Structures and organization of adenovirus cementproteins provide insights into the role of capsidmaturation in virus entry and infectionVijay S. Reddya,1 and Glen R. Nemerowb

Departments of aIntegrative Structural and Computational Biology and bImmunology and Microbial Science, The Scripps Research Institute, La Jolla, CA 92037

Edited by Thomas Shenk, Princeton University, Princeton, NJ, and approved July 8, 2014 (received for review May 7, 2014)

Adenovirus cement proteins play crucial roles in virion assembly,disassembly, cell entry, and infection. Based on a refined crystalstructure of the adenovirus virion at 3.8-Å resolution, we havedetermined the structures of all of the cement proteins (IIIa, VI,VIII, and IX) and their organization in two distinct layers. We havesignificantly revised the recent cryoelectron microscopy models forproteins IIIa and IX and show that both are located on the capsidexterior. Together, the cement proteins exclusively stabilize thehexon shell, thus rendering penton vertices the weakest links ofthe adenovirus capsid. We describe, for the first time to our knowl-edge, the structure of protein VI, a key membrane-lytic molecule,and unveil its associations with VIII and core protein V, whichtogether glue peripentonal hexons beneath the vertex regionand connect them to the rest of the capsid on the interior. Follow-ing virion maturation, the cleaved N-terminal propeptide of VI isobserved, reaching deep into the peripentonal hexon cavity, de-tached from the membrane-lytic domain, so that the latter can bereleased. Our results thus provide the molecular basis for the re-quirement of maturation cleavage of protein VI. This processis essential for untethering and release of the membrane-lytic re-gion, which is known to mediate endosome rupture and deliveryof partially disassembled virions into the host cell cytoplasm.

human adenovirus | cement protein scaffold |structure–function relationships

Human adenoviruses (HAdVs) are large (∼150 nm in di-ameter, 150-MDa) nonenveloped double-stranded DNA

(dsDNA) viruses that cause respiratory, ocular, and enteric dis-eases (1). Although these diseases are self-limiting in immuno-competent individuals, they cause significant morbidity in AIDS,cancer, and organ transplant patients with compromised immunesystems (2–4). Because of their broad cell tropism and ease ofgenome manipulation, replication-deficient or conditionally rep-licating HAdVs are also being evaluated in the clinic as potentialvaccine and gene therapy vectors (5).The capsid shell of an adenovirus (Ad) comprises multiple

copies of three major capsid proteins (MCPs; hexon, pentonbase, and fiber) and four minor/cement proteins (IIIa, VI, VIII,and IX) that are organized with pseudo-T = 25 icosahedralsymmetry (Fig. 1 A and B). In addition, six other proteins (V,VII, μ, IVa2, terminal protein, and adenovirus protease) areencapsidated along with the 36-kb dsDNA genome inside thecapsid (Fig. 1A). The crystal structures of all three MCPs areknown, and so is their organization in the capsid from priorX-ray crystallography (6–8) and cryoelectron microscopy (cryo-EM) analyses (9, 10). Recently, high-resolution structures ofrecombinant HAdV5 vectors have been determined using cryo-EM (11) and X-ray methods (12) that revealed the structuresand organization of some of the cement proteins. Both studiesagree closely on the organization of the MCPs and confirm theearlier cryo-EM observations (9, 10, 13), but neither providedsignificant information on the structure and location of proteinVI, which serves key roles in the virus life cycle. Of note,

however, the two studies differ significantly in their assignmentsof the cement proteins IIIa and IX. Recent cryo-EM studiesreported that only protein IX molecules form “triskelion” as wellas “four-helix bundle” (4-HLXB) structures and mediate thenetwork of interactions between hexon subunits on the capsidexterior (11, 14, 15). They also suggested that the densities as-cribed to α-helices beneath the vertex region belong to proteinIIIa. However, based on our X-ray crystallographic data andconsidering the principles of quasi-equivalence (16), we earliersuggested that although the IX molecules form triskelion struc-tures, it is rather unlikely that the C termini of IX would form4-HLXB structures (12). Instead, we proposed that this 4-HLXBis most likely derived from a subdomain of IIIa (12).Here we report a revised interpretation, a paradigm shift, of

the structures and locations of all of the cement proteins basedon the refined crystal structure of Ad5F35 (HAdV5 vectorencoding the type 35 fiber) that includes detailed models for theordered regions of all four cement proteins (IIIa, VI, VIII, andIX). Additionally, we identified a segment of core protein V,which associates closely with protein VI. The 4-HLXB structureon the capsid exterior is a subdomain of IIIa (amino acids 101–355) that mediates interactions between group-of-nine (GON)(17) hexons (Fig. 1 C and D). The backbone path of each IXmolecule is reversed from what was assigned by the cryo-EM

Significance

Adenoviruses cause acute respiratory, ocular, and enteric dis-eases, with significant health concerns for immunocompromisedindividuals. Replication-deficient adenoviruses are among themost frequently used vectors for human gene therapy. How-ever, the structural details of these large (150-MDa) and complexviral vectors remain obscure. In this study, we determined thecrystal structures of all cement proteins in the context of theentire virion and, in the process, revised the existing cementprotein structures and their locations. Significantly, our resultsrevealed the structure of protein VI, for the first time to ourknowledge, with its cleaved propeptide sequestered withinperipentonal hexons. This permits untethering and release ofthe membrane-lytic segment, thereby providing the molecularbasis for maturation cleavage of protein VI in adenovirus-medi-ated endosome disruption.

