ABSTRACT
The structure of adenovirus was determined to a resolution of 6 Å by cryoelectron microscopy (cryoEM) single-particle image reconstruction. Docking of the hexon and penton base crystal structures into the cryoEM density established that α-helices of 10 or more residues are resolved as rods. A difference map was calculated by subtracting a pseudoatomic capsid from the cryoEM reconstruction. The resulting density was analyzed in terms of observed α-helices and secondary structure predictions for the additional capsid proteins that currently lack atomic resolution structures (proteins IIIa, VI, VIII, and IX). Protein IIIa, which is predicted to be highly α-helical, is assigned to a cluster of helices observed below the penton base on the inner capsid surface. Protein VI is present in ∼1.5 copies per hexon trimer and is predicted to have two long α-helices, one of which appears to lie inside the hexon cavity. Protein VIII is cleaved by the adenovirus protease into two fragments of 7.6 and 12.1 kDa, and the larger fragment is predicted to have one long α-helix, in agreement with the observed density for protein VIII on the inner capsid surface. Protein IX is predicted to have one long α-helix, which also has a strongly indicated propensity for coiled-coil formation. A region of density near the facet edge is now resolved as a four-helix bundle and is assigned to four copies of the C-terminal α-helix from protein IX.
Adenovirus (Ad) is a common etiologic agent of respiratory, gastrointestinal, and ocular infections. Ad vectors also have substantial potential for diverse gene therapy approaches to treat cancer and other chronic diseases, as well as promise for vaccine delivery (33). The lack of an atomic resolution structure of an intact Ad virion has hindered further development of practical applications, as well as limited our understanding of Ad cell entry. The mature Ad virion has an icosahedral protein capsid composed of three major capsid proteins (hexon, penton base, and fiber) and four additional capsid proteins (proteins IIIa, VI, VIII, and IX). The capsid is ∼926 Å in diameter, not including the fibers that vary in length from 120 to 315 Å depending on the serotype. The double-stranded DNA Ad genome is packaged inside of the capsid, together with four additional proteins (V, VII, mu, and terminal protein) and multiple copies of the Ad protease. Cryoelectron microscopy (cryoEM) structures of Ad and Ad vectors (12, 37), Ad/receptor (7), and Ad/antibody (46) complexes, combined with crystal structures for hexon (36), penton base (53), and fiber (45), have provided a wealth of structural information on Ad, but significant gaps remain in our knowledge of the virus capsid structure. In particular, the structure and locations of the capsid proteins IIIa and VI, which play important roles in the virus life cycle, are poorly defined.
Protein IIIa participates in viral assembly and maturation as several temperature-sensitive mutants that affect protein IIIa are defective for assembly (3, 8). In addition, protein IIIa likely plays a role in cell entry since it is one of the earliest components released from the virion and one of the most thermally labile capsid components (14, 32, 50). Protein IIIa is present in 60 copies per virion and is cleaved near the C terminus by the Ad protease during virus maturation to form a mature protein of 570 residues with a molecular mass of 63.5 kDa (20, 40). Early immunoprecipitation studies with anti-IIIa antibodies suggested an external location for a least one domain of protein IIIa (10). In contrast, argon plasma etching experiments and cross-linking studies pointed to an internal position for protein IIIa between the Ad capsid and core (10, 29). The first cryoEM difference map at ∼25-Å resolution tentatively assigned one domain of protein IIIa to density observed at the facet edge on the exterior of the capsid and a second domain to density on the capsid interior (40).
At least three distinct roles in Ad biology are served by protein VI, which is present in 369 ± 34 copies per virion (20). Protein VI is cleaved near both the N and the C termini by the Ad protease to form a mature protein of 22.1 kDa (20). The first biological role of protein VI is that the cleaved C-terminal peptide serves as a cofactor for the Ad protease (23, 49). Second, protein VI regulates hexon import into the nucleus during Ad assembly (51). Third, protein VI may participate in endosomal disruption after cell entry (50). During Ad cell entry the fiber is thought to be released while the virion is still at the cell surface (28), and further viral dissociation occurs in the acidified environment of the endosome (14), followed by disruption of the endosomal membrane and escape of the partially disassembled virion into the cytoplasm. Wiethoff et al. showed that protein VI has membrane lytic activity and is probably exposed upon partial disassembly of the Ad capsid, leading to subsequent endosome disruption (50). However, the conformational changes that the Ad capsid must undergo in order to release protein VI are not well understood.
Consideration of the capsid protein stoichiometry in the mature virion indicates one to two copies of protein VI per hexon trimer, with ∼369 copies of protein VI to 240 hexon trimers per virion. Direct association between protein VI and hexon has been demonstrated (24, 25). Protein VI is thought to be located on the inner capsid surface and has been reported to bind DNA (35). A genetic analysis of the Ad virus type 5 (Ad5) ts147 temperature-sensitive mutant, which has a defect in nuclear import of hexon, indicates that hexon residue 776 (which is mutated from G to D in ts147) is most likely a key residue in the hexon-protein VI interaction (51). This residue is within the cavity of the hexon trimer and is exposed to the interior of the virion. At nonpermissive temperatures the ts147 mutant accumulates newly synthesized hexon trimers in the cytoplasm, but the nuclear import of hexon, which is necessary for viral assembly in the nucleus, is impaired (19). A second phenotype of ts147 at nonpermissive temperatures is that the precursor to protein VI is unstable and is degraded. This latter phenotype could also explain the defect in hexon import into the nucleus as protein VI shuttles between the nucleus and the cytoplasm and links hexon to the importin α/β-dependent nuclear import pathway (51).
