Superimposition of Viral Protein Structures: A Means to Decipher the Phylogenies of Viruses

Abstract

Superimposition of protein structures is key in unravelling structural homology across proteins whose sequence similarity is lost. Structural comparison provides insights into protein function and evolution. Here, we review some of the original findings and thoughts that have led to the current established structure-based phylogeny of viruses: starting from the original observation that the major capsid proteins of plant and animal viruses possess similar folds, to the idea that each virus has an innate “self”. This latter idea fueled the conceptualization of the PRD1-adenovirus lineage whose members possess a major capsid protein (innate “self”) with a double jelly roll fold. Based on this approach, long-range viral evolutionary relationships can be detected allowing the virosphere to be classified in four structure-based lineages. However, this process is not without its challenges or limitations. As an example of these hurdles, we finally touch on the difficulty of establishing structural “self” traits for enveloped viruses showcasing the coronaviruses but also the power of structure-based analysis in the understanding of emerging viruses

1. Introduction

Viruses mutate at a faster rate than cellular organisms—about four orders of magnitude [1]. Therefore the traceability of evolutionary relationships by sequence similarity is easily lost even within the same virus family. Sequence similarity can be assessed at DNA/RNA or protein level, and several tools have been developed to decipher whether two (or more) proteins are homologous or descend from a common ancestor (homology; divergent evolution) [2]. Interpretation of high sequence similarity between two proteins nevertheless cannot rule out the different scenario driven by convergent evolution (analogy or coincidence). Thus, common ancestry should never be inferred by using a single number or score, but rather, it should involve the fulfilment of other constraints such as the biological function, cellular localization, and/or the gene locus within the genome. In the virus world, some of these constraints become muddled-up and comparison of sequences (linear strings of a combination of either four (nucleotides) or twenty (amino acids) letters) becomes of limited usefulness for unravelling primitive precursors. On the other hand, the process of comparing three-dimensional (3D) objects is an activity that humans show at very young age, from about 3 years old, and which helps children (us, once upon a time) in establishing new object categories, an invaluable learning process of the outside world [3]. This same activity but comparing the structures of two or more proteins allows us to identify the salience of conserved 3D features even when the similarity in their primary sequences has become undetectable over the course of time. As the 3D structure entails structure-function constraints tuned along the course of evolution, the structure-based methods represent a powerful means for phylogenetic classification, not only of the protein world but also of the virosphere [4,5]. The elucidation of virus common ancestry together with current genome sequencing capabilities and the advance of cryo-electron microscopy (cryo-EM) for biomolecular structure determination [6,7] will also unlock the complexity of the intertwined evolutionary relationship with cellular organisms.

Here, we briefly review some of the groundwork concepts that have led to the quantitative comparison of 3D protein structures and its implications for the structure-based taxonomy of viruses as emerged from the initial systematic study of major capsid proteins (MCPs) possessing a double β-barrel topology (or double jelly-roll; DJR). Finally, we glance at the challenge of a structure-based classification for enveloped viruses and highlight the impact that structural comparisons entail on our understanding of how evolution can change structural motifs leading to the emergence of new viruses as in the case of SARS-CoV-2.

2. Structural Superimposition of Proteins and Homology of Viral Coat Proteins

The necessity of comparing 3D protein structures stretches back to the early sixties thanks to success in the determination of the hemoglobin crystal structure and the full establishment of macromolecular X-ray crystallography as a leading technique in the structural biology field [8,9]. In light of these results Huber and colleagues performed a structural alignment study between insect (invertebrate) hemoglobin and whale (vertebrate) myoglobin showing their high structural similarity despite the low amino acid sequence identity (~20%) [10]. They used a least-square minimization algorithm to minimize the sum of the square of the distances across the equivalent Cαs in the corresponding backbones [10]. The concept of superimposing equivalent residues when these are not visually intuitive is not trivial and so is the definition of a metric for the “goodness” of the alignment. A method for defining equivalent residues when these are not visually obvious (as in the case of myoglobins) was proposed in 1976 by Rossmann and Argos and aimed at identifying structural homology across proteins (HOMOLOGY software) [11,12]. They proposed to use a “probabilistic” approach to define the subset of equivalent backbone atoms in comparing sets to overcome the difficulty of this assignment. Once these were defined, the degree of structural parallelism was calculated. Since then several tools such as COMPARER [13], Secondary Structure Matching (SSM) [14], Dali [15], Structure Homology Program (SHP) [16], and Homologous Structure Finder (HSF) [17], to mention a few, have been developed to address the structural superimposition problem. The “probabilistic” method proposed by Rossmann and Argos is at the base of some of these tools such as SHP and HSF software which have been extensively used for establishing the structure-based classification of viruses (see below).

