Viruses are ubiquitous companions of cellular life forms: it appears that every cellular organism studied has its own viruses or, at least, virus-like selfish genetic elements 1. Recent environmental studies have shown that viruses, primarily, bacteriophages, are "most abundant biological entities on the planet" 2, with the total number of virus particles exceeding the number of cells by at least an order of magnitude 34. Viruses actively move between biomes and are thought to be major agents of evolution by virtue of their capacity to operate as vehicles of horizontal gene transfer (HGT) 5.

A remarkable feature of viruses is the diversity of their genetic cycles, in a sharp contrast to the uniformity of the cellular genetic cycle 6789 (Fig. 1). Viruses with different genome strategies span a vast range of genome sizes (the genomes of the largest known virus, the mimivirus, and the smallest viruses, e.g., circoviruses, differ by three orders of magnitude) and show a non-uniform and non-trivial distribution among the host taxa (Fig. 1). For example, the extraordinary diversity of double-stranded (ds) DNA bacteriophages is in a stark contrast to the absence of bona fide dsDNA viruses in plants. Conversely, RNA viruses are extremely abundant and diverse in plants and animals but are currently represented by only two compact families in bacteria, and so far have not been detected in archaea (Fig. 1).

Given the variety of genetic strategies, genome complexity, and global ecology of viruses, the problem of virus evolution inevitably digresses into a web of interlocking questions. What qualifies as a virus? Are viruses as a whole monophyletic, i.e., ultimately descend from a single primordial virus or polyphyletic, i.e., have multiple origins? If viruses are polyphyletic, how many independent lineages are there? What is the age distribution of different groups of viruses – are they ancient or have they been emerging continuously throughout life's evolution? And, perhaps, the most fundamental questions of all: what is the ultimate origin of viruses and what are the relationships between evolution of viruses and cellular life forms? The recent advances of comparative genomics create the unprecedented opportunity to start tackling these issues by inferring some of the likely answers from sequence and structure data analysis. Here, we address several of these questions but, primarily, the last two, most general ones, in an attempt to outline a coherent scenario of virus origin and evolution and delineate the connections between the evolution of viruses and cellular life forms.

Comparative genomics provides no evidence of a monophyletic origin of all viruses. Many virus groups simply share no common genes, effectively, ruling out any conventional notion of common origin. When applied to viruses, the notion of "common genes" is not a simple one because commonality is not necessarily limited to clear-cut orthologous relationships between genes that translate into highly significant sequence similarity. Instead, as discussed in the next sections, distant homologous relationships among viral proteins and between viral proteins and their homologs from cellular life forms could convey more complex but important messages on evolution of viruses. This complexity notwithstanding, cases of major virus groups abound that either share no homologous genes under any definition or have in common only distantly related domains with obviously distinct evolutionary trajectories. For example, most of the viruses of hyperthermophilic crenarchaea have literally no genes in common with any other viruses 1011, whereas RNA viruses share with DNA viruses and plasmids that replicate via the rolling circle mechanism only extremely distant domains in their respective replication proteins.

By contrast, monophyly of several large classes of viruses, including vast assemblages of RNA viruses and complex DNA viruses, can be demonstrated with confidence (Table 1). Some of these monophyletic classes of viruses cross the boundaries set by genome strategies: thus, the retroid class includes both RNA viruses and viruses, mobile elements, and plasmids with DNA genomes, and the rolling circle replication (RCR) class combines ssDNA and dsDNA viruses and plasmids. Furthermore, based on similarities in the structure of RNA replication complexes, along with the presence of homologous, even if distant, replication enzymes, it has been suggested that positive-strand RNA viruses, double-stranded RNA viruses, and retroid elements all have a common origin 9. On the whole, however, the conclusion seems inevitable that viruses comprise several distinct lines of descent (Table 1).

Sequence analysis of viral proteins reveals several categories of virus genes that markedly differ in their provenance. The optimal granularity of classification might be subject to debate but at least 5 classes that can be assorted into three larger categories seem to be readily distinguishable.

Genes with readily detectable homologs in cellular life forms

1. Genes with closely related homologs in cellular organisms (typically, the host of the given virus) present in a narrow group of viruses.

2. Genes that are conserved within a major group of viruses or even several groups and have relatively distant cellular homologs.

Virus-specific genes

3. ORFans, i.e., genes without detectable homologs except, possibly, in closely related viruses.

4. Virus-specific genes that are conserved in a (relatively) broad group of viruses but have no detectable homologs in cellular life forms.

Viral hallmark genes

5. Genes shared by many diverse groups of viruses, with only distant homologs in cellular organisms, and with strong indications of monophyly of all viral members of the respective gene families, – we would like to coin the phrase "viral hallmark genes" to denote these genes that can be viewed as distinguishing characters of the "virus state".

The relative contributions of each of these classes of genes to the gene sets of different viruses strongly depend on the viral genome size which differs by more than three orders of magnitude. Viruses with small genomes, such as most of the RNA viruses, often have only a few genes, the majority of which belong to the hallmark class. By contrast, in viruses with large genomes, e.g., poxviruses, all 5 classes are broadly represented. In order to illustrate the diversity of viral "genomescapes" more concretely, we show in Table 2 the decomposition of the gene sets of three viruses with a small, an intermediate-sized, and a large genomes, respectively, into the 5 classes. Notably, moderate-sized and large genomes of bacteriophages and archaeal viruses are dominated by ORFans that often comprise >80% of the genes in these viruses. Rapidly evolving phage ORFans are thought to supply many, if not most, of the ORFans found in prokaryotic genomes (the lack of detectable sequence conservation notwithstanding), hence playing an important role in evolution of prokaryotes 12.

The evolutionary origins of the 5 classes of viral genes are likely to be very different. The least controversial are the two classes of genes with readily detectable homologs from cellular life forms that appear to represent, respectively, relatively recent (class 1) and ancient (class 2) acquisitions from the genomes of the cellular hosts. Where do virus-specific genes come from is a much harder question. In the absence of direct evidence, the default hypothesis, probably, should be that these genes evolved from cellular genes as a result of dramatic acceleration of evolution linked to the emergence of new, virus-specific functions, such that all traces of the ancestral relationships are obliterated. This notion is compatible with the fact that many, probably, most class 4 genes (virus-specific genes conserved within a group of viruses) are virion components (e.g., see the vaccinia virus case in Table 2), a quintessential viral function. The hallmark genes that cross the barriers between extremely diverse virus lineages seem to be of the greatest interest and relevance for the problem of the ultimate origins of viruses, at least, in the context of the long argument we attempt to develop here. Thus, we discuss the distribution among viruses, evolution and significance of these genes in a separate section.

