven of these helices (green) were identified with high degree of confidence, whereas two relatively short helices (orange) had low HELIX-HUNTER correlation values. Helix 5 and 8 (orange) were fairly short, and the density in this region was not well resolved, as judged by a low HELIX-HUNTER correlation value. d, P8 subunit and a ribbon diagram of the homologous β-sandwich motif from BTV VP7. The location of the homologousfold was determined by FOLDHUNTER9. e, A proposed model for the fold of P8. A black line denotes the boundary of the P8 subunit. Helices are repre-sented using the same criterion as described in (c), whereas strands of the sheets are displayed as blue arrows. The β-strands are modeled based on thearrangement of strands from BTV VP7 and, thus, are speculative. Solid red lines represent the connectivity of the helices, whereas the dashed red linesrepresent the speculative β-strand connections. The asterick in (e) refers to the putative hinge region between the upper and lower domain of P8.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

870 nature structural biology • volume 8 number 10 • october 2001

(Table 1) are consistent, the exact length of the predictedhelices varies by up to seven residues. The assignment ofsequence to the putative helices from the HELIXHUNTERanalysis (Table 1) was proposed based on the match of helixlength and continuity of the mass density between sequentialhelices at various contour levels (Fig. 2e). In the N-terminus,five helices (1–5) and their connections could be assigned.However, in the C-terminal helices, the mass densities of onlyhelices 6, 7 and 9 are contiguous. Helix 8 was left unassigned,because it was poorly resolved in sequence and structure aswell as not connected to the densities at the C-terminal region.This helix assignment was further supported by the match inthe sequential ordering to that of the helices in the BTV outercapsid protein VP7, as assessed by two commonly used meth-ods of comparing protein structures9,15,16. For instance, six of

the helices in RDV P8 (helices 1, 2, 4, 6, 7 and 9) matched well,with <5 Å root mean square (r.m.s) deviation based on helicalcentroids, to the corresponding helices in the BTV VP7(Table 1). The N-terminal helices appeared visually to matchthe corresponding BTV helices better than those of the C-terminus, suggesting a domain movement in regards to theN- and C-terminal domains.

The secondary structure prediction methods also showedthat the middle region of the P8 sequence (128–315) con-tained β-strands. However, the variability of the predictionswithin this region is considerably higher than the helix-richterminal regions. Using fold recognition17, this region waspredicted to be structurally homologous to the β-sandwichdomain of the BTV VP7 (ref. 4). Although individual β-strands cannot be resolved in our 6.8 Å map, the flat, con-

Fig. 3 Fold model of the inner shellprotein P3. a,b, P3 and HELIX-HUNTER9 results. The identifiedhelices, represented as cylinders of5 Å diameter, are shown superim-posed on the surface views of P3Aand P3B structures. Length, resolv-ability, position, connectivity and rel-ative agreement with consensussecondary structure prediction wereused to qualitatively judge the fideli-ty of helix assignment. The less reli-able helices are colored in orange.Helices 13, 14 and 15 were betterordered in P3A than P3B. Helix 20 wasobserved only in P3A, whereas helix21 was resolved only in P3B. Helices inthe dimerization domain (helices 27,29 and 30) were in regions dominat-ed by β-sheets and loops and, thus,more speculative in assignment.c,d, Proposed folds for P3A and P3B,respectively. The β-sheets are repre-sented by light blue parallelograms.The red lines represent the connectiv-ity between the secondary structureelements. Note that some of the con-nectivities through the β-sheets aredrawn as dotted lines because theindividual strands are not resolved.The numbers on the helices corre-spond to those identified in (g).e,f, Representative P3A densities.Regions from the β-sheet-rich dimer-ization domain (e) and the predomi-nantly helical carapace domain (f) areenlarged and rendered using twothresholds: 2 σ for the wire frame and3 σ for the solid density. g, P3 sec-ondary structural prediction usingSSPRO14. Shown is the assignment ofthe predicted helices with respect tothe identified helices in our 3D map.The bracketed region (red) refers tothe insertion sequence compared tothe corresponding region in VP3 ofBTV.

a

b

c

d

e

f

g

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

nature structural biology • volume 8 number 10 • october 2001 871

tinuous shapes in the upper domain exhibit features typical ofβ-sheets at this resolution (Fig. 2b). Using FOLDHUNTER9,the homologous β-sandwich fold was localized to the upperdomain of the RDV P8 subunit (Fig. 2d). Although the overallmatch appears reasonable, an obvious mismatch in the outer-most portion of the P8 upper region can be mapped back to acorresponding difference between the β-sheet domain of RDVP8 and BTV VP7 protein sequences4. This structural differencein the outermost region of P8 suggests a possible involvementin specific host cell receptor recognition, attachment and/orentry.