The eighties saw further successes of virus crystallography with the determination of plant viruses such as Tomato bushy stunt virus and Southern bean mosaic virus [18,19]. Later on animal-infecting viruses of the Picornaviridae family revealed noticeable structural similarities in the corresponding MCP folds despite the lack of sequence similarity—strikingly not only within viruses infecting the same host but also across viruses infecting different hosts [20,21]. These MCPs possessed eight β-strands arranged with the long axis of the β-barrel tangentially to the virus surface (Figure 1A). At the time, there were not many virus structures solved and therefore it was not clear whether the detected similarities were the remains of a common origin (divergent evolution) or they appeared simply because there was no other way of forming a suitable viral capsid (convergent evolution). All the above viruses were relatively small icosahedral particles (about 300 Å in diameter) possessing a single-stranded (ss) RNA genome. The similarity of the three MCPs, VP1, VP2, and VP3, of human rhinovirus 14 to the structure of those of the plant viruses led Rossmann and colleagues to put forward the idea that they diverged from a common ancestor (Figure 1A) [20]; at the time a hypothesis that—as we now realise—led to several important implications in virus phylogeny.

3. Major Capsid Proteins with a Double β-Barrel Fold: The Conceptualization of Viral Lineages

The observation suggested by Rossmann and co-workers that certain viruses may share a common descent pullulated the field for some time. In 2003, thanks to the boost of viral structures determined by X-ray, Chapman and Liljas carried out a comprehensive analysis of the topology of viral protein folds available [25]. Significantly, they noted the recurrence of certain folds in virions belonging to distinct taxonomic viral families and postulated the ramifications that this recurrence could have for inferring phylogenetic relationships [25]. Nonetheless, a systematic and unifying approach through which these ideas could be tested was still lacking, at least until the X-ray structure of the MCP P3 of lipid-containing phage PRD1 was determined [26]. The structure showed that both bacteria-infecting PRD1 and eukaryotic-infecting adenoviruses possess a similar MCP with a vertical double β-barrel fold, a finding that opened up a treasure chest of new discoveries. This fold entails two single β-barrel connected by a short linker; each β-barrel consists of eight β-strands arranged in two four-stranded antiparallel β-sheets packed together and with their long axis perpendicular to the capsid shell (Figure 1B). In opposition to all previously studied viruses with tangential β-barrels MCPs (e.g., Figure 1A), both PRD1 and adenovirus pointed at a different capsomer morphology and organization. The capsomers displayed a pseudo-hexameric appearance generated by the trimerization of the corresponding MCPs (Figure 1C); they could also be readily fitted into the low resolution cryo-EM map of Paramecium bursaria Chlorella virus (PBCV-1), a virus infecting algae [27].

On the basis that coat protein topology and virion architecture could be used as fossil fingerprints (as “innate self” traits of these viruses), it was proposed that PRD1, adenovirus, and PBCV-1, the largest of the three with a diameter of ~1900 Å, would constitute a viral lineage [27]. Confirmation of this proposal came with the crystal structure of PBCV-1’s coat protein Vp54 as possessing a double β-barrel fold by the Rossmann group [28], and with the determination of the X-ray structure of the whole PRD1 at 4.2 Å resolution (Figure 1C, right) [23,24]. Several MCPs and virion structures elucidated to atomic details have expanded our understanding of the virosphere [28,29,30,31,32,33,34,35]. A clear example of this endeavor are the coat proteins with the so-called HK97 fold characterized basically by two domains: the P-domain, with a long α-helix and a three-stranded β-sheet, and the A-domain, with two α-helices and a β-sheet, and found in viruses infecting different domains of life [36]. With the accumulation of structural data, the idea emerged that viruses could be structurally classified in four lineages, initially called PRD1-like, HK97-like, BTV-like, and picorna-like [37]. The classification of the virosphere using the above four structure-based viral lineages accounts for more than 30 viral families (Figure 2A), in contrast to the 22 accounted by the five-rank Linnean-like structure used by the International Committee on Taxonomy of Virus (ICTV) until 2017 [38,39]. Very recently, however, in light of the accruing of structure-function and phylogenomic evidence of evolutionary relationships among what had been considered “distantly related viruses” (thanks also to the increased speed of genome sequencing, computing power, and more sensitive algorithms) has made unavoidable the revisiting of the virus classification [40,41,42,43]. Indeed, the ICTV has recognized the usefulness of extending its Code to a 15-rank classification hierarchy which reflects the proposal of a genome-based megataxonomy of the virus world that introduce the ranks of Realm, Kingdom, and Phylum among others [44,45].