Although there are no traceable vertical relationships between large groups of viruses outside the major classes listed in Table 1, a considerable number of genes that encode proteins with key roles in genome replication, expression, and encapsidation are shared by overlapping arrays of seemingly unrelated groups of viruses. As already noted above, some of these widespread viral genes are genuine viral hallmarks that are found in a variety of diverse viruses (although never in all viruses) but not in cellular life forms except as easily recognizable proviruses or mobile elements (Table 3, Fig. 1). The two proteins that are most widely dispersed among viruses are the jelly-roll capsid protein (JRC) 131415 and the superfamily 3 helicase (S3H) 16. Each of these proteins crosses the boundary between RNA and DNA viruses and spans an astonishing range of virus groups, from some of the smallest positive-strand RNA viruses to the nucleo-cytoplasmic large DNA viruses (NCLDV), the class of viruses that includes the giant mimivirus (Table 3). Other proteins listed in Table 3 are not as common as JRC or S3H but still form multiple, unexpected connections between groups of viruses that otherwise appear to be unrelated. A case in point is the rolling circle replication (RCR) initiation endonuclease (RCRE) which unites a great variety of small ss and dsDNA replicons, including viruses, plasmids, and transposable elements that reproduce in animals, plants, bacteria, and archaea 17181920. Notably, a more recent and more careful analysis has shown that the DNA-binding domain of the replication protein of polyoma and papilloma viruses (e.g., the T antigen of SV40) is a derived form of the RCRE that has lost the catalytic amino acid residues 18. Thus, through this detailed analysis of one of the hallmark proteins, the well-known connection between a variety of small ssDNA-replicons is extended to a group of similar-sized dsDNA-replicons. A similar expansion of the set of viral groups covered by a particular hallmark gene resulted from the detailed analyses of the packaging ATPase and the archaeo-eukaryotic primase (Table 3).

Replication of positive-strand RNA viruses, dsRNA viruses, negative-strand RNA viruses, and retroid viruses/elements is catalyzed by another idiosyncratic class of viral enzymes, RNA-dependent RNA polymerases (RdRp) and reverse transcriptases (RT). The positive-strand RNA virus RdRp and the RT form a monophyletic cluster within the vast class of the so-called palm-domains that are characteristic of numerous polymerases 21222324. The RdRps of dsRNA viruses and negative-strand RNA viruses are likely to be highly diverged derivatives of the same polymerase domain, an old conjecture 242526 that has been vindicated by the recent determination of the structure of a dsRNA bacteriophage RdRp 2728.

The palm domain is likely to be the primordial protein polymerase that emerged from the RNA world where nucleotide polymerization was catalyzed by ribozymes 29. This is supported not only by the wide spread of this domain in modern life forms but also by the structural and, by inference, evolutionary link between the palm domain and the RNA-recognition-motif (RRM) domain, an ancient RNA-binding domain that might have, initially, facilitated replication of ribozymes 30. The palm-domain RdRps and RTs are excluded from the regular life cycles of cellular life forms, although most eukaryotic genomes encompass numerous copies of RT-containing retroelements, and prokaryotes have some such elements as well 3132. These elements, however, are selfish and, from the evolutionary standpoint, virus-like. The most notable incursion of an RT into the cellular domain is the catalytic subunit of the eukaryotic telomerase, the essential enzyme that is involved in the replication of chromosome ends 3334.

The list of viral hallmark genes given in Table 3 is a conservative one. There well might be other genes that merit the hallmark status but for which clear evidence is hard to come up with. An important case in point is the B-family DNA polymerase that is the main replication enzyme of numerous dsDNA viruses of bacteria and eukaryotes. Homologs of these DNA polymerases are found in all archaeal and eukaryotic genomes, so that monophyly of all viral polymerases does not seem to be demonstrable in phylogenetic analyses 3536. However, this potentially could be explained by relatively (with respect to cellular homologs) fast evolution of the polymerases in various viral lineages, which would obscure their common origin. Furthermore, monophyly of the polymerases of all viruses that employ a protein-primed mechanism of dsDNA replication (animal adenoviruses and tailed phages like PRD1 or φ29) has been claimed 37. Thus, although this currently cannot be shown convincingly, it seems possible (and, as discussed below, even likely) that the DNA polymerase is a viral hallmark gene in disguise. More generally, further sequencing of viral genomes, combined with comprehensive comparative analysis, might reveal additional genes that, despite relatively limited spread among viral lineages, will qualify as hallmark genes.

The combination of features of viral hallmark proteins is highly unusual and demands an evolutionary explanation. Indeed, the hallmark genes are, without exception, responsible for essential, central aspects of the viral life cycles, including genome replication, virion formation, and packaging of the genome DNA into the virion (Table 3). These genes span sets of extremely diverse classes of viruses, often possessing different reproduction strategies and differing by three orders of magnitude in genome size. Finally, all viral hallmark genes have remote homologs in cellular life forms (Table 3) but the viral versions appear to be monophyletic.

Three hypotheses on the origin of viral hallmark genes immediately come to mind. The first possibility is that the notion of hallmark virus proteins is based on an artifact. The argument, that is commonly invoked in discussions of unexpected patterns of homologous relationships and could well be waged against the notion of viral hallmark proteins, is that genuine orthologs of these proteins (direct evolutionary counterparts, typically, with the same function) actually do exist in cellular life forms but are not detectable due to rapid sequence divergence between viral and cellular proteins. However, this reasoning does not seem to survive closer scrutiny. Firstly, the conservation of the hallmark proteins in extremely diverse classes of viruses with widely different replication/expression strategies (Table 3) but not in cellular life forms is hardly compatible with the rapid divergence interpretation. Indeed, for this to be the case, acceleration of evolution of the hallmark genes in diverse classes of viruses should occur in such a manner that the similarity between viral proteins survived, whereas the similarity between viral proteins and their hypothetical cellular orthologs vanished. Parallel conservation of hallmark protein sequences might be perceived in the case of structural protein, such as JRC, but is hardly imaginable for proteins involved in replication of structurally very different genomes, such as S3H that is conserved among viruses with RNA, ssDNA, and dsDNA genomes. Furthermore, for most of the hallmark proteins, distant and functionally distinct homologs from cellular organisms are detectable (S3H and viral RNA-dependent polymerases are the primary examples) which makes the existence of elusive orthologs extremely unlikely.

The other two hypotheses accept the hallmark viral proteins as reality but offer contrasting evolutionary scenarios to account for their existence and spread.

1. One hypothesis would posit that the hallmark genes comprise the heritage of a "last universal common ancestor of viruses" (LUCAV). This scenario implies that, despite all evidence to the contrary (see above) all extant viruses are monophyletic, although their subsequent evolution involved massive gene loss in some lineages as well as extensive acquisition of new genes from the hosts in others.

2. By contrast, under the hypothesis of polyphyletic origin of viruses, the spread of the hallmark genes across the range of virus groups could be explained by horizontal gene transfer (HGT).

Upon closer inspection, none of these hypotheses seems to be a viable general explanation for the existence and distribution of the viral hallmark genes. Indeed, the relatively small number and the mosaic spread of the hallmark genes (Table 3) do not seem to be conducive to the LUCAV notion although it is apparent that a great number of diverse viruses, if not all of them, share some common history. Conversely, the extremely distant similarity between the hallmark proteins from diverse virus groups with dramatically different replication strategies is poorly compatible with an HGT scenario.

Here, we outline a scenario of virus origin and evolution that does not involve a LUCAV but integrates aspects of the common origin and HGT hypotheses and is naturally linked to specific models of evolution of cells. The simplest explanation of the fact that the hallmark proteins involved in viral replication and virion formation are present in a broad variety of viruses but not in any cellular life forms is that the latter never had these genes in the first place. The alternative that we consider most likely is that the hallmark genes antedate cells and descend directly from a primordial gene pool. It is thought that, in such a primordial pool, selection would act primarily on functions directly involved in replication 3839 which is compatible with the properties of the majority of hallmark genes (Table 3). Given the spread of the hallmark genes among numerous groups of dramatically different viruses, a crucial corollary is that the major lineages of viruses themselves derive from the same, precellular stage of evolution. This corollary serves as the foundation for the concept of an ancient Virus World, which we envisage as an uninterrupted flow of genetic information through an enormous variety of selfish elements, from the precellular stage of evolution to this day; we discuss the Virus World in the rest of this article.

