Adenoviruses [5] have long been of interest in the basic virology field. They are present in most vertebrates [6]. As experimental systems, they have been extremely useful for investigating many fundamental processes in the eukaryotic cell life, such as splicing and apoptosis. In humans they cause usually mild respiratory, gastrointestinal and eye infections. Although the most common type of AdV infection is subclinical, AdV-induced diseases are responsible for significant morbidity in immunocompromized patients and other susceptible populations [7]. There are at present no approved antiadenoviral drug therapies [8]. Adenoviruses have recently shown promise as vectors for gene transfer into mammalian cells, vaccine delivery and oncolysis [9,10,11,12,13,14,15], although improvements in vector design are still needed [16,17]. Thus, there is currently a three-fold interest on AdV structural studies: first, because of their pathogenicity and the lack of anti-adenoviral drugs; second, because of their potential therapeutic use and the need to improve their efficiency as vectors; third, because of their role as experimental models for complex virus assembly.

Adenoviruses are icosahedral, non-enveloped viruses with a dsDNA genome. Very soon after their discovery [18], it became clear that AdV structural organization was not simple (Figure 1). An impressive amount of experimental work has been performed along this time to converge in our current, detailed knowledge of the AdV capsid organization. In this section I will briefly highlight the main milestones that took us to the situation comprehensively reviewed by W. C. Russell in 2009 [19]. Then, in the rest of the sections I will describe the latest advances in the understanding of adenovirus structure and assembly.

Early electron microscopy (EM) images of AdV allowed counting the number of capsomers forming the capsid [20]. The idea that viruses should be built from identical subunits distributed in highly symmetrical architectures, due to gene economy constraints, had already been proposed [21]. Indeed, these first electron microscopy images showed the AdV capsomers distributed according to the 532 symmetry group, proving that it was an icosahedron. When Caspar and Klug elaborated their theoretical framework for the structure of spherical viruses [22], it was clear that the adenovirus capsid corresponded to a T = 25 triangulation number icosahedral geometry. This predicted that the capsid should have 60 × 25 = 1500 structural subunits: 12 pentamers forming the vertices, plus 240 hexamers. It was assumed that all structural subunits would be chemically identical—but it soon became apparent that the AdV capsid composition did not conform to this assumption.

By combining chromatographic purification with EM of the soluble AdV antigens and of complete virions, Valentine and Pereira [23] (commented by W.C. Russell as a direct witness in [24]) showed that the 6-fold and 5-fold coordinated capsomers were two chemically different entities. According to their neighborhood in the capsid, these were later denominated “hexons” and “pentons” respectively [25]. Valentine and Pereira also produced the first striking images of the extended fibers protruding from each vertex. Later on, the complexity was extended when the virion was shown to contain at least nine different polypeptides [26]. These were named using roman numbers in order of decreasing molecular weight, as revealed by SDS electrophoresis. In this terminology, hexons correspond with polypeptide II, and pentons are composed by polypeptide III (penton base) and IV (fiber).

The next puzzle came from the AdV capsid disruption pattern. Upon mild dissociation conditions, stable substructures composed by nine hexons would reproducibly be generated. These were termed Groups of Nine hexons, or GONs. GONs contain the central hexons in each facet, but not the ones surrounding the penton, or peripentonals [29]. Now these nine hexons, which by definition occupy 6-fold coordinated positions in the T = 25 icosahedron, turned out to be organized in a p3 net, without any hint of symmetry higher than 3-fold [30]. Further, exhaustive physico-chemical analyses showed that hexons were trimers, and not hexamers, of polypeptide II [31]. Two questions arose from these studies: how could a trimeric protein fulfill the geometrical role of a hexamer? And, what made hexons in the GON different from the peripentonals?