Today, the first established structure-based PRD1-adenovirus lineage accounts for over 10 virus members infecting organisms across the three domains of life (Figure 2B). Remarkably, this lineage includes ssDNA viruses such as Flavobacterium-infecting, lipid-containing phage (FLiP) [46], non-icosahedral viruses such as Vaccinia [32,35], and virophages such as Sputnik, a satellite non-membrane-containing virus of the giant Mimivirus [47], which leads us to the question of whether or when—as the descent character of the double β-barrel is present in most polintons/mavericks sequences [48]—polintons will be also included in the phylogentic-tree of this lineage.

The size distribution of member viruses of the above lineage (from ~600 to ~2400 Å in diameter) shows how the pseudo-hexameric morphology of the capsomers is elegantly used to fulfill the requirement for building an icosahedron according to the assembly principles proposed by Caspar and Klug [49,50]. The footprint of these capsomers, which occupy the hexavalent positions on a planar hexagonal lattice, is an invariant trait of this lineage and measures approximately 75 Å in diameter (Figure 3, left) [33]. This dimension ultimately recapitulates the relative packing and angular orientation of the individual β-barrels, V1 and V2, which is reflected in the angular aperture, of helices FG1-α and FG2-α (~69° as estimated for PRD1 P3 in CHIMERA [23,51]) and is mainly conserved across all the PRD1-adenovirus members (Figure 2B and Figure 3; <66.9° > ±7.3° as estimated across 26 MCP structures derived either through X-ray or cryo-EM). The structural invariance of this footprint allows the identification of the double β-barrel fold even at intermediate resolution, as occurred in the case of the African Swine Fever virus (ASFV) p72 MCP whose 8280 copies compose the virion outermost capsid shell (Figure 3, right) [52]. The ASFV p72 double β-barrel and the MCPs of the other large nucleocytoplasmic large DNA virus (NCLDV), PBCV-1 and Faustovirus cluster together [26,33,53,54,55,56] (Figure 2B). This NCLDV clustering indicates that some structural differences in the core consensus DJR exist when compared to the DJR core of the smaller MCPs of PRD1 or PM2. Structural alignment carried out with the HSF software [17] highlights that the two V1 and V2 jelly rolls in NCLDV MCPs are further spaced. Three β-sheets at the base of, and connecting, the two jelly rolls appear to act as a spacer between the two domains. Except PM2 P2, which represents the minimalist double β-barrel core module, all MCPs are crowned by large and structured insertions in loops connecting, for example, the DE and FG β-strands above the two V1 and V2 β-barrels, and have extensions at the N- and C-terminal ends (Figure 1B and Figure 2B). The former protrude toward the outside of the virus while the latter point to the interior of the virion leading to a three module structure (Figure 2B; e.g., adenovirus); each module with specific functionalities. The core double β-barrel is key for assembling the pseudo-hexameric capsomers, the elemental building blocks for generating virions with increasing sizes (Figure 1C). The interior-module allows the rotational registering across capsomers and the anchoring of the capsomers to the underneath structural constituents and, depending on the virus type, to the membrane vesicle, to membrane-associated proteins, or to the genome (Figure 1B, left) [27,28,29,57]. The exterior module, constituted by the turret domains or extended loops above the V1 and V2 towers, is also in some cases, such as adenovirus hexon, faustovirus and ASFV p72 MCPs, involved in trimer/capsomer stabilization [22,33,53]. However, in the case of adenovirus hexon (Figure 1B), it has been shown that these structural elements extending outward from the virion surface contain the so-called “hypervariable regions”, which play a crucial role in eliciting the cell immune response [58]. These sites offer the possibility to be genetically modified or re-purposed with important consequences in vaccine and gene therapy applications in cases of pre-existing immunity or the delivery is hampered for specific adeno-vector serotypes [59,60].