Before we discuss the full scope of the emerging concept of the origin of viruses from the precellular gene pool, it is necessary to briefly examine the existing trains of thought on virus origin and evolution. Traditionally, these ideas have revolved around three themes: i) origin of viruses from primordial genetic elements, ii) degeneration of unicellular organisms to the virus state, and iii) "escaped genes" hypotheses deriving viruses from genes of cellular organisms that have switched to the selfish mode of reproduction (reviewed in 404142434445) (Table 4). Historically, it is remarkable that Felix d'Herelle, the discoverer of bacteriophages and one of the founders of virology, proposed that phages might be evolutionary precursors of cells as early as 1922 46. Furthermore, in J.B.S. Haldane's 1928 classic on the origin of life 47, an early, "viral" stage of evolution is considered as an integral part of the proposed scenario for the emergence of the first life forms from the primary soup (we revisit Haldane's prescient speculation toward the end of this article). However, the "primordial" hypothesis is habitually dismissed on the grounds that all extant viruses are intracellular parasites, so viruses could not exist before the emergence of modern-type cells although antiquity of viruses has been propounded based on the lack of cellular homologs for many virus genes 4348495051. By contrast, the high prevalence of host-related genes (as opposed to virus-specific genes) in many viruses (particularly, those with large genomes) might be construed as support for the "escaped genes" or even the "cell degeneration" hypotheses.

The existence of the hallmark virus genes seems to effectively falsify both the cell degeneration and the escaped-genes concepts of viral evolution. With regard to the cell degeneration hypothesis, let us consider the NCLDV, the class of large viruses to which the cell degeneration concept might most readily apply and, indeed, was, in the wake of the discovery of the giant mimivirus 505152. Among the 11 signature genes that are shared by all NCLDVs (53 and Table 1), three crucial ones (JRC, S3H, and the packaging ATPase) are virus hallmark genes. A clear inference is that even the simplest, ancestral NCLDV would not be functional without these genes. However, cellular derivation of this ancestral NCLDV would have to invoke decidedly non-parsimonious, ad hoc scenarios, such as concerted loss of all hallmark genes from all known cellular life forms or their derivation from an extinct major lineage of cell evolution. The same line of logic essentially refutes the escaped genes concept inasmuch as the hallmark genes had no cellular "home" to escape from. Again, to save "escaped genes", an extinct cellular domain would have to be postulated.

Two recent conceptual developments in the study of origin and evolution of viruses deserve special attention (Table 4; see also discussion below). First, Bamford and coworkers 131415 and, independently, Johnson and coworkers 54 capitalized on the conservation of the structure of the jelly-roll capsid protein in a wide variety of viruses to propose the idea of an ancient virus lineage spanning all three domains of cellular life (archaea, bacteria, and eukaryotes). Second, Forterre presented an elaborate scheme of virus-cell coevolution from the earliest stages of life's evolution. Under this concept, viruses emerged independently within three lineages of RNA-based cells (the progenitors of archaea, bacteria, and eukaryotes) and "invented" DNA replication that was subsequently captured from different viruses by each type of host cells, in three independent transitions to DNA genomes 4355.

A crucial, even if fairly obvious, aspect of viral evolution is that it is inextricably linked to the evolution of the hosts which, when traced back to the earliest stages of life's evolution, attests to the necessity to consider scenarios of virus origins in conjunction with models for the origin and early evolution of cells. This dramatically ups the ante for the study of virus evolution, bringing it to the center stage of evolutionary biology 5657. Therein, however, seem to lie some of the major problems encountered by the current hypotheses of virus evolution (Table 4). The concept of an ancient virus lineage advocated by Bamford and coworkers, in addition to being based on the broad spread of a single gene (JRC), which is construed as the "self" of a virus 13, simply leaves a glaring gap as the connection between viral and cellular evolution is not considered such that it remains unclear in what kind of cellular environment the early evolution of the primordial viral lineage took place.

By contrast, Forterre's concept is embedded within a specific scenario of cellular evolution, and we believe that this is the valid approach to the analysis of the origins and evolution of viruses and cells – the only chance to achieve understanding of these difficult problems is to consider them in conjunction. However, the specific scenario of cellular evolution favored by Forterre appears to be poorly compatible with the results of comparative genomics, thus compromising the model of virus evolution associated with it. Indeed, Forterre considers three lineages of RNA-cells evolving from the Last Universal Common Ancestor (LUCA) of the known life forms and giving rise to bacteria, archaea, and eukaryotes, with the transition to DNA-cells mediated by domain-specific viruses. This scenario, while complete with respect to the evolution of both cells and viruses, encounters at least three major, probably, insurmountable problems. First, it is dubious at best that the combination of complexity and genetic stability required of an even a minimal cell – a complement of a few hundred genes that is accurately transmitted over many cellular generations – is attainable with an RNA genome. Indeed, given the inherent instability of RNA, such a genome would have to consist of several hundred RNA molecules. Accurate partitioning of this multipartite genome between daughter cells would require an RNA segregation system of unprecedented precision; thus, faithful vertical inheritance of the genome is hardly imaginable. Stochastic segregation of RNA segments in a polyploid cell might seem to be a straightforward solution to this problem but a quick calculation shows that this is unfeasible. Indeed, if the probability of a the two daughter RNA segments being segregated into the two daughter cells is 1\2, then the probability that, say, 100 RNA segments in a reasonable multipartite genome are all correctly segregated (i.e., none of the segments is missing in either of the daughter cells) is (1/2)¹⁰⁰ ≈ 10^-30. Thus, ploidy of ~10³⁰ would be required for accurate genome segregation; in other words, the RNA cell would have to contain many tons of RNA. Thus, it is much more likely that the first fully-fledged cells already had DNA genomes resembling (even if quantitatively simpler than) those of modern archaea and bacteria. Forterre's proposal is based on the well-known observation that the DNA replication systems of archaea and bacteria are, largely, unrelated, making it unlikely that LUCA had a DNA genome 585960. We believe that a far more plausible implication of the disparity of DNA replication machineries in archaea and eukaryotes is a non-cellular LUCA, a hypothesis that finds crucial support in the fact that the membrane lipids and membrane biogenesis systems, as well as cell walls, are also distinct and, largely, unrelated in archaea and bacteria 39616263.

For efforts on reconstructing the non-cellular LUCA, a key guiding principle is that, although modern-type cells, presumably, did not exist at this stage of evolution, some form of compartmentalization was required to ensure concentrations of substrates and genetic elements sufficient for effective replication and, consequently, evolution. Furthermore, as discussed in greater detail below, compartmentalization is a necessary condition of selection among evolving ensembles of genetic elements. Following this principle, a specific model has been elaborated under which early evolution, from abiogenic syntheses of complex organic molecules to the emergence of archaeal and bacterial cells, unraveled within networks of inorganic compartments that are found at hydrothermal vents and consists, primarily, of iron sulfide 3963. Here, in order to be concrete, we attempt a reconstruction of the earliest events in the evolution of primordial virus-like entities within the framework of this model although our general conclusions do not seem to hinge on any particular model of organization of the ancient, precellular life.