The answer to the first question came from the hexon structure solved by X-ray diffraction. A 6 Å resolution crystallographic model showed that the trimeric capsomer had a pseudo 6-fold hexagonal base, ideally suited to establish a close-packed protein shell for protecting the viral genome [32]. In the opposite side, the trimer had three clearly marked towers that were recognizable in negative staining images of GONs and other subviral structures, and were twisted with respect to the hexagonal base. This facilitated the determination of hexon orientations in the 240 capsomers of the icosahedral facets [33,34]. Chain tracing in the electron density map revealed that the hexagonal shape was achieved by repetition of a structural motif in the base of each hexon monomer: an 8-stranded β-barrel with a “jellyroll” topology [35]. When the hexon homotrimer, rather than the monomer, was considered as the basic building block, it was realized that the icosahedral asymmetric unit (AU) was formed by four independent hexons, placed in four different environments. This is in contrast with the 25 different environments in the AU of a T = 25 icosahedron predicted by Caspar and Klug quasi-equivalence theory. Also, because of the trimeric nature of hexons, the AU is composed by 4 × 3 (hexons) +1 (penton) = 13 independent polypeptides, instead of the predicted 25. It is in this sense that the AdV capsid is often described as pseudo T = 25.

The question of why the nine hexon trimers in the GON behaved differently from the peripentonal hexons in disruption studies was solved when it was shown that GONs were formed by two different viral components: hexon, and polypeptide IX [36]. The copy number of IX was determined by ³⁵S labeling stoichiometric studies [37]. There are 240 copies of polypeptide IX per virion, with 12 copies per GON. That is, polypeptide IX is associated most intimately with hexons in the GON. The location of polypeptide IX in the GON was directly observed for the first time in difference maps where a GON model constructed from the crystal structure of hexon was subtracted from 2D EM average images of negatively stained GONs [38]. Four trimers of IX were found reinforcing the interactions between hexons at the icosahedral and a set of local 3-fold symmetry axes that are present only within the GON. No similar binding environment appears between the GON and the peripentonal hexons. Thus, the location of IX explained the defined capsid disruption pattern.

A more complete model for the distribution of minor components in the AdV capsid was obtained when the combination of X-ray and EM data was extended to the third dimension. The first 3D image of the whole virion was obtained from alignment and averaging of only 29 individual virion projections from frozen-hydrated samples, and reached a resolution of 35 Å [39]. The atomic model of hexon, filtered to the same resolution, was fitted to the four independent positions in the AU, to obtain a 3D density model for all hexon copies in the particle. This “hexon only” density was subtracted from the cryoEM map, to reveal the molecular envelope of other icosahedrally ordered capsid components [40]. This was a pioneer study, in which it was shown that it was possible to combine 3D data from X-ray diffraction and EM, even if the resolution range of the two techniques was at the time one order of magnitude away, and the physics originating the signal were quite different. Even at the limited resolution attained, the difference maps showed for the first time the shape of the penton complex, and the location of outer and inner densities reinforcing the capsid at specific positions. These were interpreted with the help of the known copy numbers and molecular weights, and assigned to minor coat proteins IIIa, VI, and IX.

At the end of the 20^th century and beginning of the 21^st, methodological and computing advances resulted in a remarkable take off for structural biology techniques. With regard to the AdV structural components, protein crystallography yielded the atomic structures of the fiber knob [41] and shaft [27], as well as penton base, alone and in complex with the N-terminal fiber peptide [42]. These structures were obtained from isolated proteins or domains; to see how they related to each other within the virion context, they were fitted into cryoEM maps moving fast beyond subnanometer resolution: first at 10, then at 9, then at 6 Å resolution [43,44,45]. Because of the higher level of detail available in these maps, it was possible to start interpreting the difference densities in terms of secondary structure elements, particularly at 6 Å resolution, if long helices were predicted. The improved models resulted in reassigned locations for polypeptides IIIa, VI, VIII, and the C-terminal domain of IX.