The PRD1-adenovirus lineage is expanding; a new tentative family of viruses called Autolykiviridae may also possess a MCP with a DJR fold [61]. Within the current ICTV 15-rank classification, the PRD1-adenovirus structure-based lineage constitutes the new taxon rank of Kingdom with Bamfordvirae as taxon name (for the reasons behind this name please see Dennis Bamford’s webpage, Helsinki University), which in turn, falls within the Realm of the Varidnaviria [44,45].

4. Viruses with Vertical Single β-barrel MCPs: A Clade of the PRD1-Adenovirus Lineage

In 2006, the identification of icosahedral archaeal tailless membrane-containing Haloarcula hispanica virus 1 (SH1) encoding two MCPs but forming capsomers with a resemblance to that of PRD1 (e.g., similar pseudo-hexagonal footprint) raised the question of how the virus would assemble [62,63]. SH1 pseudo-hexameric capsomers displayed both two- and three-turret morphologies whilst capsomers of Thermus bacteriophage P23-77, at the time the only other virus identified possessing two MCPs, a two-turret morphology. Both viruses share the same triangulation (T) number, pseudo T = 28. When the X-ray structures of VP16 and VP17 MCPs of P23-77 were solved, they showed that each of the MCPs adopts a vertical β-barrel (or vertical single jelly-roll, vSJR) fold. Later, the characterization at 11 Å by cryo-EM of the archaeal lipid-containing Haloarcula hispanica icosahedral virus 2 (HHIV-2) suggested not only MCPs with a vSJR fold but also the presence of additional proteins in between the capsid and the membrane vesicle [64,65]. None of the above studies, however, explained the assembly process of these new viruses for which a novel taxonomic family, Spherolipoviridae, was approved by the ICTV in 2015 [39,66].

Their assembly mechanism was finally unraveled in 2019 when the cryo-EM structures of the latest discovered membrane-containing archaeal Haloarcula californiae icosahedral virus 1 (HCIV-1), HHIV-2 and SH1 were determined at ~3.8 Å resolution (Figure 4A,B) [67,68]. The virions’ structures showed two distinct proteins with an α- and an α/β-fold located beneath the two- and three-tower capsomers, respectively, and homopentameric membrane proteins at the vertices [67]. They orchestrate the positioning relative to the membrane of penton proteins (VP9) and the pre-formed vertical single β-barrel MCP heterodimers (VP4 and VP7), acting as global-positioning-system (GPS) proteins [67]. These heterodimers (at least in HCIV-1, HHIV-2 and SH1) mimic the relative angular orientation of the individual β-barrels in DJR MCPs of the PRD1-adenovirus lineage (Figure 4B,C). However, the characteristics FG1-α and FG2-α helices which are common motifs in all DJR (Figure 1B, Figure 2B and Figure 3, left) are not equivalently oriented in the vSJR MCPs known so far (Figure 4B, bottom). Their different length and orientation, in particular for the vSJR VP7 may reflect the requirement of a fine tuning in the formation of heterodimers and consequently in the rotational registering and packing of the individual jelly-roll forming the pseudo-hexameric capsomers (Figure 4C). Structural superimposition of the vSJR MCPs onto the individual V1 and V2 jelly-roll of the minimalist DJR PM2 P2 and the vertical β-barrel proteins composing the penton of the vertical DJR and SJR viruses groups the vSJR used for the formation of the capsomers apart from those used for the penton, albeit with the PM2 P2 V1 and V2 closer to certain penton proteins (Figure 4D). This clustering supports the idea of a specialization of interactions according to their location within the capsid shell.

Thus, assembly of viruses with two vertical single β-barrel MCPs relies, apart from the protein at the vertices, on homo- and/or hetero-dimerization motifs in the MCPs and on additional molecular guiding proteins in between the capsid protein shell and the outer membrane leaflet. The fusion event of the two consecutive MCP genes leading to the double jelly rolls not only simplify the assembly but more importantly appear to represent so far the sine qua non for the uncoupling of the assembly from the presence of the membrane as seen in adenovirus [67].