Second, the model by viral overtake meets a stumbling block: if three different viruses brought the three distinct DNA replication machineries into the three cell lineages, why have not the hallmark genes that are essential for viral DNA replication, in particular S3H and the viral-type primase, entered the genomes of any of those cellular lineages? One could argue that, in each case, an unusual virus, not carrying any of these hallmark genes, was involved, but this happening three times independently stretches credulity.

The third major issue that is ignored in Forterre's hypothesis and similar, three-domain scenarios of cellular evolution 6465, is the readily demonstrable relationship between eukaryotes and the two prokaryotic domains. This relationship follows the divide between the two principal functional categories of genes in the eukaryotic cell, the informational genes that are, almost invariably, most closely related to archaeal homologs, and the operational genes that are of bacterial provenance 666768. By far the simplest, most parsimonious explanation of this dichotomy is that the eukaryotic cell emerged as a result of a symbiosis between an archaeon and a bacterium. Given the overwhelming evidence of the origin of mitochondria and related organelles (hydrogenosomes and mitosomes) from α-proteobacteria, the nature of the partners and the direction of the symbiosis appear clear: an α-proteobacterium invaded an archaeal cell 697071. In addition to the genomic evidence, there is a clear biochemical rationale for the symbiosis to actually comprise the onset of eukaryogenesis such that the invading α-proteobacterium became an anaerobic symbiont that originally supplied hydrogen to the methanogenic, archaeal host and subsequently gave rise to aerobic mitochondria 71. The massive transfer of the symbiont's genes to the host genome gave rise to the mosaic provenance of the eukaryotic genomes. The argument against the symbiotic models of eukaryogenesis and for three-domain models stems primarily from the existence of numerous eukaryote-specific proteins (ESPs) without readily detectable prokaryotic homologs 647273. However, the abundance of ESPs hardly can be taken as evidence of a third domain of life, distinct from archaea and bacteria, and comprising the pre-symbiosis eukaryotic line of descent. First, although many of the ESPs are proteins with important biological functions, they do not belong to the core of either the informational or the metabolic proteins repertoires of eukaryotes that are of archaeal and bacterial descent, respectively. Second, more often than not, prokaryotic homologs of an apparent ESP can be identified by using sensitive techniques of protein motif analysis and, particularly, structural comparison 74. Thus, it seems likely that many, if not most, ESPs are cases of accelerated evolution in eukaryotes which accompanies the emergence of novel, eukaryote-specific functional systems, such as the cytoskeleton and ubiquitin signaling. The prevalence and nature of "true" eukaryotic innovations, i.e., proteins that evolved through processes other than descent with (perhaps, radical) modification from archaeal or bacterial proteins, remain to be characterized. Clearly, however, even if such novelties turn out to be numerically prominent, they are, typically, involved in ancillary functions and hardly can be construed as the heritage of a distinct primary line of cell evolution 74.

In the rest of this article, we consider evolution of viruses in conjunction with these two central concepts of cellular evolution: i) the existence of an early non-cellular but confined stage in life's evolution, that probably encompassed LUCA, and ii) the origin of the eukaryotic cell as a result of fusion of an archaeon and a bacterium. We argue that the scenario of virus origin and evolution informed by comparative genomics forms a coherent whole with these notions of early cell evolution and provides supportive feedback for them.

Taken together, all these lines of evidence and reasoning suggest that the principal classes of prokaryotic viruses, including positive-strand RNA viruses, retroid elements, and several groups of DNA viruses, emerged within the primordial genetic pool where mixing and matching of diverse genetic elements was incomparably more extensive than it is in any modern biological community 3975 (Fig. 2).

As indicated above, we present a scenario of viral origins under the recent model of the emergence of cells and genomes within networks of inorganic compartments 3963. These compartments are envisaged being inhabited by diverse populations of genetic elements, initially, segments of self-replicating RNA, subsequently, larger and more complex RNA molecules encompassing one or a few protein-coding genes, and later yet, also DNA segments of gradually increasing size. Thus, early life forms, including LUCA, are perceived as ensembles of genetic elements inhabiting a system of inorganic compartments. This model explains the lack of homology between the membranes, membrane biogenesis systems, and the DNA replication machineries of archaea and bacteria by delineating a LUCA that had neither a membrane nor DNA replication. The model also outlines the processes that might have enabled selection and evolutionary change in such a system. Under this scenario, a transition gradually took place from selection at the level of individual genetic elements to selection for ensembles of such elements encoding both enzymes directly involved in replication and proteins responsible for accessory functions, such as translation and nucleic acid precursor synthesis. From selection for gene ensembles, there is a direct path to selection for compartment contents such that compartments sustaining rapid replication of genetic elements would infect adjacent compartment and, effectively, propagate their "genomes" 39.

This model implies that, at the early stages of evolution, including the LUCA stage, the entire genetic system was, in a sense, "virus-like". Initially, all RNA segments in the population would be completely selfish, and there would be no distinction between parasitic, "viral" elements and those that would give rise to genomes of cellular life forms. However, this distinction would emerge as soon as the first "selfish cooperatives" – relatively stable ensembles of co-inherited genetic elements 39 – evolve. Inevitably, the selfish cooperatives would harbor parasitic genetic elements that would divert the resources of the ensemble for their own replication. The separation of parasitic elements from cooperators would be cemented by the emergence of physical linkage of the latter: genes that encode components of the replication machinery and are physically linked to genes for accessory functions would be locked into ensembles of cooperators, whereas solitary genes (or small gene arrays) encoding replication functions would occupy the parasitic niche. From a complementary standpoint, within the inorganic compartment model, the distinction between the progenitors of viruses and the ancestors of cellular life forms may be described as one between primary infectious agents, that routinely moved between compartments, and increasingly complex and "sedentary" ensembles of selfish cooperators that would spill from one compartment into another on much rarer occasions.

Like other models of the early stages of evolution of biological complexity, this scenario faces the problem of takeover by selfish elements 767778. If the primordial parasites become too aggressive, they would kill off their hosts within a compartment and could survive only by infecting a new compartment. Devastating "epidemics" sweeping through entire networks and eventually eliminating all their content are imaginable, and indeed, this might have been the fate of many, if not most, primordial "organisms". The evolution of those primordial life forms that survived would involve, firstly, emergence of temperate virus-like agents that do not kill the host, and secondly, early invention of primitive defense mechanisms, likely, based on RNA interference (RNAi). The ubiquity of both temperate selfish elements and RNAi-based defense systems 798081 in all life forms is compatible with the origin of these phenomena at a very early stage of evolution.