The locations of protein VIII and IX within the virion are better characterized than proteins IIIa and VI, but their structures have not yet been determined at high resolution. A cryoEM analysis of Ad particles at ∼10-Å resolution has provided a convincing assignment for protein VIII on the inner capsid surface (12). Protein VIII is known to be associated with hexons and is present in ∼120 copies per virion (40). Comparison of the cryoEM structures of wild-type Ad5 and a mutant virus lacking protein IX that had undergone single or multiple freeze-thaw cycles enabled Fabry et al. to observe that there are two similar density regions per asymmetric unit on the inner capsid surface (12). These density regions were approximately the correct volume for protein VIII and in the correct copy number, which helped to support the assignment. Mass spectrometry indicates at least two Ad protease cleavage sites in the protein VIII precursor, leading to 7.6- and 12.1-kDa protein fragments in the mature virion (20). The precursor form of protein VIII has been found to be useful in reversed-phased high-pressure liquid chromatography analyses to quantitate the percentage of empty capsids in Ad preparations (42, 47). This implies that the mature Ad virion contains only the two cleaved fragments of protein VIII, with no remaining copies of the precursor form of protein VIII.
The main locations of protein IX within the virion were deduced by scanning transmission electron microscopy (STEM) of capsid dissociation fragments (13). Protein IX is present in 240 copies per virion, and four trimers of protein IX are known to stabilize the group of nine hexons in the center of each icosahedral facet. Protein IX is a relatively small capsid protein of 14.4 kDa, but it has gained prominence as a platform for ligand addition for the purposes of vector retargeting and fluorescence labeling (30). Recent studies have shown that only the conserved N-terminal domain of protein IX (amino acids [aa] 1 to 39) is necessary for stabilization of the Ad capsid (34, 48). A cryoEM study at 9-Å resolution from our laboratory indicated that the locations identified by STEM for protein IX are likely to correspond to only the N-terminal viral interaction domains (37). For the predicted C-terminal α-helix of protein IX, two conformations were proposed, either binding on the capsid surface or extending away from the capsid.
The Ad35F vector selected for this structural study transduces human hematopoietic cells via association of its fiber with CD46 (37). The Ad35F vector is comprised of an Ad5 capsid pseudotyped with an Ad35 fiber and has the normal tropism of Ad35. CD46 is a member of the family of complement regulatory proteins and an attachment receptor for several microbial pathogens as well as for a subset of Ad serotypes. The 6-Å resolution cryoEM reconstruction of Ad35F presented here allows visualization of α-helices within the Ad capsid, as demonstrated by docking of the hexon and penton base crystal structures into the cryoEM density. Secondary structure prediction for the Ad capsid proteins without atomic resolution structures (IIIa, VI, VIII, and IX), combined with visualization of α-helices within the capsid, has led to new and more precise assignments for regions of proteins IIIa, VI, VIII, and IX. The implications of these structural results for Ad capsid disassembly during cell entry and the release of protein VI inside cell endosomes are discussed.
MATERIALS AND METHODS
CryoEM imaging.The Ad35F sample was prepared as described previously (39). CryoEM grids were prepared as described by Saban et al. (37). The data collection was performed on an FEI Polara (300 kV; FEG) electron microscope at liquid nitrogen temperature and 300 kV with a Gatan UltraScan 4kx4k charge-coupled device camera as described for dataset 1 in reference 37. Datasets 1 and 2 were collected with an absolute magnification of 253,654×, and dataset 3 was collected with an absolute magnification of 397,878×. Leginon (41) was used for semiautomatic collection of dataset 2. Datasets 1, 2, and 3 contained 1,511, 4,657, and 2,015 micrographs, respectively. The refined pixel sizes are 0.589 Å for datasets 1 and 2 and 0.375 Å for dataset 3 on the molecular scale without binning.
Image processing.The entire data set consisted of 6,880 particle images. Dataset 1 particle images were selected manually with QVIEW (38) as described in reference 37. An automated particle picking program, Virus (1), was used for datasets 2 and 3. Particle images were first binned to 2502 pixels for the initial rounds of refinement, subsequently binned to 5002 pixels for image processing with a ∼2-Å pixel size (2.36 Å for datasets 1 and 2; 2.25 Å for dataset 3), and later binned to 7502 pixels for image processing with a 1.5-Å pixel size on the molecular scale. The program CTFFIND3 (27) was used to determine the initial defocus and astigmatism parameters for each particle image. The previously published 9-Å resolution reconstruction of Ad35F (37) was used as a starting model for image processing with the FREALIGN v.7 program (15). The FREALIGN code was modified to allow the input of angular and translational step sizes for refinement and the input of externally determined particle centers. The IMAGIC-5 (44) centering algorithm, Align, was used to determine initial particle centers.
Multiple rounds of refinement were performed with FREALIGN, first with refinement of the five orientational parameters (three angular and two translation) and later with additional refinement of the defocus and astigmatism parameters. The orientational angles were refined to within 0.0005° and X- and Y-centers were refined to within 0.0025 pixels. No inverse temperature factor was applied to the data. During refinement, the particle images were masked with an outer radius of 463 Å, which effectively cut off the fibers. For Fourier shell correlation (FSC) resolution assessment, inner and outer masks at radii 300 and 463 Å were applied to prevent the inclusion of density from the disordered core and flexible fibers. Refinement was performed on two SGI Altix computers, each with 16-processors and 32 GB of RAM and on the Vampire 64-bit Linux cluster at Vanderbilt. The final reconstruction was calculated from 2,630 particle images.