Author contributions: V.S.R. and G.R.N. designed research; V.S.R. and G.R.N. performedresearch; V.S.R. and G.R.N. contributed new reagents/analytic tools; V.S.R. analyzed data;and V.S.R. and G.R.N. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

Data deposition: The coordinates and structure factors of the refined Ad5F35 structurereported in this paper have been deposited in the Protein Data Bank, www.pdb.org (PDBID code 4CWU).1To whom correspondence should be addressed. Email: reddyv@scripps.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1408462111/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1408462111 PNAS | August 12, 2014 | vol. 111 | no. 32 | 11715–11720

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Apr

il 18

, 202

0

studies (11, 14). Proteins V, VI, and VIII form a ternary complexthat stabilizes the adjacent peripentonal hexons (PPHs) un-derneath each of the 12 vertex regions (Fig. 1D). This complexwas incorrectly assigned to protein IIIa in cryo-EM studies (11,14). Following virion maturation, the cleaved propeptide(s) ofVI (pVIn; amino acids 1–33) is observed in the inner cavities ofthe PPHs, in agreement with recent interpretations from hy-drogen–deuterium exchange mass spectrometry studies (18).

ResultsProtein IIIa. We resolved three segments of IIIa (amino acids 9–25, 48–209, and 252–355) with good certainty, aided by the use ofa recently devised method for evaluating and scoring assignedsequences to features of experimental density. Unweighted bestmatch scores (UWBMSs; 0.82 and 0.80), weighted best matchscores (WBMSs; 0.64 and 0.60), and corresponding reliabilityindices (RIs; 3.2 and 2.1) are high for two large segments (aminoacids 48–209 and 252–355) (SI Appendix, Table S4). UWBMS/WBMS values represent the fraction of residues in the poly-peptide for which the electron density matches the size of theside chain. RI values indicate the significance of the best WBMSvalues relative to the second-best WBMS values. RI values greaterthan 1 (RI >> 1) indicate a higher confidence in the assignedsequences, whereas RI = 1 indicates the lack of specificity to thesequence assignment for the peptide. Details of calculating BMSsand RIs are described in SI Appendix,Materials and Methods. Eventhough UWBMS and WBMS values are high (0.88 and 0.76) forthe short peptide (amino acids 9–25), which was built into anisland of density, the corresponding RI value (1.2) is marginal.This resolved domain of IIIa mediates interactions between

the hexons, including PPHs, along the group-of-twelve (GOT)hexon–GOT interfaces on the capsid exterior (Fig. 2 A and B).The characteristic signature of the IIIa structure is an antipar-allel, four-helix bundle at its core and a long extended poly-peptide formed by N-terminal residues (Fig. 2B and SI Appendix,Fig. S1B). The helical bundle comprises two long helix-turn-helixsegments (amino acids 103–200 and 262–342) with short turnsat residues 147–156 and 301–309, whereas a longer connection(amino acids 210–253) is disordered. Interestingly, the visibleN-terminal segment of IIIa (amino acids 9–25 and 48–102) adopts

an extended structure and mediates interactions primarily betweena pair of PPHs and reaches all of the way to the penton vertices(Fig. 2A and SI Appendix, Fig. S2A). The remaining C-terminalresidues (356–585) are disordered, and we surmise that they likelyremain on the capsid exterior at the GON–GON interfaces nearthe twofold axis. In spanning ∼168 Å from the penton vertices tonear the twofold axis, the ordered region of IIIa appears to func-tion like the tape measure protein (P30) in PRD1 (19), but on thecapsid exterior. The location and interactions of IIIa concur withearly observations that it is surface-accessible and located near thePPHs and penton base (20, 21). Furthermore, the location of theN terminus of IIIa near the PPH and penton base also agrees withrecent studies involving N-terminal labeling of IIIa (22), eventhough it was suggested that protein IIIa resides inside the capsid.Association of the four-helix bundle with neighboring hexons

is facilitated mostly by hydrogen bond and charge–chargeinteractions (SI Appendix, Table S1). Roughly two-thirds of theinteractions involving the ordered regions of IIIa occur withhexon-2 and a PPH from the reference GOT/facet, and the re-mainder are with hexon-4′ and hexon-3′ from adjacent facets (SIAppendix, Fig. S2A). The four-helix bundle is oriented ∼35°relative to the base of the hexons, tilted down near the twofoldaxis and up toward the PPHs (SI Appendix, Fig. S2B). Two IIIamolecules (strict twofold-related), one from each facet, stabilizethe unique edge interface between a pair of facets. A total of sixIIIa molecules surround the border of each facet along three edgeson the exterior (Fig. 1C and SI Appendix, Fig. S1A), three of whichuniquely belong to each facet, one in each icosahedral asymmetricunit, resulting in 60 IIIa molecules that stabilize the entire virion.The structure and location of IIIa differ from recent cryo-EM