From the very beginning, the parasites would have different levels of genome complexity, from the minimal replicable unit, like in viroids, to more complex entities that might encode one or more proteins facilitating their own replication. Capsid proteins, in particular, JRC were the crucial innovation that might have led from virus-like genetic elements to entities resembling true viruses (as long as viruses are defined as intracellular parasites, we are not entitled to speak of viruses evolving prior to the appearance of membrane-bounded cells; however, this is a semantic problem only). Under this model, the advantage conferred on a selfish genetic element by encapsidation is obvious: stabilization of the genome and increased likelihood of transfer between compartments and even between different networks of compartments. Given this advantage of the capsid and the extensive reassortment and recombination that is thought to have reigned in the primordial gene pool, it is not surprising that different genes that could contribute to virus replication, such as polymerases, helicases, ATPases, and nucleases, combined with the JRC gene on multiple occasions, and several such combinations were propagated by selection to give rise to distinct viral lineages. Alternative capsid proteins that form helical capsids in diverse RNA and DNA viruses 82 might have evolved already at this early stage of evolution. However, the capsid is by no means a pre-requisite of evolutionary success for selfish genetic elements. Indeed, virus-like entities, such as the prokaryotic retroelements (retrons and group II introns) and various small plasmids, have maintained the capsid-less, selfish life style from their inferred emergence within the primordial gene pool to this day.

Thus, under this scenario, virus-like entities are older than typical, modern-type cells: agents resembling modern viruses, some of them even with capsids, existed before the emergence of membrane-bounded cells with large dsDNA genomes. Moreover, these agents would, effectively, have a virus-like life style because the primordial life forms are perceived as non-cellular but compartmentalized and would support ancient virus-like agents moving between compartments much like modern cells support viruses. The origin of ancient virus-like agents, the ancestors of the major classes of modern viruses, would follow the stages of evolution of genetic systems: RNA-only selfish agents resembling group I introns or viroids might have emerged in the primitive RNA world. The RNA-protein world begot RNA viruses, the postulated stage of an RT-based, mixed RNA-DNA system spawned retroid elements, and the DNA stage yielded several lineages of DNA viruses (Fig. 2). It appears likely that the primordial gene pool harbored an extraordinary variety of virus-like entities. When archaeal and bacterial cells escaped the compartment networks, they would carry with them only a fraction of these viruses that subsequently evolved in the cellular context to produce the modern diversity of the virus world.

General considerations on the course of evolution from the RNA world to modern-type systems, together with the spread of different virus classes among modern life forms (Table 1 and Fig. 1), suggest that the following classes of virus-like selfish elements emerged from the primordial gene pool and infected the first bacteria and/or archaea (Fig. 2): i) RNA-only elements, such as group I introns, ii) positive-strand RNA viruses (and, possibly, dsRNA viruses as well), iii) retroid elements similar to prokaryotic retrons and group II introns, iv) small RCR replicons, v) at least one, and probably, more groups of larger, dsDNA viruses ancestral to the order Caudovirales (tailed phages). Incidentally, the logic of this inference suggests that the dsDNA viruses emerging from the primordial gene pool already possessed DNA polymerases and, accordingly, that the DNA polymerase is a true viral hallmark gene, notwithstanding the difficulty in obtaining strong evidence of this (see above).

To conclude the discussion of viral origin from the primordial genetic pool, we would like to emphasize, once again, the tight coupling between the earliest stages of viral and cellular evolution. Indeed, the presence of viral hallmark genes in extremely diverse groups of viruses, combined with a variety of other genes, constitutes strong evidence of extensive reassortment/recombination associated with the origin of viruses, which is best compatible with a primordial gene pool with rampant gene mixing and matching. Thus, viral comparative genomics seems to provide substantial support for the non-cellular model of early evolution. However, it should be noticed that the virus world concept does not necessarily require a non-cellular LUCA; the concept would survive even if LUCA was, actually, a cell. What is germane to our model is the existence of an advanced pre-cellular stage of evolution at which substantial genetic diversity was already attained; whether LUCA existed at that or at a later stage, while an extremely important and intriguing issue in itself, is not central to our argument.

Origin of the eukaryotic viruses is a distinct and fascinating problem. Two features of eukaryotic viruses are most relevant for this discussion:

i) with the sole exception of large dsDNA viruses, all major classes of viruses display greater diversity in eukaryotes than in prokaryotes;

ii) although eukaryotic viruses share a substantial number of genes with bacteriophages and other selfish genetic elements of prokaryotes, the relationships between prokaryotic and eukaryotic viral genomes are always complex, to the extent that direct, vertical links between specific groups of eukaryotic and prokaryotic viruses often are not traceable (Table 5).

This implicates the emerging eukaryotic cell as a second, after the primordial gene pool, melting pot of virus evolution, in which extensive mixing and matching of viral and cellular genes molded a new domain of the virus world (Fig. 3).

As discussed above, it seems most likely that the eukaryotic cell emerged via an archaeo-bacterial fusion. Specifically, the most parsimonious scenario seems to be that engulfment of an α-proteobacterium by a methanogenic archaeon, leading to the origin of the first, initially, anaerobic mitochondriate cell, had been the starting point of eukaryogenesis 7071. Conceivably, mitochondrial symbiogenesis set off a dramatic series of events that led to the emergence of the eukaryotic cell in a relatively short time. The mitochondria, probably, have spawned the invasion of group II introns into the host genes; this onslaught of introns could have been the driving force behind the origin of the nucleus 70. The mitochondria also donated numerous genes that integrated into the host genome, including genes coding for components of the essential organelles of the eukaryotic cells, such as the endoplasmic reticulum, the nucleus, and the bacterial-type plasma membrane that displaced the original archaeal membrane 6683. Furthermore, several bacteriophage genes, in all likelihood, from the mitochondrial endosymbiont's phages, have been recruited for replication and expression of the mitochondrial genome 8485. Even more notably, the catalytic subunit of the telomerase, the pan-eukaryotic enzyme that is essential for the replication of linear chromosomes appears to have evolved from a prokaryotic retroid element 3334. Thus, prokaryotic selfish elements apparently supplied at least one key component of the eukaryotic cell replication machinery.

The relationships between bacteriophage genes and those of eukaryotic viruses (Table 5) imply that the mitochondria and, possibly, other, transient endosymbionts also contributed many genes from their phage gene pool to the emerging eukaryotic viruses 5385. Based on the current knowledge of bacterial and archaeal virus genomics, bacteriophages of the endosymbiont(s) played a much greater role in the origin of eukaryotic viruses than archaeal selfish elements (Table 5). In particular, eukaryotic RNA viruses apparently could have been derived only from the respective phages inasmuch as no archaeal RNA viruses have been so far discovered. Retroid elements are also most characteristic of bacteria, α-proteobacteria in particular (those few retroids that have been discovered in isolated archaeal species are thought to be the results of relatively recent HGT from bacteria), such that the remarkable proliferation of these elements in eukaryotes, most likely, was spawned by mitochondrial endosymbiosis 86. The case of DNA viruses is more complicated because both bacteriophages and archaeal viruses and plasmids share genes with eukaryotic DNA viruses (Table 5). Thus, different groups of eukaryotic DNA viruses might have inherited genes from either bacteriophages or archaeal viruses (and other selfish elements), or a mixture thereof.