It remains to be seen whether one day we will identify a virus with only one vertical single jelly-roll MCP capable of forming pseudo-hexameric capsomers prefiguring the assembly of viruses with two vertical single jelly rolls.

5. The Open Question of A Structure-Based Classification of Enveloped Viruses and the Example of Current SARS-CoV-2

The identification of a structural “self” in enveloped viruses remains a challenge [5,38]. While glycoproteins decorating the viral envelope are usually categorized in three structural classes of fusion proteins (I–III) [69], their fold is a weak determinant for inferring distant viral relationships (Figure 5A). Firstly, it has been shown that class II viral glycoproteins have counterparts in cellular proteins, demonstrating the possibility of horizontal gene exchange [70], and secondly, viral glycoproteins are under environmental selection pressure for the cellular receptors and thus inclined to higher sequence and/or structural variability. This is noticeable when comparing the available structures of the receptor-binding domain (RBD) within the spike glycoprotein S (a class I viral fusogen) across the members of the Coronaviridae family, as shown early this year [71,72].

However, with the emerging of SARS-CoV-2, a further manifestation of evolution at work, the power of sequence and structural analysis has revealed in record time the determinants for the recognition of the RBD for the human angiotensin-converting enzyme 2 (ACE2) [74,75]. The known RBDs adopt either a twisted five-stranded antiparallel β-sheet with short connecting helices and loops or a β-sandwich (Figure 5B). Superimposition of the available RBD-receptor complexes highlights the loops connecting the β-strands as the regions more susceptible to retain mutations (so-called receptor-binding-motifs) that render the RBD the master key for operating different cellular “locks” (e.g., ACE2, aminopeptidase N (APN), dipeptidyl peptidase 4 (DDP4) but also 9-O-acetylated neuraminic acid, carcinoembryonic antigen-related adhesion molecule 1, heparin sulfate, and α-2,3-linked sialic acid [76]) (Figure 5B,C).

Thus, the flexible loops not strictly possessing a structural role in the (RBD) protein fold are more prone to accommodate amino acid changes and possible elaborations (Figure 5B,C) as similarly observed in the conformational diversity of the connecting loops at the top of the vDJR/vSJR of the PRD1-adenovirus lineage (Figure 2B).

In the search of “self” candidates, the fold of the nucleocapsid protein has also been considered. In the case of negative-sense RNA viruses two distinct α-helical folds for this protein could cluster apart viruses with segmented genome from those with non-segmented genome (Figure 5C, left) [38]. For other enveloped viruses, a classification based on the nucleocapsid protein structure remains arduous. Therefore, whether the nucleocapsid is a reliable self for inferring common ancestry remains to be seen.

As for positive-sense ssRNA SARS-CoV-2 virus, the RNA binding-domain and the C-terminal domain of the nucleocapsid have been structurally determined but neither of them displays a topological resemblance to those of other viral families (Figure 5C, right) [73]. Its sequence, however, is highly conserved among human coronaviruses (25% < identity < 91%) making the nucleocapsid a cross-reactive antigen in immunological tests of patient sera as opposed to the specificity shown by the SARS-CoV-2 RBD antigen (unpublished).

Finally, all RNA viruses (enveloped and not) and reverse-transcribing viruses encode respectively an RNA-dependent RNA polymerase (RdRp) and a reverse-transcriptase, which allows for these genes to be used as markers in phylogenomic analyses [40]. Both above enzymes as well as the gene encoding for DJR MCPs have been suggested as “hallmark genes” for the virus classification currently adopted by the ICTV [45].

6. Conclusions

The journey of discovering evolutionary relationships through a structure-based approach across different viruses has not stopped since the initial observation that some animal and plant RNA viruses MCPs—with limited sequence identity—share a similar fold [20]. In fact, this journey is experiencing a rapid acceleration with the increased capabilities in genome sequencing, in structure determination and in more sensitive phylogenetic methods.

Deciphering sequence and structural homologies does not only emerge from the need of the human mind to categorize the complexity that surrounds us as a means to understand it, but also from its power in impacting on the fight against viral diseases.