studies (11, 14), where the 4-HLXB domain was ascribed to theC termini from four molecules of protein IX. Comparison of theouter cement protein (IX) models from the cryo-EM study (11)[Protein Data Bank (PDB) ID code 3IYN] with that of the IIIaand IX triskelion structures here highlights the differences (SIAppendix, Fig. S3). Helices in the 4-HLXB from the X-raystructure are consistently longer than the corresponding helicesseen in the deposited cryo-EM model (PDB ID code 3IYN) (11)(SI Appendix, Figs. S18 and S20). However, to our surprise, thedensity from the cryo-EM reconstruction [Electron Microscopy

Fig. 1. Structure and organization of human adeno-virus. (A) A schematic illustration of the organization ofcapsid and core proteins in human adenovirus. Thelocations of various proteins are represented by dif-ferent-colored symbols and the corresponding namesare shown (Right). The indicated locations of the coreproteins are approximate. Shown in blue-colored let-tering are the proteins whose structures have beenidentified in this study. (B) Overall organization ofhexon and penton base subunits exhibiting pseudo-T =25 icosahedral symmetry. Structurally unique hexons(1–4) are color-coded in light blue, pink, green, andkhaki, respectively. Penton vertices are shown in ma-genta. Outer cement proteins IIIa and IX are shown inpurple and blue, respectively. Fiber molecules associ-ated with the penton base are disordered. The outlineof the triangular icosahedral facet is shown as a graytriangle, whereas the border of the GON hexons is in-dicated by yellow-colored rope. (C) An exterior view ofthe triangular icosahedral facet that comprises 12hexons along with penton base vertices shown in ma-genta. Color representations are the same as in B. (D)An interior view of the facet in C, with three minorproteins, V (green), VI (red), and VIII (orange). It isnoteworthy that a copy of V, VI, and VIII forms a ter-nary complex beneath the vertices, whereas VIII (or-ange) molecules are arranged as staples along theborder (yellow-colored rope) of the GON hexons.

11716 | www.pnas.org/cgi/doi/10.1073/pnas.1408462111 Reddy and Nemerow

Dow

nloa

ded

by g

uest

on

Apr

il 18

, 202

0

Data Bank (EMDB) ID code emd_5172] is consistent with thelonger helices seen in IIIa (SI Appendix, Fig. S20), even thoughthe connections between the helices are not clear. Furthermore,the cryo-EM models for protein IIIa (amino acids 7–300) wereincorrectly assigned to underneath the vertex region (11) thatactually belongs to a ternary complex of proteins VI and VIII anda fragment of core protein V (see below) (SI Appendix, Fig. S4).

Protein IX. Protein IX is the most flexible molecule among thecement proteins, displaying a primarily extended structure (Fig.2C and SI Appendix, Figs. S5–S7), perhaps dictated by theinteracting hexon subunits. We traced four structurally inde-pendent molecules of IX, but at varying lengths: IX-P (aminoacids 19–75), IX-Q (amino acids 12–92), IX-R (amino acids 29–77), and IX-S (amino acids 21–78) (SI Appendix, Fig. S5). TheUWBMS, WBMS, and RI scores for the longest of the IXmolecules (IX-Q) are 0.95, 0.9, and 3.5, respectively, which arethe best match/confidence scores attained for any of the cementproteins in this study (SI Appendix, Table S4). The characteristicfeature of protein IX is that it forms triskelion structure(s)consisting of three IX molecules. Two types of triskelions,termed IX-Q3 and IX-I3, occur in the Ad capsid shell. IX-Q3 isformed by three molecules of IX (IX-P, IX-Q, and IX-R) thatare related by quasi (local)-threefold symmetry and primarilystabilize hexons 2, 3, and 4. Furthermore, the N terminus of IX-Pclasps the nearest PPH (SI Appendix, Fig. S8). The second(IX-I3) is formed by three IX-S molecules that are related by ico-sahedral (strict) threefold symmetry and stabilize three hexon-3trimers at the center of the icosahedral facet (SI Appendix, Fig.S5). Therefore, four triskelions (three IX-Q3 and one IX-I3)stabilize the hexons within a GON as well as latch onto PPHswithin the facet (Fig. 1C). Hence, the Ad capsid contains 80triskelions (240 IX molecules), 60 of which are quasi-triskelions(IX-Q3) with the remaining 20 strict triskelions (IX-I3).Importantly, the direction of the IX polypeptide is reversed

compared with previous cryo-EM reports (11, 14). Our sequenceassignments agree well after reversing the chain direction, with thesole tryptophan residue (W22) in the sequence along with Y49, F79,