The pivotal contribution of prokaryotic viruses to the origin of eukaryotic viral genomes appears to be as strongly supported by the conservation of the hallmark genes across most of the virus world as the original emergence of viruses from the primordial gene pool. Indeed, since the viral hallmark genes have never become integral parts of the cellular gene pool, the only source from which eukaryotic viruses could inherit these essential genes was the gene pool of prokaryotic viruses, plasmids, and other selfish elements. Obviously, while the hallmark genes comprise a large fraction of the gene complement in small viruses, such as most of the RNA viruses and retroid viruses/elements, in large DNA viruses, these genes form only a small genomic kernel (see Table 2 and discussion above). Accordingly, as far as small viruses are concerned, the notion of direct evolutionary derivation of eukaryotic viruses/elements from prokaryotic counterparts might be justified. In particular, although positive-strand RNA viruses of eukaryotes share only one gene, the RdRp, with RNA bacteriophages, this gene occupies more than half of the genome coding capacity in the phages and the smallest eukaryotic viruses, and its product is either the sole virus-specific component of the replication machinery of the respective viruses or, at least, constitutes its catalytic core. This prominence of the conserved RdRp gene supports the view that bacterial and eukaryotic RNA viruses are linked by a direct, vertical relationship. The same logic could apply to retroid viruses/elements, where the common denominator is the RT, and to RCR elements (viruses and plasmids) that are unified by the presence of the RCRE. However, even small viruses of eukaryotes carry signs of recombination bringing together genes from different sources, including hallmark genes that are not found next to each other in prokaryotic viruses. An obvious case in point is the juxtaposition of the RdRp gene with the genes for JRC and/or S3H in eukaryotic positive-strand RNA viruses 87. The latter two genes are not seen in RNA bacteriophages, suggesting, however counter-intuitively, that DNA phages (and plasmids, in the case of S3H), many of which possess genes for these proteins, made major contributions to the evolution of eukaryotic RNA viruses. Even more dramatically, eukaryotic retroid viruses possess, in addition to the RT, a diverse array of genes of apparently different provenance, including RNAse H, a ubiquitous cellular enzyme whose exact origin in retroviruses is hard to infer 88, the integrase, probably derived from bacterial transposons 8990, the protease of likely mitochondrial origin 91, and the capsid/envelope proteins whose ancestry remains uncertain. Thus, the conclusion is inevitable that eukaryotic retroviruses and retrotransposons evolved through a complex series of recombination events between selfish elements of diverse origins 88 and, possibly, genes of cellular origin as well.

The large dsDNA viruses of eukaryotes are a different lot altogether, with several classes of genes of different origins and ages (see discussion above and Table 2). The conserved core in each lineage of large eukaryotic dsDNA viruses consists of a small set of hallmark genes along with some less common genes shared with bacteriophages, other bacterial selfish elements, and/or bacteria, and a few genes that appear to be unique to eukaryotes and have been acquired by the viruses at an early stage of eukaryogenesis. In addition to the core genes, each group of large viruses acquired numerous genes, mostly, from eukaryotes but, in some cases, also from bacteria; often, these genes play roles in virus-host interactions, and some of them seem to have been scavenged by viruses relatively recently (Table 2; 5392). Large viruses also have many "unique", lineage-specific genes the provenance of which remains uncertain. This "multilayer" genome organization of large viruses, along with the observations that even the hallmark genes often connect the same group of eukaryotic viruses to different prokaryotic selfish elements (Table 5), does not seem to support direct derivation of these viruses from individual phages.

The above data and considerations clearly point to a great amount of genomic perturbation during the emergence of eukaryotic viruses. However, from a more general perspective, one must remember the Darwinian principle that evolution can only proceed via a succession of manageable steps each of which is associated with an increase in fitness. Hence the emergence of eukaryotic viruses involves an apparent paradox: an uninterrupted succession of viruses must connect the prokaryotic and eukaryotic domains of the virus world, even if, in most cases, this succession cannot be unequivocally identified in terms of shared genes. From this standpoint, the notion that even a single shared protein, such as the JRC, represents a lineage of virus evolution 131415 appears relevant. However, other viral hallmark genes, as well as less widespread genes shared by phages and eukaryotic viruses (Table 5), comprise additional lines of descent connecting the prokaryotic domain of the virus world with the eukaryotic domain. These lines of viral gene evolution form multiple intersections in the melting pot of eukaryogenesis: the path between the two domains was not smooth but rather involved drastic remolding of the viral genomes. Certainly, this conclusion must be taken with the caveat that the available bacteriophage genomes account for but a tiny slice of the enormous phage diversity, so it is impossible to rule out that direct progenitors of the main groups of eukaryotic viruses are lurking in the unexplored parts of the phage domain. Still, given the numerous connections at the level of individual genes between prokaryotic and eukaryotic viruses discovered so far and the lack of wholesale genomic links, we venture to extrapolate that the future, massive sequencing of phage genomes brings more of the same: we will be able to identify phage roots of additional eukaryotic viral genes but (in most cases) not genome-wide vertical relationships.

As pointed out above, the representation of different genome strategies are dramatically different between prokaryotic and eukaryotic viruses, the foremost distinction being the higher prevalence of DNA viruses in prokaryotes contrasted by the preponderance of RNA viruses in eukaryotes (Fig. 1). The cause(s) of this disparity is a major puzzle in virology but consideration of the differences in the cellular organization of prokaryotes and eukaryotes might offer a clue. Conceivably, the emergence of the nucleus substantially changed the distribution of niches available for viruses by sequestering the machineries for host cell DNA replication, repair, and transcription within the nucleus and hence hampering the exploitation of these systems by DNA viruses, a mode of reproduction that is readily accessible to bacteriophages. Hence the relative decline in DNA virus diversity in eukaryotes and the massive spread of RNA viruses into the freed niches.

Thus, the birth of the eukaryotic virus realm can be envisaged as a turbulent process of mixing and matching of genes from several different sources, the most important one being, probably, the gene pool of prokaryotic viruses, in the melting pot of the emergent eukaryotic cell. At this juncture, we must, once again, stress the importance of the links between viral and cellular evolution and the coherence of the emerging scenarios. Indeed, the cataclysmic events of eukaryogenesis appear to have precipitated the birth of a new domain of the virus world. Of course, our models of the virus world evolution depend on the adopted scenario for evolution of cells; to wit, a "eukaryotes first" scenario 6465 would call for a very different perspective on viral evolution as well. As explicated above, comparative-genomic data do not seem to be compatible with this scheme at all but, instead, strongly favor an archaeo-bacterial fusion scenario which dictates the polarity of the virus world evolution discussed in this section. Moreover, the apparent mixed contribution of bacteriophages and archaeal selfish elements to the formation of the genomes of some eukaryotic viruses, such as the NCLDV (Table 5), provides supportive feedback for the fusion models of eukaryogenesis.

Our principal conclusions are encapsulated in the concept of the ancient Virus World. We envisage this world as a distinct flow of genes that emerged from within the primordial gene pool and retained its identity ever since, the numerous gene transfers between viral and cellular genomes notwithstanding. The notion of the virus world, one of its central tenets being intense gene mixing at the precellular stage of life's evolution, explains the spread of several essential genes (which we dubbed viral hallmarks) among apparently unrelated groups of viruses, observations that otherwise might seem completely baffling. The virus world concept blurs the distinction between monophyletic and polyphyletic origins of virus classes and implies a "third path". The major classes of viruses do not have a common origin in the traditional sense but neither are they unrelated, being connected through overlapping arrays of common genes, some of which are hallmarks of the Virus World.