Generation of the pseudoatomic capsid and difference map was performed as described previously (37), with additional refinement of the microscope magnification to within ±0.25%. Specifically, a c-shell script (Gen_diffmap) was written to perform a complete difference mapping analysis. First, Gen_diffmap directed the docking of the Ad5 hexon (36) and Ad2 penton base (53) coordinates into subvolumes of the cryoEM map using the automated fitting tool, CoLoRes, in the Situs package (5). Gen_diffmap selected the best solutions corresponding to the four unique hexon positions within the asymmetric unit. The coordinates for a penton base monomer were extracted from the best solution for the pentameric form of penton base. Icosahedral symmetry was applied to the asymmetric unit to yield a pseudoatomic capsid of 240 trimeric hexons and 12 pentameric penton bases. The pseudoatomic capsid was converted into a density map and filtered to a 6.9-Å resolution with the pdb2mrc routine in EMAN v1.7 (21), normalized to the cryoEM density with the IMAGIC command Norm-Variance (44), and subtracted from the cryoEM map with the IMAGIC command Subtract-Images in order to generate a difference map. The difference map should contain density for the minor capsid proteins IIIa, VI, VIII, and IX, as well as for a small part of the fiber associated with the penton base and regions of hexon and penton base missing from the crystal structures. Lastly, the Gen_diffmap script extracted one-eighth (an octant) of the cryoEM reconstruction, the pseudoatomic capsid, and the difference map for visual inspection with the UCSF Chimera package (31). Molecular graphics images were produced by using Chimera.
Density segmentation, fitting of α-helices, and identification of protein interaction residues.The volume eraser function of Chimera (31) was used for manual density segmentation. Canonical α-helices were built with Swiss PDB viewer (16) in lengths predicted by the PsiPred web server (4) for the capsid protein without atomic resolution structures. The canonical α-helices were manually positioned into the cryoEM density with Chimera. Potential protein-protein interaction residues between the penton base and the helical cluster assigned to protein IIIa were evaluated with Chimera. Penton base residues within 10 Å of the helical cluster density were selected.
RESULTS
CryoEM structure at 6-Å resolution reveals α-helices.Our previous cryoEM structure of Ad35F at 9-Å resolution provided new information about the capsid proteins, particularly for protein IX, but a complete assignment was not possible for the additional capsid proteins (37). Acquisition of more particle images (6,880 versus 2,215 total), inclusion of more particle images in the final reconstruction (2,630 versus 964), and improvement in the image processing methods has enabled us to extend the resolution and enhance our interpretation of Ad structure (Fig. 1). The resolution of the new Ad35F reconstruction as assessed by the conservative 0.5 FSC threshold is 6.9 Å (Fig. 1b). Information extends to 6.1 Å, as indicated by the 0.3 FSC threshold.
CryoEM structure of Ad35F at 6-Å resolution. (a) Surface view along a twofold symmetry axis of the reconstruction. The density is radially color coded (red = 596 Å; blue = 316 Å). The flexible fibers were masked during image processing, and only a short portion of the fiber is reconstructed. A rotating view of the Ad35F structure is presented in Movie S1, part 1, in the supplemental material. (b) An FSC plot indicating a resolution in the range of 6.1 to 6.9 Å, at the 0.3 and 0.5 thresholds, respectively. (c) Hexon (PDB 1P30) (36) and penton base coordinates (PDB 1X9T) (53) for one asymmetric unit of the capsid after docking within the cryoEM density. The penton base monomer is yellow, the hexon in position 1 is green, position 2 is cyan, position 3 is blue, and position 4 is magenta. (d) Pseudoatomic capsid calculated by applying icosahedral symmetry to the asymmetric unit. Scale bars, 50 Å.
The 6-Å resolution map enabled a more accurate fitting of the atomic resolution structures of the Ad5 hexon (36) and Ad2 penton base (53) within the cryoEM density with the CoLoRes quantitative docking tool in the Situs software package (5). The Ad35F vector contains the Ad5 capsid proteins pseudotyped with the Ad35 fiber, and thus the Ad5 hexon crystal structure is the correct serotype and fits well within the cryoEM density. The Ad2 penton base is highly homologous to the Ad5 penton base (98% identity) and also fits well within the cryoEM density. During fitting of the hexon crystal structure the magnification was refined to within ±0.25%, whereas at 9-Å resolution the magnification could only be refined to within ±2.0% by this method. After docking the hexon and penton base crystals structures into the cryoEM density, we generated coordinates for the asymmetric unit of the Ad capsid, including four independent hexon trimers and one monomer of the penton base (Fig. 1c). Application of icosahedral symmetry to the asymmetric unit produced a pseudoatomic model for the Ad capsid containing all 240 copies of the hexon trimer and 12 copies of the pentameric penton base, representing over 5.2 million nonhydrogen atoms (Fig. 1d).
The 6-Å resolution reconstruction is significant in that α-helices longer than 10 aa are now resolved within the icosahedral capsid. The backbone of a 10-residue α-helix has nearly three complete helical turns, and this is long enough to produce a rod-like shape that is recognizable at 6-Å resolution (18). Surface representations of the major capsid proteins penton base and hexon also reveal finer detail than observed in the earlier 9-Å resolution map (Fig. 2a and 3a). The Ad35 fiber was previously noted to be flexible (37), and therefore it was masked during image reconstruction. Thus, only a short portion of the fiber shaft is visible protruding from the penton base in the reconstruction (Fig. 2a). The crystal structure of penton base (53) indicates that the monomer has two α-helices longer than 10 aa, corresponding to 10 long α-helices in the pentamer (Fig. 2b). These penton base α-helices are clearly resolved as density rods in the cryoEM reconstruction (Fig. 2c, d, and e).