and F81 that can be clearly identified in the difference and omitmaps (Fig. 3B and SI Appendix, Fig. S6). In fact, 95% of the assignedresidues matched the features of the experimental density (SIAppendix, Table S4). Also, a stretch of Ala and Ser residues oc-curring between amino acids 60 and 70 forms the center/core of thetriskelion structure and also mediates interactions between theβ-barrels of the hexons. Significantly, the revised polypeptide direc-tion allows the C termini of all IX molecules to be surface-exposed,in agreement with previous biochemical and immunoelectronmicroscopy experiments (20, 21, 23). This change in chain directionalso allows the organization of IX molecules in HAdVs to be con-sistent with the arrangement of trimeric C-terminal spikes (helicalbundles) seen in nonhuman adenoviruses (24, 25), even though thecorresponding trimeric spikes are disordered in human serotypes.

Protein VI. Here we report to our knowledge the first structure ofproteinVI, which playsmultiple roles in the adenovirus life cycle—itfunctions as a cofactor for the adenovirus protease (AVP) and asa chaperone for nuclear transport, and is essential for virus assemblyand endosome lysis (26–30). One copy of protein VI is found in-timately associated with the base of each PPHon the capsid interior.We were able to resolve residues 6–31, 34–79, and 87–157 of proteinVI, which interact primarily with the bases of PPHs; the remainingC-terminal residues (158–239) were disordered. High BMS valuesand reliability indices of 2.4, 1.9, and 1.4, respectively, for thesepeptides indicate greater confidence in the sequence assignments(SI Appendix, Table S4). A representative [noncrystallographicsymmetry (NCS)-averaged] electron density for the membrane-lyticregion of VI is shown in Fig. 3A. Protein VI has a helical core andunstructured N and C termini (Fig. 3C). The fold of protein VIappears to be distinct and has no known structural homologs. Thepresence of a helical core is in agreement with observations frommodeling and CD spectroscopy studies (29, 31). After proteolysis bythe AVP, the ends of the newly formed fragments (6–31 and 34–157) are separated by ∼24 Å, perhaps due to rearrangement fol-lowing the cleavage (Fig. 3C and SI Appendix, Fig. S9A).It appears that the majority of residues (amino acids 1–21) in the

cleaved 33-residue N-terminal propeptide are buried in the cavities

Fig. 2. Structure and location of outer cement proteins. (A) A zoomed-inview of the (partial) icosahedral facet. (B) A tube representation showing thefold of protein IIIa. Rainbow coloring highlights the flow of the polypeptidechain from the N terminus (blue) to the C terminus (red). Dashed lines rep-resent disordered residues. Selected residues are labeled. (C) Structure ofone of the protein IX triskelions. Rainbow coloring blue to red shows thetrail of the polypeptide chain from the N terminus to the C terminus for eachof the three individual IX molecules P, Q, and R.

Fig. 3. Representative electron densities and the structure of protein VI. (A)NCS-averaged 3Fo − 2Fc electron density map (contoured at 1σ) correspondingto the membrane-lytic peptide of VI. The equivalent Fo − Fc map, calculated byomitting protein VI, is shown in SI Appendix, Fig. S14. (B) Similar NCS-averagedelectron density for the N-terminal part of the IX-Q molecule. (C) Stick diagramshowing the fold of protein VI. The cleaved propeptide (pVIn) (amino acids 6–31)is shown in purple color.

Reddy and Nemerow PNAS | August 12, 2014 | vol. 111 | no. 32 | 11717

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Apr

il 18

, 202

0

of the PPHs and that the remaining ordered residues (amino acids22–31) interact with the N terminus (amino acids 5–17) of the Asubunit (Fig. 4 A and B and SI Appendix, Fig. S10 and Table S2detail the interactions between VI and the PPHs). The propeptideinteractions with the base of the hexons concur with recent hydro-gen–deuterium exchange mass spectrometry studies (18) as well aswith affinity measurements between VI and the hexon (32).Moreover, although it was not reported by Liu et al. (11), electrondensity attributable to pVIn can be seen associated with PPHs intheir high-resolution cryo-EM map deposited in the EMDB (IDcode emd_5172) (SI Appendix, Fig. S20). Owing to steric con-straints, it is unlikely that more than one copy of VI would bind toeach hexon trimer in the assembled virion. In addition, hexon-4might not accommodate even one VI molecule without causingsevere steric clashes with icosahedral symmetry-related molecules.