From both sequence and structural analysis, we proposed afold model for RDV P8 (Fig. 2e). Although the placement ofthe strands and their connectivity within the upper domain isspeculative, the fold of the lower helical domain is substantiat-ed by the well-resolved helices, their connectivities evident at alower density contour, the agreement with sequence-based sec-ondary structure prediction and the match with the corre-sponding secondary structure elements from BTV VP7 (Table1). Although the two-domain architecture of P8 is similar toVP7 of BTV, the relative orientation of these domains is differ-ent from both structural isoforms of VP7: REF and HEX,which exist in two different crystal forms18,19. P8 appears to bein a conformation with less twist (60°) between the upper andlower domains than the REF isoform (120°) but more than theHEX isoform (0°) of VP7. This difference may be influenced byvariances in flexible segments within the hinge region (asterisk,Fig. 2e). Additionally, the subunit arrangement results in a fun-nel-shaped opening at the top of the trimer, which is in con-trast to the fully filled density on the top surface of thecorresponding trimer of BTV. This unique conformation mayfacilitate the subunit interactions required for capsid stabilitywhile maintaining the requisite surface for host receptor inter-actions.

Inner capsid and P3 modelUnderlying the trimers of the outer shell are 60 P3 dimers form-ing a T = 1 inner shell (Fig. 1e). The structures of the A and Bsubunits of P3 dimer in RDV were graphically segmented fromthe inner capsid. Both P3 A and B are essentially planar, 150 Ålong and 25–35 Å thick, with an overall crescent shape. P3 can

be partitioned into three domains, referred to as the apical(close to the five-fold axis), carapace (plate-like) and dimeriza-tion (close to the two-fold axis) domains in BTV10. A grossstructural similarity to other reovirus inner capsid proteins wasevident.

Using HELIXHUNTER9, 28 helices in each of the subunitsin the asymmetric unit of the inner shell were identified(Fig. 3a,b). However, the exact lengths and positions of thecorresponding helices in P3A and P3B varied slightly, reflect-ing the conformational variations between the subunits as alsoseen in both BTV and animal reovirus. As with P8, the threeaforementioned secondary structure prediction methodsyielded similar lengths and sequence locations of the predict-ed helices (Fig. 3g). The sequence-predicted helices were thencorrelated with those visualized in our structure as describedfor P8 (Fig. 3a,b). Additionally, four β-sheets were visuallyapparent in the dimerization domain of both subunits(Fig. 3e). One extra β-sheet was seen in the carapace domainof the A subunit.

An overall similarity between the arrangement of secondarystructure elements within P3A and P3B subunits to the corre-sponding subunits in BTV10 and animal reovirus core11 wasevident and, thus, facilitated the connection of the observedstructural elements (Fig. 3c,d). In addition, the connectivity ofthe helices is apparent in our map (Fig. 3f). The proposed con-nectivity (red line, Fig. 3c,d) between helices, based on the BTVand animal reovirus core proteins, is consistent with theobserved densities in our map, whereas the connections of the β-strands (dotted red line, Fig. 3c,d) are not well resolved and,thus, more speculative.