Penton base α-helices resolved in the cryoEM reconstruction. (a) Surface representation of the penton base with a short portion of the protruding fiber indicated by a white bracket. Scale bar, 50 Å. (b) Penton base coordinates (PDB 1X9T) (53), with the α-helices of 10 or more residues shown as helical ribbons. The ribbon structure is rainbow colored with the N terminus in blue and the C terminus in red. The arrow indicates the position of the plane in panel c. (c) Crop plane through penton base density showing that the two long α-helices per monomer (PDB 1X9T) are resolved as rods within the cryoEM density (mesh). (d) A perpendicular crop plane through the penton base showing the agreement of the cryoEM density with the α-helices in the 1X9T penton base crystal structure (with a fiber peptide). (e) The same crop plane as in panel d but with the cryoEM density aligned with the 1X9P penton base crystal structure (without fiber). The arrowhead indicates the conformational switch region affected by fiber binding for one penton base monomer, which includes the α-helix (aa 482 to 491) shown in gold. A rotating view of the penton base/fiber complex and the fit of the penton base crystal structure (1X9T) into the cryoEM density are presented in Movie S1, parts 2 and 3, in the supplemental material.
Hexon α-helices resolved in the cryoEM reconstruction. (a) Surface representation of a hexon, in position 4 of the asymmetric unit, with the sharp protrusions of the HVR4 region (aa 251 to 256) indicated by arrows. Scale bar, 50 Å. (b) Enlarged side view of the hexon tower density (mesh) with the docked hexon trimer coordinates (PDB 1P30) (36) (cyan) and a model for loop residues (aa 251 to 256) (red). Scale bar, 25 Å. (c) Hexon trimer coordinates with the α-helices of 10 or more residues shown as helical ribbons. The ribbon structure is rainbow colored, with the N terminus in blue and the C terminus in red. The arrows indicate the positions of the four planes in panel d. (d) Crop planes through a position 4 hexon showing that all of the moderately long α-helices in the hexon trimer are resolved as rods within the cryoEM density (mesh).
Two alternative crystal structures of the penton base have been determined: one in complex with an N-terminal fiber peptide (PDB 1X9T) and another without fiber (PDB 1X9P) (53). A conformational switch was noted in penton base residues 482 to 505, which includes one of the two moderately long α-helices. The cryoEM density of penton base is clearly more consistent with the conformation of 1X9T (with fiber peptide) than that of 1X9P (without fiber) (Fig. 2d and e), which is as expected since the Ad35F vector includes the Ad35 fiber. This result demonstrates the level of accuracy with which the 6-Å resolution cryoEM structure of Ad35F can be interpreted.
Surface representations of hexon in the 6-Å resolution cryoEM reconstruction (Fig. 3a) show a sharp protrusion for the HVR4 hypervariable region (36). The hexon crystal structure is missing six residues (aa 251 to 256) in the HVR4 region due to disorder (36). The cryoEM density suggests that these residues form an elongated loop protruding from the side of the hexon. The hexon PDB coordinate file was submitted to the CODA protein loop prediction web server (9), and several reasonable model loop structures were returned. The top ranked model generated by CODA fit best within the cryoEM density (Fig. 3b). The crystal structure of the hexon indicates that each monomer has six α-helices longer than 10 aa, corresponding to 18 long α-helices in the trimer (Fig. 3c). The agreement between the cryoEM density rods and the α-helices in the crystal structure is shown for a hexon in position 4 of the asymmetric unit (Fig. 3d). In addition, all of these α-helices are resolved as rods within each of the three other independent hexons in the asymmetric unit (data not shown). These results for the hexon and penton base demonstrate that the resolution achieved for this cryoEM structure is sufficient to reliably identify α-helices within the Ad capsid. The results also provide confidence for interpretation of the α-helices observed in the remaining cryoEM density for the additional capsid proteins.
Secondary structure predictions for the additional capsid proteins.In addition to hexon, penton base, and fiber, the Ad capsid also contains proteins IIIa, VI, VIII, and IX. At present, no atomic resolution structural information is available for these four proteins. Difference mapping, or subtraction of the pseudoatomic capsid of the hexon and penton base from the cryoEM reconstruction, should reveal the density for the additional capsid proteins if they are packaged with icosahedral symmetry. The observation of α-helices within the cryoEM reconstruction opens up the possibility of assigning density to the additional capsid proteins based on their predicted α-helical content, as well as their known copy numbers and expected capsid locations.
PsiPred secondary structure prediction (4) was performed for proteins IIIa, VI, VIII, and IX of Ad5, since the capsid of the Ad35F vector is comprised of Ad5 proteins pseudotyped with the Ad35 fiber. Only predicted α-helices of 10 or more residues are reported here, since these are the α-helices that we expect to resolve at 6-Å resolution. Protein IIIa is likely to be highly α-helical, with 14 α-helices in the N-terminal two-thirds of the sequence (aa 1 to 400) and 2 α-helices in the C-terminal third (aa 401 to 570). The first half of the mature form of protein VI is also likely to be α-helical, with two predicted α-helices of 17 and 43 residues. Submitting the full-length sequence of the precursor to protein VIII to the PsiPred server indicates one long α-helix of 24 residues in the 12.1-kDa fragment (aa 1 to 111 of the precursor) and no long α-helices in the 7.6-kDa fragment (aa 158 to 227 of the precursor). For protein IX the PsiPred server indicates no α-helices for the viral interaction domain (aa 1 to 39), two possible α-helices of 10 and 20 residues in the alanine-rich low complexity region (aa 60 to 90) in the middle of the sequence, and one long α-helix of 40 residues (aa 92 to 131) in the C-terminal half of the sequence. The COILS, PAIRCOILS, and MULTICOILS web servers (2, 22, 52) all indicate a strong propensity for coiled-coil formation for the C-terminal α-helix of protein IX. In contrast, the coiled-coil web servers indicate that coiled coils in proteins IIIa, VI, and VIII are unlikely.