Of note, most of the residues (34–50) in the membrane-lytic re-gion (amino acids 34–54) of VI that were predicted to form anamphipathic helix (29) do not conform to the helical structure in thecontext of the virion (Fig. 3C and SI Appendix, Fig. S9A). However,like other amphipathic helices (33), they may in fact adopt a helicalconformation after they come in contact with the endosomalmembrane, following their release from the partially disassembledvirion(s). Also, a key leucine (L40) that has been shown to impactendosome rupture (31) may influence the above helix formation (SIAppendix, Fig. S9A). In addition to its role in cell entry, the orderedportion of the mature VI molecule (amino acids 34–157) stabilizesthe capsid by gluing two PPHs together and joining them withhexon-4 of an adjacent GON hexon (Fig. 4A). The residues fol-lowing the membrane-lytic region (amino acids 68–79) closely in-teract with the neighboring PPH (Fig. 4A). The first helix of thehelix-turn-helix (amino acids 87–101) wedges in between two PPHsnear the vertex region, thus acting as “molecular glue.” In addition,residues 139–145 of the unstructured region mediate interactions atthe opposite end of this PPH–PPH interface. The C-terminal resi-dues (amino acids 145–157) join the PPHs to the adjacent GON(Fig. 4A). A set of five such glue-like interactions joins five PPHssurrounding the penton vertices to the five GON tiles (Fig. 5A).Notably, one VI molecule together with one V and one VIII-Aform a close ternary complex (Fig. 5B) that bolsters interactionsbetween the PPHs and the neighboring GONs.Residues 135–157 of a second molecule of VI-B, along with

another molecule of VIII-B, associate with hexon-2, whereas the

Fig. 4. Interactions between protein VI and the PPHs. (A) A hybrid (surfaceand tube) representation illustrating associations mediated by one copy ofVI (red tube) gluing the adjacent PPHs (1,1′) and connecting them to hexon-4′ arising from the neighboring GON tile. The cleaved propeptide of VI(purple tube) remains associated with PPH-1 inside the hexon cavity. Certainresidues of VI are identified with blue-colored labels. The penton base (PB) isshown in magenta. (B) A close-up view of the propeptide (purple tube)interactions with PPH-1, shown as a gray ribbon. A few residues that areinvolved in propeptide (purple) and hexon (green) interactions are labeled.Names of hexon subunits (A and C) are shown in parentheses.

Fig. 5. Organization of cement proteins underneath the vertex region. (A)Five copies of the ternary complex formed by proteins V (green), VI (red),and VIII (orange) glue the peripentonal hexons (gray) to each other andconnect them to the adjacent GONs. One of the copies is highlighted bya black oval. (B) A stick diagram showing the structure of the ternary com-plex. Color assignments of the proteins are as in A, with the exception thatthe propeptide of VI (pVIn) is shown in purple and the C-terminal fragmentof VIII (VIII-3) is shown in orchid, whereas the N-terminal fragment of VIII(VIII-1) is shown in orange.

11718 | www.pnas.org/cgi/doi/10.1073/pnas.1408462111 Reddy and Nemerow

Dow

nloa

ded

by g

uest

on

Apr

il 18

, 202

0

rest of VI-B, including the N-terminal propeptide, is disordered(SI Appendix, Fig. S11B). This binary complex between VI-B andVIII-B mediates interactions between the same adjacent GON(facet) that are stabilized by the above (V–VI–VIII) ternarycomplex (SI Appendix, Fig. S11A). This accounts for 120 of ∼350molecules of VI believed to be present in the adenovirus virionbased on biochemical (34) and mass spectrometry proteomicsstudies (35). However, another 60 molecules of VI could steri-cally be accommodated associating with 60 copies of hexon-3(i.e., one VI per hexon), thereby increasing the number of VImolecules interacting with the hexons to 180. The remaining(∼170) molecules of protein VI, if indeed present, might be as-sociated with the nucleoprotein core of HAdV.

Protein VIII. Based on biochemical and structural studies it wasshown that two structurally distinct copies of protein VIII (des-ignated A and B) interact with the bases of the hexons on thecapsid interior (13, 14, 20, 21). Protein VIII is processed by AVPat two locations (amino acids 111 and 157), resulting in threefragments (amino acids 1–111, 112–157, and 158–227). We couldtrace residues 33–90 of fragment 1; 163–215 in fragment 3 inVIII-A; and residues 31–86 of fragment 1 in VIII-B (SI Appendix,Figs. S9C and S11B). Reliability indices for fragments 1 and 3 ofVIII-A are 1.9 and 1.5, respectively, and along with high UWBMSvalues indicate that ∼90% of assigned residues match the ex-perimental density (SI Appendix, Table S4). No density was ob-served corresponding to fragment 2 in either VIII-A or VIII-B,suggesting that it might have been released from the virion afterthe processing by AVP. The structure of fragment 1 mostly agreeswith that seen in the high-resolution cryo-EM reconstruction(11), but the fragment 3 structure differs completely from thatreported by cryo-EM (SI Appendix, Fig. S12).As described above, VIII-A forms a ternary complex with

proteins VI and V that interacts with PPHs, whereas a secondmolecule of VIII (VIII-B) forms a binary complex with a shortfragment of VI-B that closely associates with hexon-2 in thereference GON. Both unique copies of VIII (A and B) line theborder of the GONs and mediate interactions between hexonsfrom adjacent GONs and PPHs (Fig. 1D and SI Appendix, Fig.S11). A total of 120 copies of VIII glue the PPHs and GONstogether on the interior of the Ad capsid.