The Virus World, apparently, remained continuous in time, in terms of uninterrupted inheritance of substantial sets of virus-specific genes, since its emergence from the primordial gene pool at the precellular stage of evolution to this day. The temporal continuity of the Virus World is complemented by its current interconnectedness as demonstrated by numerous observations of intervirus HGT, especially, the rampant gene exchange among phages 939495. The routes of HGT within the virus world can be most unexpected as spectacularly illustrated by the recent discovery of the homology of the surface glycoproteins of herpesviruses (large dsDNA viruses) and rhabdoviruses (negative-strand RNA viruses) 9697. In yet another dimension, the discovery of extensive movement of viruses between environments 2 attests to the spatial continuity of the virus community on earth. Hence the Virus World appears to be a spatial-temporal continuum that transcends the entire history of life on this planet.

The continuity of the Virus World notwithstanding, during eukaryogenesis, one of the most remarkable, turbulent transitions in life's evolution, it has gone through a second melting pot that gave rise to the realm of eukaryotic viruses. In this new domain of the Virus World, a few primeval lineages (primarily, viruses with small genomes) retained their identity, whereas most have been remolded into novel ones, albeit linked to the prokaryotic domain of the virus world through a variety of genes.

Despite the upheaval associated with eukaryogenesis, we submit that, in the Virus World, all viruses evolve from other viruses, even if via circuitous routes: Omnis virus e virus. Granted, this requires an important caveat: viruses do exist that seem to be disconnected form the rest of the virus world. Examples include several viruses of hyperthermophilic Crenarchaeota 11 and some of the insect polydnaviruses 98. In terms of the classes of virus genes delineated here (see Table 2), the genomes of these viruses consist, mainly, of class 1 genes (relatively recent acquisitions from the host) and class 3 genes (ORFans). These viruses, although they are exceptions rather than the rule, might be viewed as a challenge to the Virus World concept, and their genome composition seems to lend some support to the escaped-genes hypothesis. However, given the strong evidence of the continuity of the virus world provided by the results of comparative-genomic analyses of other viruses, we are inclined to believe that these agents evolved from other, "normal" viruses via step-by-step gene substitution and loss of original viral genes. The escaped-genes scenario for the origin of these viruses could be falsified by delineation of the path leading from "normal" to "unique" viruses via gradual loss of hallmark genes in related viral genomes. Hints at such a solution might be coming from the discovery of the JRC and the hallmark packaging ATPase in some of the crenarchaeal viruses 1154 and of the latter gene in one of the polydnavirus genomes (EVK, unpublished observations).

Having formulated the concept of the ancient Virus World, it makes sense to define the denizens of this world in more biologically relevant terms. The most meaningful definition appears to be that the Virus World consists of selfish genetic elements whose reproduction depends on certain basic capacities of cellular life forms, such as translation and energy transformation. This is an inclusive definition that accommodates not only traditional viruses but also a plethora of other selfish genetic elements, in particular, viroids that do not even have genes, and various kinds of mobile elements that integrate into the host genomes but propagate in the selfish mode. Once such elements lose their ability to reproduce autonomously, they emigrate from the virus world to the cellular world. Most often, having no function, they perish upon immigration but, on some occasions, the deteriorating selfish elements are functionally integrated into the host genome (especially, regulatory regions) which could be an important factor of the host's evolution 99100. Under this concept, the eukaryotic spliceosomal introns are a major incursion of the virus world into the cellular world. Indeed, the selfish nature of self-splicing group I and group II introns and their legitimate place in the virus world are evident, and the myriad spliceosomal introns of eukaryotes are believed to have evolved from group II introns, probably, coming from the mitochondrial endosymbiont, at the dawn of eukaryotic evolution 70101102. In complex eukaryotes, such as mammals, introns comprise ~30% of the genome; another ~60% of the genome consists of intergenic regions that are chock-full of mobile elements in various stages of decay. Thus, paradoxically, the bulk of our genome seems to be derived from the ancient Virus World.

A crucial aspect of the Virus World concept is its inextricable connection with certain classes of models for the evolution of cells. In particular, the results of viral comparative genomics appear to demand a complex, pre-cellular stage of life's evolution, characterized by extensive mixing and matching of genetic elements, and an archaeo-bacterial fusion model of eukaryogenesis. Under this view, the various replication and expression strategies employed by viruses (Fig. 1) emerge as relics of an ancient, "experimental" stage of evolution. Two distinct versions of one of these experimental designs, the invention of dsDNA replication, were particularly successful – so much so that they gave rise to the two cell types that survived throughout the rest of life's history. It is interesting to note that this concept of pre-cellular evolution reverberates with Haldane's early vision: "...life may have remained in the virus stage for many millions of years before a suitable assemblage of elementary units was brought together in the first cell" 47. The concept of the virus world and its connections with the evolution of cells has an important corollary: further mapping of the viral world will not only reveal the vagaries of virus evolution but should substantially inform the study of the origin of different types of cells.

W. Ford Doolittle, Dalhousie University

Fond as I am of many of Dr. Koonin's imaginative ideas, I'm not so keen on this one. My reasons are three-fold.

First, Koonin et al. acknowledge that HGT among and between viruses and hosts is one of the major forces in the evolution of their genomes. Their viral hallmark genes have had plenty of opportunity in the last three-four billion years to make their way into one or another host genome. That they have not done so must mean that the proteins they encode are of no use to hosts. And if they are of no use to hosts, then that in itself is sufficient explanation for their exclusive presence in viruses. It is irrelevant to the question of how long these genes have been associated with viruses.

Author response: Certainly, this is an ingenious and pertinent argument. However, matters are not quite so simple. Firstly, it is not absolutely true that cellular life forms have not recruited viral hallmark genes for their own purposes. A clear counterexample (it is discussed in the present paper but I will rehash the issue in brief) is the telomerase, a reverse transcriptase that apparently has been acquired by an ancestral eukaryote from a mobile retroelement and employed in an essential eukaryotic function, the replication of chromosomal ends. Let us note the pattern here: reverse transcriptase is a hallmark of the virus world that is common in both prokaryotic and eukaryotic selfish elements but performs an essential cellular function in eukaryotes only. It is hard to deny that this pattern is strongly suggestive of the origin of the gene in question within the virus world, with subsequent recruitment for the cellular function (telomerase) in eukaryotes. Granted, this is an exception. Most of the viral hallmark genes, indeed, are amiss from cellular life forms, which does seem to suggests that they are of no use to cells or might even be deleterious. In some cases, e.g., the JRC, this is understandable (cells do not need capsids), in others, the reasons remain mysterious, e.g., S3H. However, is exclusive presence of these genes in a broad variety of viruses (and this is how the hallmark genes are defined), indeed, irrelevant for the antiquity of their association with viruses? Certainly, the hallmark genes must have entered the virus world by one route or another, whenever that might have happened, time-wise. If the source of these genes was not the primordial gene pool, they must have come from genomes of cellular organisms. That means that these genes, at some time, have been of some use to cellular organisms. It seems extremely far fetched to suppose that, after being acquired by a virus, these genes all of a sudden became useless for cells. A different, probably, more viable version of a late, cellular origin of the hallmark genes might be considered (as, indeed, pointed out by Doolittle in the second part of his comment). This alternative involves acquisition of the progenitor of a hallmark gene from a cellular genome by a particular virus accompanied by a dramatic acceleration of evolution caused by the functional alteration. The gene, then, would sweep the viral world. The assumption here is that there is a set of functional constraints that is common to a variety of viruses. This seems to be plausible for some of the hallmark genes but not others, e.g., it does stand to reason that the capsids have common design even in very different viruses but, in the cases of S3H or the primase, a viral common denominator hardly can be gleaned, which does not bode well for the HGT scenario. We continue with this line of reasoning in our response to Doolittle's second point.