Additional protein density on the capsid exterior.In our earlier 9-Å resolution difference map we observed four trimeric regions of density per icosahedral facet and also a large density lobe near each facet edge on the exterior of the capsid (37). The trimeric density regions were confidently assigned to protein IX, since these sites had previously been shown by STEM to be occupied by protein IX (13). More precisely, volume analysis indicated that the trimeric regions corresponded to only the N-terminal virus interaction domains of protein IX. The density lobes observed at the facet edges were tentatively assigned as a domain of protein IIIa, in accord with an earlier cryoEM interpretation made at ∼25-Å resolution (40). This assignment for protein IIIa was based mainly on the fact that the copy number of these lobes was in agreement with that of protein IIIa, with 60 copies per virion (20, 40), and also on an early observation that at least a portion of protein IIIa might be exposed on the outside of the capsid (10).
The 6-Å resolution difference map also shows the four trimeric density regions in the middle of the facet and density lobes at the facet edges (Fig. 4a). The trimeric density regions now display more structural detail. There is no indication for α-helices within these regions, which is consistent with the PsiPred secondary structure prediction for the N-terminal region of protein IX. Therefore, we interpret each trimeric region to be three copies of the N-terminal region of protein IX possibly up to residue 56, which is just before the first predicted α-helix.
Protein density on the exterior of the capsid. (a) A region of the cryoEM difference map (red) and pseudoatomic capsid (colored as in Fig. 1c) roughly corresponding to one icosahedral facet. Four trimeric density regions and three helical bundles are observed. Density from missing residues in the hexon and penton base crystal structures has been removed except for the hexon HVR4 loops (pink) that connect to the helical bundles. (b) Enlarged views of the helical bundle shown in two views related by a 180° rotation. The density is displayed with a lower isosurface in the 180° view to show the weak density corresponding to the fourth α-helix. The connected hexon HVR4 density is shown in pink. Four α-helices, each 40 residues long (red), are shown fit within the density. The bracket indicates the 5-Å ladder-like spacing corresponding to the expected side chain spacing for successive α-helical turns. (c) Enlarged view of one trimeric density region and a nearby helical bundle shown with a low isosurface to reveal that they appear to be almost connected. Two α-helices of 10 and 20 residues (green) are shown fit within the connecting density. (d) Diagram of our interpretation of the difference map density on the exterior of the capsid. The four trimeric density regions are assigned to four trimers of the N-terminal capsid interaction domain of protein IX (aa 1 to 56) (blue), the weak density connections are assigned to the low-complexity midsection of protein IX (aa 57 to 91) (green), and the helical bundles are assigned to the C-terminal predicted α-helix of protein IX (aa 92 to 131) (red). The black dashed lines indicate presumably extended (non α-helical) connections between the central trimeric density region and the helical bundles. Note that one of the four helices in each bundle appears to come from a copy of protein IX in an adjacent facet. Scale bars, 50 Å. Rotating views of the density regions assigned to protein IX are presented in Movie S2, parts 1, 2, and 3 in the supplemental material.
One of the most dramatic differences between the 6- and 9-Å resolution cryoEM structures is that at higher resolution the density lobe at the facet edge is resolved into a four-helix bundle (Fig. 4b). The density lobe appears to be four moderately long α-helices twisting around each other as in a left-handed coiled coil. The cryoEM density clearly shows a 5-Å ladder-like spacing between the α-helices, which is consistent with the side chain packing from sequential turns of an α-helix. The crystal structure of the tetrameric coiled-coil domain of Sendai virus phosphoprotein (PDB 1EZJ) (43) served as good model for the observed coiled-coil density. The 63-residue α-helix in the Sendai protein coiled coil was shortened to 40 residues, as predicted for the C-terminal protein IX α-helix, and manually fit within the cryoEM density (Fig. 4b). The shortened four-helix bundle from the Sendai protein fit within the cryoEM density rods without any further modification. When displayed at a low isosurface value to include even weak density the length of the Ad helical bundle appears to be ∼40 residues long. Given that our secondary structural analysis indicates that of all of the additional capsid proteins only protein IX has a strong propensity for coiled-coil formation, we reassigned the density at the facet edges to the C-terminal region of protein IX.
We observed that the density for one of the α-helices in the Ad coiled coil is stronger than that of the other three α-helices. In fact, the fourth α-helix on the back side of the coiled coil is quite weak. These observations may be explained by the fact that only one of the α-helices, the one with the strongest density, appears to interact with a neighboring hexon. A clear connection is seen between the strongest α-helix and the HVR4 loop of a hexon in position four of the asymmetric unit (Fig. 4b). Our model for the hexon HVR4 loop (Fig. 3b) allows us to predict that HVR4 residues K251 and N254 are the most likely side chains to interact with the coiled coil, since the other side chains in the model point in the wrong direction. The weak density for three of the four α-helices suggests partial occupancy for these sites. Thus, our early proposal that the C-terminal domain of protein IX may adopt two conformations, either binding on the capsid surface or extending away from the capsid (37), is still valid. We now can elaborate on this and state that the C-terminal α-helices of protein IX may either cluster into the helical bundles at the facet edges, or possibly sit at other low occupancy sites on the capsid surface, or extend away from the capsid surface. Both low-occupancy sites and extended conformations would be difficult to reconstruct by cryoEM single-particle methods.