Protein V.Even though protein V is considered a core protein andalong with protein VII is known to interact with the Ad genome,it also interacts with protein VI based on cross-linking experi-ments (21, 36). We could trace 72 (amino acids 208–219 and236–295) of the 368 residues of protein V (SI Appendix, Fig.S9B). The ordered region of V includes two short helices (amino

acids 208–219 and 259–271), with the rest forming an extendedstructure. Even though the BMS values (0.9 and 0.85) are high,the reliability indices (1.0 and 1.3) are marginal for the respectivepeptides, particularly for the short segment of 12 residues (aminoacids 208–219) (SI Appendix, Table S4). This short peptide,which forms a helix, was built into an island of density foundclose to the larger segment (amino acids 236–295). Significantly,85% of sequence assignments for the larger segment of 60 res-idues agree well with the protein V sequence, which is more thanany other cement protein sequence on the capsid interior. Thevisible C terminus of V (amino acids 289–295) interacts with VI(amino acids 103–115) as the former reaches underneath thevertex region (Fig. 5A). Based on this organization, the disor-dered and highly basic N-terminal part of V likely interacts withthe genome, which is also disordered in the HAdV crystalstructure. The structurally ordered region of protein V lies be-neath protein VI (closer to the center of the particle) and doesnot directly interact with the hexon subunits but likely mediatesinteractions between the (disordered) nucleoprotein core andthe capsid shell through its interactions with protein VI. ProteinV also interacts with the C-terminal fragment 3 of VIII-A informing the V–VI–VIII ternary complex that stabilizes the PPHsand links them with adjacent GONs (Fig. 5 and SI Appendix,Fig. S11A).

DiscussionThe refined crystallographic model of HAdV, one of the largestbiomolecules determined by X-ray crystallography, reveals thestructures and organization of all of the cement proteins andthe partially ordered core protein V, which together stabilizethe Ad capsid shell. The cement proteins are organized in twodistinct layers (Fig. 6), which exclusively stabilize the hexon shellincluding PPHs, thus rendering the penton vertices the weakestlinks of the HAdV capsid and enabling their release into the low-pH endosome. The outer layer is formed by proteins IX and IIIa,which mediate interactions within and between GON/GOThexons, respectively. Notably, the N terminus of IIIa tethers thePPHs, connecting them with GONs on the capsid exterior. Theseinteractions are consistent with experimental observations thatprotein IIIa is surface-accessible and is released along with theperipentonal hexons (21). In addition, reversal of the chain di-rection of protein IX molecules allows the surface exposure oftheir C termini, in agreement with immunoelectron microscopyresults (23). Moreover, protein IX triskelion structures in alllikelihood are enforced by the adjoining hexon subunits, as thereare hardly any interactions between individual IX molecules.Together, IIIa and IX form a nearly contiguous hexagonal lattice

Fig. 6. Organization of cement proteins in humanadenovirus. (A) Organization of the five differentminor/cement proteins IIIa (purple), IX (blue), V (green),VI (red), and VIII (orange) that are present in multiplecopies. Color assignments of inner and outer cementproteins are indicated (Lower). (B) Cross-section show-ing the double-layered organization of the cementproteins. The outer layer is composed of proteins IIIaand IX and the inner layer is formed by proteins V, VI,and VIII. The encapsidated DNA is not ordered in theaveraged electron density maps, computed from dif-fraction data in the resolution range 20–3.8 Å.

Reddy and Nemerow PNAS | August 12, 2014 | vol. 111 | no. 32 | 11719

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Apr

il 18

, 202

0

(SI Appendix, Fig. S13A), which surrounds the hexons withineach GON. The inner cement protein layer (SI Appendix, Fig.S13B), although not as contiguous as the outer layer, is formedby proteins V, VI, and VIII, where VI and VIII closely associatewith the bases of the hexons by stapling them along the GON–

PPH and GON–GON interfaces. However, the ordered seg-ments of protein V seen in the crystal structure do not directlycontact the hexon shell. The locations of the “hot spots” of ce-ment protein interactions that surround the hexon subunits arecomplementary in the upper and lower layers. Interestingly,some of the distinct features observed in the X-ray structurecompared with the deposited cryo-EM model (PDB ID code3IYN) (e.g., longer helices in IIIa and the sequestered pVInpeptide inside the PPH) are supported by the cryo-EM density(SI Appendix, Fig. S20), but were not previously considered (11).Most importantly, the structural models for VI and V provide

insights into the need for proteolytic processing of VI so it can bereleased, in turn leading to partial exposure of the DNA. Pro-teolytic processing of VI is necessary for untethering and releaseof the membrane-lytic region, which is implicated in endosomelysis (29, 30). Consistent with this cleavage requirement, abrogationof such processing of VI by the viral protease, as in the case of Ad-ts1 mutant viruses, makes them noninfectious (37). Furthermore,the close apposition of five copies of protein VI near the icosa-hedral vertices allows cross-linking of lysine residues K45 and K70from neighboring molecules (∼4.5 Å apart), and the closeproximity of proteins V and VI would allow the cross-linking be-tween V and the dimer of VI (Fig. 5 and SI Appendix, Fig. S15) (21,36). The release of the mature VI molecules would occur con-comitant with the loss of peripentonal hexons, followed by therelease of partially disassembled virions into the cytoplasm.