One of the new assignments for minor coat protein locations concerned a putative 4-helix bundle located between hexon towers at the icosahedron edges. This density was originally thought to correspond to polypeptide IIIa [40], but in the new studies it was attributed to the C-terminal domain of IX [44]. Polypeptide IIIa was now proposed to occupy a position under the vertices inside the virion. Because this reassignment implied a radical change in the accessibility of polypeptide IIIa to ligands such as antibodies or cellular receptors, and may have created confusion in the field, it is worthwhile to summarize here the evidence supporting it. Viruses lacking polypeptide IX also lack all capsid external densities not attributable to hexons or pentons [45,46]. However, for a while it was not clear if this meant that polypeptide IX was the only externally located minor capsid protein, or if the absence of IX produced the loss of polypeptide IIIa. Exogenous peptides or protein domains fused to the C-terminus of polypeptide IX confirmed its location at the facet edges, and indicated an antiparallel arrangement for its C-terminal 4-helix bundle [47,48]. Finally, a peptide mapping study using viruses with N-terminally tagged IIIa confirmed the location of this polypeptide at an internal position below the vertex region [49].

Most of the structural knowledge on AdV comes from studies on the human types 2 (HAdV-2) and 5 (HAdV-5). In the last years, the range of available AdV known structures has expanded to include the fiber knobs of other AdV species infecting humans or animals; knobs bound to their receptors; hexons of simian and avian AdV; and various fiber chimeric constructs (Table 1). These have all been obtained from isolated proteins, outside of the virion context. No structural information on any of the minor coat proteins has been reported using this strategy. It is likely that the minor coat proteins need the virion context to fold properly.

CryoEM has yielded maps of canine AdV, and the Atadenovirus ovine AdV [50,51]. Although the different species and genera differ in some of the minor capsid components, the general virion architecture is conserved. This includes particle size, hexon packing, external reinforcement by minor proteins at the 3-fold axes in the GONs, and internal densities beneath the vertices. CryoEM has also contributed to our current understanding on receptor binding and neutralization for HAdV [52,53,54,55,56,57,58,59,60].

A particularly interesting aspect of AdV was revealed by the observation that this virus, which infect vertebrates, has a striking structural similarity to PRD1, a bacteriophage with an internal membrane [87]. Previously, geometrical considerations based on the low resolution structure of AdV hexon had predicted that pseudo-hexagonal trimers could be used to build larger icosahedral capsids [33]. This prediction has now been experimentally confirmed. In the last years, more members of the PRD1-AdV family have been described or predicted, and the lineage now extends from viruses infecting bacteria or archaea, to the large nucleo-cytoplasmic DNA viruses such as Asfarvirus, Iridovirus and the giant Mimivirus [88]. All these viruses are built from the same kind of double jelly roll, pseudo-hexagonal capsomers arranged in different tiling systems, with triangulation numbers ranging between T = 21 and T = 169 (for those actually solved), and reaching up to 972 < T < 1200 for the giant Mimivirus [89]. Intriguingly, even a scaffold protein of the non-icosahedral vaccinia virus folds as a double jelly–roll pseudohexamer, indicating a possible common ancestor with icosahedral dsDNA viruses [90]. Therefore, information obtained on AdV is likely to apply to a large family of highly complex viruses.

The story of AdV structural characterization runs parallel to that of two structural biology techniques: X-ray crystallography, and electron microscopy. AdV was one of the first viruses to be imaged at the electron microscope [20]; hexon was the first animal virus protein to be crystallized [91], and the longest polypeptide solved by X-ray diffraction in its time [35]. AdV was one of the first experimental systems to benefit from the combination of both imaging techniques, at a time when it was not clear how meaningful such a combination would be. From the first interpretations of negative staining EM micrographs using a low resolution crystallographic model of hexon [92], to the difference imaging between GON EM images and hexon projections [38]; to the sophisticated quasi-atomic models and difference maps derived from three-dimensional cryoEM [40,43,44,45], both techniques were successfully put to the test to keep unveiling new details on this complex virus. The story converged when, in a parallel tour de force, the complete virion was solved at atomic resolution both by X-ray diffraction and by cryoEM [3,4].

AdV is the largest complex solved by protein crystallography, and the largest, and one of the few so far, solved at atomic resolution by cryoEM [93]. Both techniques achieved similar nominal resolutions (3.5 Å for X-ray and 3.6 Å for cryoEM). At this level of detail, it is possible to observe side chains. Therefore, the polypeptide chains can be traced in the different regions of the density map with high precision. Interestingly, for AdV, tracing in the cryoEM density map proved more informative than in the X-ray diffraction density map, even at similar levels of resolution. This section summarizes the technical developments that resulted in final success for both techniques.