As additional support for our new assignment for the C-terminal domain of protein IX, we noted that at low isosurface values the trimeric density region closest to the helical bundle almost connects to the helical bundle (Fig. 4c). This density connection can be modeled as two α-helices of 10 and 20 residues, as predicted for the middle low-complexity region of protein IX. Other partial connection arms are observed in the difference map, including an arm from the trimeric density along the facet edge that extends to the adjacent facet. We propose that the low-complexity region can adopt different conformations, as would be needed to span different distances between the various trimeric density regions and the helical bundles. No connections are observed between the trimeric density region in the center of the facets and the helical bundles; however, if these connecting peptide regions are in an extended (non-α-helical) conformation, we would not expect to observe them at this resolution. It appears that at least one of the four α-helices in the bundle may come from a neighboring facet. Our interpretation of the protein density in the difference map on the exterior of the capsid surface is shown schematically in Fig. 4d. If all four sites of all 60 helical bundles in the capsid are occupied, then this would account for 240 (4 × 60) C-terminal domains of protein IX, in agreement with the copy number for protein IX in the Ad capsid.
Internal capsid density assigned to protein VIII.The 10-Å resolution cryoEM analysis of Fabry et al. led to an assignment for protein VIII to two elongated regions density per asymmetric unit on the inner capsid surface (12). We concur with this assignment, since the 6-Å resolution difference map also shows two elongated and structurally similar density regions in the asymmetric unit on the inner capsid surface (Fig. 5). However, we note that the two independent copies of protein VIII are difficult to discern in the Ad35F 6-Å difference map without density segmentation. All of the density on the inner capsid surface in the 6-Å difference map is shown in Fig. 5a for one icosahedral facet. The facet includes three asymmetric units, and thus we expect to see six copies of protein VIII. Three copies of protein VIII are well resolved and surround the three copies of hexon in position 3 at the center of the facet. The other three copies of protein VIII are partially obscured by large clusters of α-helices below the penton bases at the corners of the facet. When these α-helical clusters are computationally removed, the additional copies of protein VIII become apparent. The two independent copies of the protein VIII in the asymmetric unit are shown in Fig. 5b. Each copy of protein VIII appears to interact with four hexons, although these are hexons in different positions (2, 3, and 4) or (1, 2, and 4) depending on the protein VIII site. When the two independent volumes of protein VIII density are overlaid, they are nearly identical three-dimensional structures (Fig. 5c). Examination of the protein VIII density indicates only one long α-helix of ∼24 residues (Fig. 5d), as predicted by PsiPred for the 12.1-kDa fragment of protein VIII.
Protein density on the interior of the capsid including protein VIII. (a) A region of the cryoEM difference map (red) and pseudoatomic capsid (colored as in Fig. 1c) roughly corresponding to one icosahedral facet viewed from inside of the capsid. Three separate triangular density regions previously assigned to protein VIII, as well as clusters of multiple α-helices below the penton base, are observed. (b) Two independent copies of protein VIII (red) in the asymmetric unit. Density from other nearby proteins has been removed to reveal the similar three-dimensional structure of the two copies of protein VIII. (c) Overlap of the two independent copies of protein VIII density (red and cyan). (d) Fit of a 24-residue α-helix into the cryoEM density (mesh). Note that in this view other regions of protein VIII appear rod-shaped; however, interactive rotation of the density reveals that only one long α-helix is present as modeled. Scale bars, 50 Å.
Density assigned to protein VI within the hexon cavities.The hexon trimer has a large central cavity that faces the viral core and residue 776, which has been implicated as a key residue in the hexon-protein VI interaction (51), is in a loop between two α-helices within the cavity (Fig. 6). Thus, we carefully examined the difference map for any density within the hexon cavities. At a high isosurface value only a few small, disconnected regions of density are observed in the hexon cavities, and these disconnected regions appear as dots in Fig. 5a. At a moderate isosurface value a disk-shaped region of density is observed at the top of each hexon cavity, as well as three rods of density near the bottom of the cavity (Fig. 6). The disk-shaped density is surrounded by hydrophobic hexon residues at the top of the cavity, including copies of L463, L467, W468, and F471 from the three hexon monomers. The three rods of density could correspond to α-helices, although this is difficult to determine since the density is so weak inside of the hexon cavity. The density rods come within 10 Å of the hexon loop with residue 776, which has been implicated in protein VI binding (51). At low isosurface the density disk at the top of the cavity connects to the three rods at the bottom of the cavity. Thus, we tentatively assign all of the density within the hexon cavity to protein VI.
Density within the cavity of the hexon trimer tentatively assigned as protein VI. (a) Regions of the cryoEM difference map (red) near the hexon trimer in position 4 of the asymmetric unit. Only a thin plane of density through hexon is shown to reveal the central cavity. The hexon is oriented so that the exterior capsid surface is at the top, and the interior surface is at the bottom. The hexon loop implicated in binding to protein VI is shown in green (aa 770 to 787). The density disk at the top of the hexon cavity is indicated by an arrowhead, and two of the three density rods near the base of the cavity are marked with arrows. (b) A perpendicular thin plane of density showing the close proximity of the hexon loop (aa 770 to 787) (green) to the rod-like density indicated by arrows. A rotating view of the density within the hexon cavity tentatively assigned to protein VI is presented in Movie S2, part 4, in the supplemental material.