Materials and MethodsA form of recombinant human adenovirus serotype 5 vector displayinga short and flexible fiber fromHAdV serotype 35, termed Ad5F35 (also knownas Ad35F), was purified and crystallized as described previously (38). Dif-fraction data were collected at beamline 23-ID-D, General Medical Sciencesand Cancer Institute’s structural biology facility (GM/CA), Advanced PhotonSource. The crystal structure was determined using molecular replacementmethods using Phaser (39). Notably, high-resolution features of the electrondensity maps were significantly improved by omitting data at resolutionslower than 20 Å, which is akin to B-factor sharpening of electron densitymaps. All of the initial cement protein backbone traces were built into dif-ference (Fo − Fc) and/or omit (5Fo − 4Fc) maps. Only polyalanine traces ofthe cement proteins were used to calculate electron density maps, to avoidany model bias. We devised a method for evaluating the reliability ofassigned amino acid sequences to the experimental electron density (see SIAppendix, Materials and Methods for details). The reliability scores for mostof the sequence assignments of the cement proteins are high. The X-raymodel was refined using the program X-PLOR (40). Further details of ex-perimental procedures are described in SI Appendix, Materials and Methods.

ACKNOWLEDGMENTS. We are grateful to Drs. Jack Johnson, Ian Wilson, andTimothy Baker for their comments on the manuscript, as well as to Dr. CrystalMoyer for careful reading of it; to Tina-Marie Mullen for her assistance withvector production; and to Drs. Kundhavai Natchiar and Nhung Huynh fortheir support with data collection. We acknowledge the cryo-EM maps andmodels received from Drs. Phoebe Stewart and Guy Schoehn. We also thankDr. Bob Fischetti and colleagues at the GM/CA beamline at the AdvancedPhoton Source (APS), Chicago, for technical support with synchrotron datacollection. We also thank Dr. Tom Goddard for his expert advice on usingspecial features of the Chimera program. We acknowledge support fromNational Institutes of Health (NIH) Grants AI070771 and AI103692 (to V.S.R.)and HL054352 (to G.R.N.). GM/CA has been funded with federal funds fromthe NIH (Y1-CO-1020, Y1-GM-11040), and the APS is supported by the USDepartment of Energy under Contract DE-AC02-06CH11357.

1. Wold WSM, Horwitz MS (2007) Adenoviruses. Fields Virology, eds Knipe DM,Howley PM (Lippincott Williams & Wilkins, Philadelphia), 5th Ed, Vol 2, pp 2395–2436.

2. Hierholzer JC (1992) Adenoviruses in the immunocompromised host. Clin MicrobiolRev 5(3):262–274.

3. Hoffman JA (2006) Adenoviral disease in pediatric solid organ transplant recipients.Pediatr Transplant 10(1):17–25.

4. Lenaerts L, De Clercq E, Naesens L (2008) Clinical features and treatment of adeno-virus infections. Rev Med Virol 18(6):357–374.

5. Campos SK, Barry MA (2007) Current advances and future challenges in adenoviralvector biology and targeting. Curr Gene Ther 7(3):189–204.

6. Rux JJ, Burnett RM (2000) Type-specific epitope locations revealed by X-ray crystal-lographic study of adenovirus type 5 hexon. Mol Ther 1(1):18–30.

7. Zubieta C, Schoehn G, Chroboczek J, Cusack S (2005) The structure of the humanadenovirus 2 penton. Mol Cell 17(1):121–135.

8. van Raaij MJ, Mitraki A, Lavigne G, Cusack S (1999) A triple beta-spiral in the adenovirusfibre shaft reveals a new structural motif for a fibrous protein. Nature 401(6756):935–938.

9. Stewart PL, Burnett RM, Cyrklaff M, Fuller SD (1991) Image reconstruction reveals thecomplex molecular organization of adenovirus. Cell 67(1):145–154.

10. Stewart PL, Fuller SD, Burnett RM (1993) Difference imaging of adenovirus: Bridgingthe resolution gap between X-ray crystallography and electron microscopy. EMBO J12(7):2589–2599.

11. Liu H, et al. (2010) Atomic structure of human adenovirus by cryo-EM reveals inter-actions among protein networks. Science 329(5995):1038–1043.

12. Reddy VS, Natchiar SK, Stewart PL, Nemerow GR (2010) Crystal structure of humanadenovirus at 3.5 Å resolution. Science 329(5995):1071–1075.

13. Fabry CM, et al. (2005) A quasi-atomic model of human adenovirus type 5 capsid.EMBO J 24(9):1645–1654.

14. Saban SD, Silvestry M, Nemerow GR, Stewart PL (2006) Visualization of alpha-helicesin a 6-Ångstrom resolution cryoelectron microscopy structure of adenovirus allowsrefinement of capsid protein assignments. J Virol 80(24):12049–12059.