The observed differences in the detailed spatial arrangementof the secondary structure elements between P3A and P3B maybe related to their different functional and structural roles. Theapical domains of P3 A and B, although highly similar, exhibitdifferences in a stretch of three short helices (helices 13–15).These helices in P3A appear to be well ordered and located nearthe five-fold axis. However, the corresponding helices in P3Bare less resolved. The clustering of these P3A variable helicalregions about the five-fold axis results in the formation of apore along this axis (Fig. 1e). Thus, these helices are likelyinvolved in the nascent mRNA transcription and extrusion dur-

Fig. 4 Molecular interactions between the shells. a, A zoom-in view of the inner shell and the five unique trimers (color codes specified in Fig. 1) illus-trating the interaction surfaces between the inner and outer capsid. P3A is colored in green, and P3B is colored in purple. b, Same as in (a) but thetrimers are replaced with contour lines to reveal the contact surfaces of the trimers on the P3 molecules. c, Sites of interaction of P8 on the P3 A–Bdimer (gray). Regions of continuous density at relatively high threshold (>2 σ) between proteins are colored. Exclusive interactions between P3 andthe T and P trimer are colored in red and light blue, respectively, whereas multiple but specific interactions between P3 and the other trimers are col-ored in dark blue.

a b c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

872 nature structural biology • volume 8 number 10 • october 2001

all asymmetric units. The nonequivalent and specific interac-tions allow the T = 1 inner and T = 13l outer shells to maintaina unique and stable geometrical relationship throughout RNAreplication and release.

ConclusionOur analysis suggests the overall folds of the RDV capsid shellproteins appear similar among different members in the samevirus family3, even though they have no significant sequencesimilarity. This is indicative of the divergent evolution of theseviruses from a common ancestor in order to adapt to varioushosts and environments. By maintaining a similar capsid archi-tecture and protein structure, the specific interactions necessaryfor capsid stability can be maintained, whereas the structuraldifferences among the outer shell proteins may play key roles inconferring host specificity during infection.

The structure determination of RDV provides the firstdemonstration of deriving a high-resolution model for themolecular components in a large assembly without a crystal,through the integration of electron cryomicroscopy withbioinformatic tools. We showed that at 6.8 Å resolution,although only helices and β-sheets of proteins are discernable,establishing a high-resolution fold model is feasible. The con-struction of these models epitomizes the integration of infor-mation, spanning de novo structure assignment (P8 lowerdomain), domain localization through fold recognition (P8upper domain) and model-aided fold assignment (P3). Suchan integrated approach is generally applicable to other largebiological machines.

MethodsElectron cryomicroscopy. A JEOL4000 electron cryomicroscopewith a Gatan cryoholder was used to record focal pairs of purifiedRDV at 400 keV as described8. Individual micrographs were digitizedat 1.35 Å pixel–1 on a SCAI scanner (Z/I Imaging) and subsequentlyaveraged to 2.7 Å pixel–1. From 81 focal pairs, >4,000 pairs of RDVparticle images were selected for further analysis. Particle imagesfrom the far-from-focus micrographs were used to determine the ini-tial orientation parameters for the corresponding particles in theclose-to-focus micrograph8,22,23. The orientation and center parame-ters of each close-to-focus particle were iteratively refined asdescribed23,24. A contrast transfer function correction with an ampli-tude contrast of 7% and a Fourier amplitude damping factor of100 Å2 was applied to each particle image prior to 3D merging in

ing viral replication, a process common among dsRNA virusesoccurring near the five-fold axis2,20,21. The conformational flexi-bility of these helices could conceivably allow for the expansionand/or contraction of the pore necessary during and aftermRNA release.

Like the apical domain, the carapace domain helices are wellconserved between P3A and P3B. However, there is an obviousdifference in a region of the carapace domain involved in theintersubunit contact. In P3B, a 30 Å long helix (helix 1), alsoseen in an equivalent position in the animal reovirus11, pro-trudes from the back of the carapace domain. In P3B, this helixinserts into a pocket formed by helices 22–25 of P3A (Figs 1e,3c,d) but is not visible in the corresponding location in P3A,similar to both the corresponding inner capsid proteins of BTVand the animal reovirus core.

The dimerization domains have the least amount of similaritybetween P3A and P3B, and our model suggests that both arecomposed mainly of loops and β-sheets. The packing of theinner capsid subunits in conjunction with the potentially flexiblenature of this region might be responsible for the conformation-al differences.