All of the hexons in the capsid have similar densities within their cavities. The density within the cavities also connects to density that folds over onto the bottom of the hexons, indicating that perhaps not all of the mature form of protein VI fits within the hexon cavity. A consideration of the stoichiometry indicates 720 possible binding sites for protein VI, or three potential binding sites in each of the 240 hexon trimers. However, mass spectrometry indicates only ∼369 copies of protein VI per virion (20). Thus, it seems likely that there is partial occupancy of all 720 possible binding sites for protein VI. This may explain the weak cryoEM density for protein VI in the Ad35F reconstruction.
A new assignment for protein IIIa on the inner capsid surface.A second major difference between the 6- and 9-Å resolution cryoEM structures is that the density below the penton base on the inner capsid surface is now resolved as large clusters of α-helices (Fig. 7a). At 9-Å resolution this density was speculated to correspond to protein IIIa, VI, or VIII or to the Ad protease (37). Now that α-helices are resolved at 6-Å resolution a more informed assignment can be made. Protein IIIa is unique among the Ad capsid proteins in that secondary structure prediction indicates that it is a highly α-helical protein with an α-helical content of >50%. PsiPred predicts at least 14 α-helices in the N-terminal two-thirds and 2 α-helices in the C-terminal one-third of the sequence. Each of the five helical clusters below the penton base contains at least 13 α-helices, in reasonable agreement with the predicted number for the N-terminal two-thirds of protein IIIa. The copy number of this helical cluster is also consistent with mass spectrometry data for protein IIIa, with 60 copies of protein IIIa per virion, or one copy per asymmetric unit (20).
Internal capsid density assigned to protein IIIa. (a) Density in the difference map below the penton base includes protein VIII and clusters of multiple α-helices. One-fifth of the density from the difference map is shown in red, and the remainder is in gray. A region of the pseudoatomic capsid is shown colored as in Fig. 1c. (b) Enlarged view of one-fifth of the density from the difference map. This includes one cluster of α-helices assigned to the N-terminal region (aa 1 to 400) of one protein IIIa monomer (red), together with one copy of protein VIII (gray), and unidentified density, possibly protein VI, at the base of a hexon (orange). (c) Fit of 13 α-helices, ranging in length from 12 to 23 residues, into the α-helical density assigned to protein IIIa (mesh). Scale bars, 50 Å. A rotating view of the density assigned to protein IIIa, together with the 13 modeled α-helices, is presented in Movie S2, part 5, in the supplemental material.
Now that α-helices are resolved in the cryoEM density, none of the other alternative protein candidates for the density below the penton base seems reasonable. The crystal structure of the Ad protease (26) clearly does not fit within this α-helical density. Protein VIII is now accounted for in the difference map. Even if the current protein VIII assignment is incorrect, two copies of protein VIII per asymmetric unit would only contribute two long α-helices per asymmetric unit, which is not enough to account for the observed helical cluster. Protein VI is a possible, but highly unlikely, candidate for the helical cluster. PsiPred predicts two long α-helices in protein VI and, given that there are ∼369 copies of protein VI per virion, this would imply ∼6 copies per asymmetric unit that could in principle account for 12 α-helices per asymmetric unit. However, the helical cluster appears much more like a folded protein domain than a multimer with two helices contributed from each of six copies of protein VI. Also, assigning the helical cluster to a multimer of protein VI is not very compatible with the ts147 mutant data, indicating that protein VI binds in the hexon cavity.
Therefore, the most likely assignment for the helical cluster below the penton base is the N-terminal domain of protein IIIa, which has 14 predicted α-helices. Before attempting to fit α-helices within the helical density, we segmented one helical cluster from its neighboring fivefold related clusters and from the nearby copy of protein VIII (Fig. 7b). In addition, we segmented away a small region of density on the bottom of the hexon. A similar density is found on the bottom of all copies of hexon in the capsid, and thus it is likely to be protein VI or perhaps a core component that associates with the capsid.
After segmentation, 13 α-helices ranging in length from 12 to 23 residues were fit within the helical cluster (Fig. 7c). This accounts for all but one of the predicted α-helices in the N-terminal two-thirds of protein IIIa. Clearly, at this resolution there is not a unique solution for placement of the α-helices within the density since in general the length of an α-helix can only be predicted to within a few residues and also since there are multiple predicted α-helices for protein IIIa with the same or similar length. Nevertheless, the fit of α-helices presented in Fig. 7c supports the assignment of the N-terminal region of protein IIIa to the helical cluster of density below the penton base.
Consideration of the fivefold symmetry of the vertex region implies that five copies of protein IIIa bind in five symmetrically arranged areas on the penton base as shown in Fig. 7a. These five copies of protein IIIa nearly cover the entire accessible inner surface area of penton base. The pseudoatomic capsid enables us to predict which residues in the penton base are likely to interact with the helical cluster assigned to protein IIIa. The protein IIIa-penton base interface involves ∼38 penton base residues, including residue 52, which is the first ordered residue in the crystal structure (Fig. 8). It seems likely that the N-terminal tail of the penton base (aa 1 to 51) is also involved in the interaction. The PsiPred server predicts just one α-helix with a high level of confidence for the N-terminal tail of the penton base, and it is only seven residues long. Thus, we presume that, at most, one short helical density rod in the helical cluster assigned to protein IIIa might correspond to the N-terminal tail of the penton base. This new assignment for protein IIIa below the penton base on the inner capsid surface is in accord with the biochemical findings that protein IIIa is one of the more labile capsid components and is released shortly after the penton base (14, 32, 50).