15. Fabry CM, et al. (2009) The C-terminal domains of adenovirus serotype 5 protein IX as-semble into an antiparallel structure on the facets of the capsid. J Virol 83(2):1135–1139.

16. Caspar DLD, Klug A (1962) Physical principles in the construction of regular viruses.Cold Spring Harb Symp Quant Biol 27:1–24.

17. Smith KO, Gehle WD, Trousdale MD (1965) Architecture of the adenovirus capsid.J Bacteriol 90(1):254–261.

18. Snijder J, et al. (2014) The cleaved N-terminus of pVI binds peripentonal hexons inmature adenovirus. J Mol Biol 426(9):1971–1979.

19. Abrescia NG, et al. (2004) Insights into assembly from structural analysis of bacte-riophage PRD1. Nature 432(7013):68–74.

20. Everitt E, Sundquist B, Pettersson U, Philipson L (1973) Structural proteins of ad-enoviruses. X. Isolation and topography of low molecular weight antigens from thevirion of adenovirus type 2. Virology 52(1):130–147.

21. Everitt E, Lutter L, Philipson L (1975) Structural proteins of adenoviruses. XII. Location andneighbor relationship among proteins of adenovirion type 2 as revealed by enzymaticiodination, immunoprecipitation and chemical cross-linking. Virology 67(1):197–208.

22. San Martín C, et al. (2008) Localization of the N-terminus of minor coat protein IIIa inthe adenovirus capsid. J Mol Biol 383(4):923–934.

23. Akalu A, Liebermann H, Bauer U, Granzow H, Seidel W (1999) The subgenus-specificC-terminal region of protein IX is located on the surface of the adenovirus capsid.J Virol 73(7):6182–6187.

24. Schoehn G, et al. (2008) Three-dimensional structure of canine adenovirus serotype 2capsid. J Virol 82(7):3192–3203.

25. Cheng L, et al. (2014) Cryo-EM structures of two bovine adenovirus type 3 inter-mediates. Virology 450-451:174–181.

26. Mangel WF, McGrath WJ, Toledo DL, Anderson CW (1993) Viral DNA and a viral peptidecan act as cofactors of adenovirus virion proteinase activity. Nature 361(6409):274–275.

27. Wodrich H, et al. (2003) Switch from capsid protein import to adenovirus assembly bycleavage of nuclear transport signals. EMBO J 22(23):6245–6255.

28. Greber UF, Willetts M, Webster P, Helenius A (1993) Stepwise dismantling of ade-novirus 2 during entry into cells. Cell 75(3):477–486.

29. Wiethoff CM, Wodrich H, Gerace L, Nemerow GR (2005) Adenovirus protein VI me-diates membrane disruption following capsid disassembly. J Virol 79(4):1992–2000.

30. Maier O, Galan DL, Wodrich H, Wiethoff CM (2010) An N-terminal domain of ade-novirus protein VI fragments membranes by inducing positive membrane curvature.Virology 402(1):11–19.

31. Moyer CL, Wiethoff CM, Maier O, Smith JG, Nemerow GR (2011) Functional genetic andbiophysical analyses of membrane disruption by human adenovirus. J Virol 85(6):2631–2641.

32. Graziano V, et al. (2013) Regulation of a viral proteinase by a peptide and DNA inone-dimensional space: II. Adenovirus proteinase is activated in an unusual one-dimensional biochemical reaction. J Biol Chem 288(3):2068–2080.

33. Drin G, Antonny B (2010) Amphipathic helices and membrane curvature. FEBS Lett584(9):1840–1847.

34. van Oostrum J, Burnett RM (1985) Molecular composition of the adenovirus type 2virion. J Virol 56(2):439–448.

35. Benevento M, et al. (2014) Adenovirus composition, proteolysis, and disassembly studiedby in-depth qualitative and quantitative proteomics. J Biol Chem 289(16):11421–11430.

36. Chatterjee PK, Vayda ME, Flint SJ (1985) Interactions among the three adenoviruscore proteins. J Virol 55(2):379–386.

37. Weber J (1976) Genetic analysis of adenovirus type 2 III. Temperature sensitivity ofprocessing viral proteins. J Virol 17(2):462–471.

38. Reddy VS, et al. (2010) Crystallization and preliminary X-ray diffraction analysis ofhuman adenovirus. Virology 402(1):209–214.

39. McCoy AJ, et al. (2007) Phaser crystallographic software. J Appl Crystallogr 40(Pt 4):658–674.

40. Brunger AT (1992) X-PLOR, Version 3.1: A System for X-Ray Crystallography and NMR(Yale Univ Press, New Haven, CT).

11720 | www.pnas.org/cgi/doi/10.1073/pnas.1408462111 Reddy and Nemerow

Dow

nloa

ded

by g

uest

on

Apr

il 18

, 202

0