Symmetry mismatchAs shown in other members of Reoviridae3, the primary func-tion of the inner shell is to protect the RNA and provideanchoring sites for enzymes to facilitate the process of endoge-nous transcription, whereas the outer shell mediates hostspecificity and plays important roles in capsid assembly andstability. In our structure, neither the P3 dimers nor the P8trimers have extensive contacts within their respective shells(Fig. 1d,e). However, the interactions between the inner andouter capsid proteins are substantial, because each outer capsidtrimer contacts multiple inner capsid proteins. The contactsites between the two shells are mainly localized to helices 2 and3 of P8 but are more widespread in P3 (Fig. 4). Specifically, theP trimer of P8 makes exclusive contacts with the apicaldomains of P3A and P3B (red, Fig. 4c), whereas the T trimermakes contacts with the dimerization domains of three neigh-boring P3B molecules about the three-fold axis (light blue,Fig. 4c). The remaining three trimers maintain multiple con-tact sites with several neighboring P3A and P3B molecules(Fig. 4b,c). These identified contact sites are conserved across

Table 1 Helix assignment in P81

HELIXHUNTER Length SSPRO Length PSIPRED Length PHD Length Matched(Å) (Å) (Å) (Å) BTV helices2

1 28.3 4–19 24.0 3–21 27.0 5–18 21.0 12 16.1 36–48 19.5 36–49 21.0 36–45 15.0 23 22.6 61–73 19.5 60–72 19.5 61–71 16.54 16.8 132–141 15.0 131–141 16.5 135–142 12.0 55 9.3 154–161 12.0 153–160 12.0 155–159 7.56 16.6 326–335 15.0 321–337 25.5 324–337 21.0 67 23.5 345–359 22.5 341–359 28.5 341–359 28.5 78 11.0 376–391 24.0 374–381 12.0 380–385 9.09 29.5 393–414 33.0 393–413 31.5 396–416 31.5 8

1Consensus results from three secondary structure prediction methods showing helix location and length (Å) are illustrated. Helix numbering wasbased on the sequence order. The resulting helices from HELIXHUNTER9 and their corresponding predicted helices by sequence analyses are listed.The assignment of the short helices, 5 and 8, is speculative because the connections between these helices and neighboring elements are not unam-biguous.2The final column shows the six matched helices between RDV and BTV outer capsid proteins from visual inspection, DejaVu16 and COESC15 analysis.Although helix 3 and 5 of RDV do not directly correlate to an analogous BTV outer capsid helix, they can be localized in our map. Helix 8 of RDV ismatched to a sequence region based only on the size and spatial location relative to the other helices; however, no direct connections to helix 8 canbe resolved within this region of our density map.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

nature structural biology • volume 8 number 10 • october 2001 873

Fourier space25. Only the close-to-focus particle images, from 0.3 to2.2 µm underfocus, were included in the final 3D reconstruction,which was computed by merging 3,261 particles25,26. The effective res-olution was determined with the 0.5 threshold criterion for theFourier shell correlation coefficient between two independent recon-structions27,28.

Structure analysis. From the density map, slightly larger than oneasymmetric unit was graphically extracted using custom-designedmodules built in IRIS Explorer (NAG). Definitions between the cap-sid proteins, as well as their contacts, were identified visually, allow-ing the geometrically distinct subunits to be further segmentedfrom the capsid. Each of the P8 trimers was aligned to the T trimerusing FOLDHUNTER9. The local three-fold symmetry axis within thetrimers was iteratively refined and the degree of three-fold symme-try was assessed about this axis29. The aligned trimers, excluding theT trimer, were then summed to produce an averaged trimer. A P8monomer was subsequently extracted from the averaged P8 trimer.HELIXHUNTER9 was then run on all capsid subunits and the P3monomers to identify potential helices. Furthermore, the β-sheetdomain of BTV VP7 was blurred to ∼ 7 Å resolution using PDB2MRC30

and localized to RDV P8 using FOLDHUNTER9. Secondary structureprediction using SSPRO14, PsiPred12 and PHD13 was done for both P3(SWISS-PROT ID VP8_RDVF) and P8 (SWISS-PROT ID VP3_RDVF). Theresults obtained from HELIXHUNTER9 and secondary structure pre-dictions were then correlated in order to map the sequence to theobserved helices. Assignment of the helices was based on length,location, density continuity and spatial arrangement. In P3, thestructures of the corresponding BTV and reovirus proteins wereused as references during the model building process. Approximateconnections between helices and the β-sheet regions were assignedbased primarily on visually identified connectivity of the secondarystructure elements in our density map and similarities to the BTVand reovirus structures.