To describe the complex AdV structure, it is useful to depict the capsid using a series of geometrical landmarks. Each capsid facet is formed by 12 trimers of the major coat protein, hexon. As explained above, hexon trimers have a hexagonal shape, and so they can act as hexagonal bricks in the icosahedral network. The pentagonal elements at each vertex are formed by a pentamer of penton base, in a complex with a trimer of the long projecting fiber. Four hexon trimers, plus one penton base monomer, form the icosahedral AU. The general icosahedral architecture can be described as two different systems of tiles. Nine hexon trimers form the central plate of each facet or GON. Similarly, the five peripentonal hexon trimers, together with the penton base, can be considered as a second tile system, which has been designed as GOS (Group of Six) [4] (Figure 2).

As described above, the structure of AdV in its final, infective forms has been extensively studied. However, the structural aspects of the pathway leading to this final form are less well understood. Only recently cryoEM has yielded structural information of some assembly intermediates: namely, the immature (also called young) virion [148]. Like many other viruses, AdV needs to undergo a number of proteolytic cleavages to become infective. The agent responsible for proteolytic maturation is the viral L3 23K protein, or AVP [149]. AVP recognizes (M/I/L)XGX-G and (M/I/L)XGG-X sequence motifs to cleave minor capsid proteins IIIa, VI and VIII, as well as core proteins VII, μ, and the terminal protein (TP) [132] (Figure 7). Without these cleavages, the immature particle lacks infectivity because of its inability to uncoat [150,151].

One striking particularity of the adenoviral protease, which differentiates it from all other viral proteases described so far, is its use of the viral DNA as a cofactor. Indeed, the AVP enzymatic activity is increased 100 fold in the presence of DNA. Another cofactor is the C-terminal peptide of precursor polypeptide pVI (pVI_C), released upon cleavage by AVP. Together, these cofactors increase the protease catalytic rate by several orders of magnitude [124,125,152,153]. It has been proposed that AVP uses the dsDNA molecule as a 1-dimensional track to diffuse in the crowded capsid environment and reach all its target substrates [153].

A HAdV-2 thermosensitive mutant (ts1) is a useful experimental system to investigate the structural and functional aspects of maturation [154]. When grown at the non-permissive temperature (39 °C), ts1 does not package AVP [155], and produces capsids containing the unprocessed protein precursors. Viral genome packaging is unimpaired, but the virus is not infectious. The ts1 mutation consists of a single replacement of Proline 137 by Leucine in the AVP sequence [155]. The phenotype of this classic mutant has now been reproduced by genetical engineering in both HAdV-5 and HAdV-2 [156,157].

Two cryoEM studies have analyzed the structural differences between AdV mature and immature capsids by comparison of wild type (WT) and ts1 structures obtained at 10 and 8.5 Å resolution [1,2]. These studies indicated three main differences between the mature and immature virions. First, in the inner capsid surface of ts1 there are extra densities located between the ring of peripentonal hexons and those making the GON. The authors called these densities a “molecular stitch”, a structure that would contribute to hold the GOS in place during assembly, but needs to be removed afterwards to facilitate vertex release for uncoating [158]. With the data available at the time, this feature was assigned to the C-terminal peptide of pIIIa. In the light of the atomic structure published later [4], the identity of this “stitch” becomes more clear (Figure 8). Indeed, this feature is in close proximity to two regions where polypeptide chains no longer could be traced in the atomic resolution map, either because of their absence or because of disorder. These regions are: a short stretch of residues in polypeptide IIIa VIII-binding domain (residues 216–225); and the 45 residues between the cleavage sites at residues 110 and 159 in polypeptide VIII (see Section 3.5). The extra density was only observed close to the peripentonal copy of VIII, and not in the second independent copy underneath the GON. Therefore, it seems likely that the molecular stitch is a structure formed by the contribution of the central peptides of uncleaved pVIII and IIIa, adding an additional element to the complex network of interactions holding the capsid together.