Protein interactions between protein IIIa and penton base. (a) A ribbon representation of penton base (53) (yellow) viewed along the fivefold axis, and surface representations of portions of a neighboring hexon (green) and the density assigned to protein IIIa (gray). The residues in penton base that are within 10 Å of protein IIIa are in blue. (b) Same as in panel a but with the protein IIIa density removed. The first residue (aa 52) resolved in the crystal structure of penton base is shown as a red sphere. Scale bar, 50 Å.
DISCUSSION
The collection and averaging of more cryoEM particle images, as well as the use of improved image processing techniques, enabled us to determine a 6-Å resolution structure for Ad35F. This resolution is significant in that α-helices can now be resolved in the capsid. Docking of the hexon and penton base crystal structures allowed us to confirm that all α-helices of 10 or more residues are resolved as rods within the cryoEM density. One improvement that had a significant effect on the resolution was to process particle images with a smaller pixel size of 1.5 Å on the molecular scale. This increased the size of the reconstructed volume to 7503, which in turn significantly increased the memory requirement to ∼20 GB, but we found this to be essential in order to reach 6-Å resolution. A second improvement that led to higher resolution was correction for the astigmatism in the cryomicrographs, which we accomplished with the Frealign software package (15).
Two significant new features were revealed in the 6-Å resolution Ad structure that were not apparent at 9 Å. First, a density lobe at the facet edge on the exterior of the capsid that had long been assumed to be a domain of protein IIIa is now resolved into a four-helix bundle. Structure predictions indicate that only protein IX has a strong propensity for coiled-coil formation; thus, we have revised our assignment for this exterior density region to a bundle of C-terminal α-helices from protein IX. The early assignment of this density region to protein IIIa is understandable when one considers that the density of four 40-residue α-helices is quite significant (∼18 kDa). Before this density region was resolved into α-helices there was no easy way to determine whether this represented 18 kDa from a single capsid protein or an assembly of domains from multiple copies of a capsid protein. The recent cryoEM results of Fabry et al. (12) on Ad5 and an Ad protein IX deletion mutant actually support this new assignment, although it was not clear at the time. These authors found that when protein IX was deleted, regardless of whether or not the sample was subjected to a single or multiple freeze-thaw cycles, the putative “protein IIIa” density at the facet edges was also missing. Our new assignment of the density at the facet edges to a bundle of protein IX C-terminal α-helices is consistent with their published findings.
The second new feature observed at 6-Å resolution is that the large regions of density internal to the penton base are actually clusters of 13 or more long α-helices. Consideration of secondary structure prediction results for all of the capsid proteins without atomic resolution structures makes it highly probable that these clusters are N-terminal domains of protein IIIa, which are predicted to have 14 long α-helices. Interestingly, a biopanning study using a phage-display peptide library found that a peptide from protein IIIa (aa 157 to 162 for Ad5) binds to the penton base (17). The biopanning result is in agreement with our new assignment for protein IIIa.
The new assignments for proteins IX and IIIa, which are based on the observation of α-helical density and secondary structure prediction, can be tested biochemically. We have proposed that the C-terminal α-helix of protein IX forms a four-helix bundle and that one α-helix in the bundle interacts with residues in the hexon HVR4 loop. We have also proposed that the N-terminal two-thirds of protein IIIa is likely to fold into a defined domain that interacts with the inner surface of penton base and hexon and also with an unidentified region of protein VI. Protein IIIa appears to form multiple associations with each of the other proteins in the vertex region of the icosahedral capsid, suggesting a crucial role for protein IIIa as a linchpin for virus assembly and disassembly.
Careful kinetics studies of Ad cell entry have indicated that Ad particles exist in a metastable state being relatively stable outside of the cell but weakened by the dissociation of structural proteins during the cell entry process so that it can efficiently release its DNA when it reaches the nucleus (14). Wiethoff et al. reported that protein IIIa disassociates from the Ad capsid at about the same time as penton base and one-quarter of the hexons, which are presumably the peripentonal hexons next to the penton base (50). Given these previous findings and our new structural information, we propose a more complete model of Ad disassembly. When penton base dissociates from the capsid, protein IIIa would also likely be released, bringing with it the peripentonal hexons and multiple copies of protein VI. Protein VI is thought to exist in the virion as a dimer or trimer (6, 11). If multiple copies of protein VI are connected either by direct homotypic association or via interactions with other Ad proteins in the core, this would explain the observation by Wiethoff et al. that ∼80% of protein VI dissociates at the same time as protein IIIa. Protein VI is the Ad component that participates in the disruption of the endosomal membrane, and thus its release from the Ad virion is essential for cell entry. We propose that the multiple protein-protein interactions made by protein IIIa in the vertex region facilitate the release of the membrane lytic protein VI.
ACKNOWLEDGMENTS
This study was supported by grants from the National Institutes of Health (R01-AI42929 to P.L.S. and R01-HL54352 and R01-EY11431 to G.R.N.). S.D.S. and M.S. acknowledge support from the NIH Molecular Biophysics Training Grant at Vanderbilt (T32-GM008320).
We thank Lance Gritton for preparation of the Ad35F vector, the NRAMM staff for their assistance with Leginon, Niko Grigorieff for assistance with Frealign, Harry Wodrich for helpful discussions about the Ad ts147 mutant, and Jarrod Smith and the ACCRE staff at Vanderbilt for computer support.
FOOTNOTES
- Received 1 August 2006.
- Accepted 15 September 2006.
- Copyright © 2006 American Society for Microbiology