AcknowledgmentsThis research has been supported by grants from National Institutes of Health,the Robert Welch Foundation and the National Natural Science Foundation of

China. Z.H.Z. is a Pew Scholar in the Biomedical Sciences and a Basil O’ConnorStarter Scholar of the March of Dimes Foundation. M.L.B. was supported in partby Baylor Research Advocates for Student. We would like to thank B.V.V. Prasad,M.F. Schmid and M. Baker for their helpful comments on the manuscript.Supplementary animations and 3D VRML models are available athttp://ncmi.bcm.tmc.edu/∼ wjiang/rdv/.

Correspondence should be addressed to W.C. email: wah@bcm.tmc.edu

Received 24 May, 2001; accepted 23 August, 2001.

1. Gouet, P. et al. Cell 97, 481–490 (1999).2. Lawton, J.A., Estes, M.K. & Prasad, B.V. Nature Struct. Biol. 4, 118–121 (1997).3. Nibert, M.L., Schiff, L.A. & Field, B.N. in Fields Virology, Vol. 2 (eds. Fields, B.N.

et al.) 1557–1596 (Lippincott-Raven, Philadelphia; 1996).4. Lu, G. et al. J. Virol. 72, 8541–8549 (1998).5. Zhu, Y. et al. J. Virol. 71, 8899–8901. (1997).6. Wu, B. et al. Virology 271, 18–25. (2000).7. Shao, C., Zhou, Z.H. & Lu, G. Science in China 44, 192–198 (2001).8. Zhou, Z.H. et al. Science 288, 877–880 (2000).9. Jiang, W., Baker, M.L., Ludtke, S.J. & Chiu, W. J. Mol. Biol. 308, 1033–1044 (2001).

10. Grimes, J.M. et al. Nature 395, 470–477 (1998).11. Reinisch, K.M., Nibert, M.L. & Harrison, S.C. Nature 404, 960–967. (2000).12. McGuffin, L.J., Bryson, K. & Jones, D.T. Bioinformatics 16, 404–405. (2000).13. Rost, B., Sander, C. & Schneider, R. Comput. Appl. Biosci. 10, 53–60. (1994).14. Baldi, P., Brunak, S., Frasconi, P., Soda, G. & Pollastri, G. Bioinformatics 15,

937–946. (1999).15. Mizuguchi, K. & Go, N. Protein Eng. 8, 353–362. (1995).16. Kleywegt, G.J. & Jones, T.A. Methods Enzymol. 277, 525–545 (1997).17. Fisher, D. & Eisenberg, D. Protein Sci. 5, 947–955 (1996).18. Grimes, J.M., Basak, A.K., Roy, P. & Stuart, D.I. Nature 373, 167–170 (1995).19. Grimes, J.M. et al. Structure 5, 885–893 (1997).20. Pesavento, J.B., Lawton, J.A., Estes, M.K. & Prasad, B.V. Proc. Natl. Acad. Sci. USA

98, 1381–1386. (2001).21. Zhang, H. et al. J. Virol. 73, 1624–1629. (1999).22. Fuller, S.D. Cell 48, 923–934 (1987).23. Zhou, Z.H. et al. Nature Struct. Biol. 2, 1026–1030 (1995).24. Zhou, Z.H. et al. Biophys. J. 74, 576–588 (1998).25. Zhou, Z.H., Chen, D.H., Jakana, J., Rixon, F.J. & Chiu, W. J. Virol. 73, 3210–3218

(1999).26. Crowther, R.A. Phil. Trans. R Soc. Lond. B Biol. Sci. 261, 221–230 (1971).27. van Heel, M. Ultramicroscopy 21, 95–100 (1987).28. Böttcher, B., Wynne, S.A. & Crowther, R.A. Nature 386, 88–91 (1997).29. He, J., Schmid, M.F., Zhou, Z.H., Rixon, F. & Chiu, W. J. Mol. Biol. 309, 903–914

(2001).30. Ludtke, S.J., Baldwin, P.R. & Chiu, W. J. Struct. Biol. 128, 82–97 (1999).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com