The second difference observed was another extra density located inside all hexon cavities in the ts1 structure [1,2]. Weak density has been observed at this location in all sub-nanometer resolution structural studies of the mature particle, and has been attributed to polypeptide VI [3,4,44] (Section 3.5). Stronger density in 3DEM maps of ts1 therefore indicates that when the precursor form pVI is present, the interaction with hexon is different, resulting in a more uniform occupancy or ordering of the part of VI inserted within the internal hexon cavity. This interaction change relates to the different functions of VI during the viral cycle: a stronger bond to hexon would be required for the transport function in assembly, while a looser interaction would facilitate release of VI from the capsid in the endosome.

Finally, the third difference observed between mature and immature viruses concerned the core organization. This is remarkable in itself, due to the scarce data available on the disposition of DNA and accompanying proteins within the virion, and because it confers the core architecture a role in infectivity. Although some details differed, both cryoEM studies indicated that the core undergoes a transition from a more ordered to a more disorganized structure during maturation [1,2]. Disrupted ts1 virions released compact, spherical cores, hinting at an extra stabilization of the structure. This suggests that precursor proteins pVII and pµ have a much stronger dsDNA condensing activity than their mature versions. The immature core exhibits increased thermostability, and unravels forming a 12 nm thick nucleoprotein filament that exits the capsid trough a single vertex, bringing to mind the unidirectional genome packaging and possible existence of a singular vertex [2,28,159].

AdV has long been a subject of interest for structural biology, both because of its intrinsic biological relevance and because of the challenges it poses. Work by many researchers culminated in 2010 with the resolution of the atomic structure of the virion by cryoEM and crystallography. However, plenty of questions are still open to understand the structural determinants of the AdV infectious cycle. A large part of the success in structural studies comes from taking advantage of the 60-fold averaging facilitated by the icosahedral architecture. However, this same advantage may turn into a disadvantage, since information on the non-icosahedrally ordered viral components is blurred and therefore lost in the averaging process. These components, however, may play critical roles in the viral cycle. Such is the case of the viral genome, and in the case of AdV, its large entourage of DNA-bound proteins. There are no structural data for any of the core proteins (polypeptides V, VII, μ), as well as for a large part of the polypeptide IIIa chain (residues 301 to 585). It is not known how the 12 μm long molecule of dsDNA, plus over 25 MDa of protein, fit into the ~0.1 μm diameter capsid. Unlike for its structural partner bacteriophage PRD1, AdV 3DEM maps do not show a concentric ring pattern inside the capsid [100]. Further, ordering of the nucleoproteic core changes during viral maturation [1,2].

It is still not clear how AdV capsid size is determined. Theoretical studies based on the general shape of hexon, and later structural studies, have shown that capsids with different sizes and triangulation numbers can be made using the same kind of hexagonal shaped trimeric capsomers [33,88]. In bacteriophage PRD1, an extended minor coat protein runs beneath the icosahedral edges and has been proposed to act as a tape measure for capsid building [96]. In AdV, both polypeptide VIII and IX have been proposed to act as molecular rulers [4,96]. However, VIII does not have the extended conformation and edge location expected for such a protein; and capsids with the correct size can be assembled in the absence of IX [45,46,137]. Nor are the putative packaging motor IVa2 or the core components responsible for size determination, as empty capsids with uniform size are made in the absence of IVa2 and/or packaged genome [160].

Finally, another area where considerable unknowns remain is the temporal pathway of assembly. Is capsid assembly and DNA packaging sequential, or is it a simultaneous process? How would a protein-covered dsDNA molecule be translocated through a portal-type packaging complex? How about proteolytic maturation? Does it occur after the capsid is full and sealed, or is it concomitant with packaging? How could the compact state of the immature core be reconciled with packaging into a preformed capsid? It can be expected that the latest advances in structural studies reviewed here form a firm basis to build on towards elucidation of these intriguing questions.