Second, and again, Koonin et al. acknowledge the role of HGT between viral "lineages" in the evolution of their genomes. The presence of genes in diverse viruses, though consistent with their presence in some LUCAV or primordial gene pool, is not proof of it. These genes could have arisen in one lineage and been transferred to others. By "arisen" I mean evolved so dramatically away from whatever viral or host function they used to perform that we can no longer detect the relationship. This headlong erasing-all-traces-of-the-past phase would have eased up before the born-again genes began their inter-viral odyssey, explaining the "conservation of the hallmark proteins in extremely diverse classes of viruses with widely different replication/expression strategies." Koonin et al. dismiss the HGT scenario because of "the extremely distant similarity between the hallmark proteins from diverse virus groups". But they are not so extremely distant that their homology is undetectable by sequence, and no one says the HGTs have to have been recent. Koonin would not, I think, argue that the extreme dissimilarity (homology not detectable by sequence) of many eukaryotic proteins to any in prokaryotes argues for their primordial pre-cellular origin.

Author response: This point is very closely related to the first one and their separation might be somewhat arbitrary. Nevertheless, we follow the structure of Doolittle's argument and respond separately. First of all, let us notice that, beyond any doubt, HGT is an extremely important aspect of viral evolution. Not only do not we deny the role of HGT but, on the contrary, we emphasize that HGT is the glue that holds the virus world together. Furthermore, it is likely that the currently observed distribution of the hallmark genes among viruses has been affected by HGT. However, we still maintain that HGT is not the preferred and, indeed, not a good explanation at all for the general pattern of that distribution. We attempted to present the arguments in this article but let us consolidate them here. 1. The distant similarity between the versions of the hallmark genes in diverse viral lineages suggests the ancient spread of these genes in the virus world. However, we must agree with Doolittle that ancient here does not necessarily equal primordial. 2. The essentiality of the functions of the hallmark genes is not well compatible with HGT being the principal mode of their dissemination over the virus world. Indeed, a question seems inevitable: HGT of a hallmark to what? Without the given hallmark gene, e.g., the JRC or the RdRp, there would be no competent recipient virus. Thus, the HGT scenario necessarily would involve displacement of ancestral genes with the same function, e. g., of a rod-shaped capsid protein gene by the JRC gene or of a gene for a DNA polymerase with the RCRE gene. Such displacements are not impossible in principle but some will turn out to be awkward when the pre-existing viral replication machinery is poorly suited to accommodate the newcomer. More importantly, what kind of virus world does this translate into? Seemingly, one with perpetual displacement of essential genes as a result of rampant HGT. Where would these essential genes come from in the first place? They would be there as long as viruses exist – suspiciously similar to the model of the virus world discussed in this paper. 3. The conservation of the hallmark genes between prokaryotic and eukaryotic viruses seems to be another telltale sign of primordial origin of these genes. We are unaware of any extensive gene movements between prokaryotic and eukaryotic viruses in modern times. However, at some point(s) in the past, such exchanges must have happened, and as discussed in this paper, the epoch of eukaryogenesis seems to be the most likely period for these events to occur (the second "melting pot" of virus evolution). Should that be the case, the spread of hallmark genes across the virus world is at least as old as the eukaryotes. However, that is a completely unrealistic upper bound because the hallmark genes must have been already contained in bacteriophage genomes in order to contribute to the emergence of the eukaryotic viruses. Taking the reasonable uniformitarian approach to bacterial evolution, we are justified to deduce that the prevalence of the hallmark genes in the virus world is as old as bacteria. From there, it is but a small – though not necessarily easy, given that uniformitarianism is hardly applicable here, – step back to the primordial gene pool.

The final point on this second comment of Doolittle is about eukaryotes and whether or not we would take the existence of eukaryotic proteins without detectable prokaryotic homologs (or with extremely distant homologs) as evidence of a primordial, pre-cellular emergence of eukaryotes. Surely, we won't although it is notable that others make this argument quite earnestly (perhaps, not exactly pre-cellular origin of eukaryotes but, definitely, a distinct, primordial eukaryotic lineage – see 60and references therein). The difference from the situation with viruses is clear and straightforward: it is demonstrable in many cases and seems highly plausible in others that emergence of novel eukaryotic functions entails major acceleration of the evolution of the genes that were inherited from prokaryotes but were exapted for these novel functions (the cytoskeleton and the ubiquitin system are obvious cases in point). Once emerged, the novel, eukaryote-specific cellular structures were rapidly fixed and then changed minimally throughout the evolution of eukaryotes. Thus the acceleration of evolution was dramatic but very brief, explaining the chasm between highly conserved, pan-eukaryotic proteins and their (sometimes, barely recognizable) prokaryotic progenitors. In our view, this is, by far, a simpler, better explanation of the observed pattern than any claim of a "new entity", let it be cellular or pre-cellular. The case of viruses is in a stark contrast: no pan-viral genes, no perceptible set of common functional constraints across diverse viral lineages (with some possible exceptions like JRC), hence no basis for rapid acceleration upon the entrance of a hallmark gene into the viral world followed by fixation in the new functional niche and the accompanying, equally dramatic deceleration of evolution. In a sense, this is the crux of our argument: viral hallmark genes are altogether a different lot from the sets of conserved pan-eukaryotic genes (or pan-archaeal, or pan-bacterial ones). Hence a qualitatively different evolutionary scenario is called for, and we try to step up to the plate in this paper.

Third, I totally agree with Koonin et al. that the habitual dismissal of an early viral origin "on the grounds that all extant viruses are intracellular parasites" is jejune by any standards, and that virus-like entities surely predated the appearance of modern cells. But today's viruses do not have to descend directly – in the sense that any of their genes descend directly – from these entities. Adam's sins are not my sins, even though I'm pretty sure the lineage of sinning is unbroken.

Author response: It is important that we agree on the antiquity of virus-like entities and (presumably) their importance in the evolution of life from the get go. We also do not disagree on the possibility that today's viruses have nothing to do with those of old. What we do disagree about is the conception that this possibility is as realistic as the alternative outlined in this paper, namely, that the major lineages of modern viruses derive directly from the primordial, pre-cellular gene pool. Not only is there no shred of evidence in support of the presumed sweep of new genes over the virus world but there are (we believe) substantial arguments against specific scenarios that must be developed to make such a sweep credible. These arguments are summarized in the paper and in our responses to Doolittle's first two points. Thus, we believe that the origin of the viral hallmark genes and several major lineages of viruses directly from the primordial gene pool is the simplest explanation of the patterns discovered by comparative genomics of viruses, and this scenario shows strong synergy with specific models of cell evolution. The epistemological status of this conclusion is briefly considered below.

All that said, I don't disfavor publication of this ms. Evolutionary scenarios are an artform. They usefully exercise the brain, causing us to look at old data in new ways and stimulating us to collect new data. They do not have to be true!

Author response: It might be wise to refrain from an explicit philosophical discussion and simply take this last statement of Doolittle as a legitimate opinion which it is. However, we strongly (inasmuch as the very notion of a "strong disagreement" is still relevant in post-postmodern philosophy) disagree with this agnostic stance (which we take as being serious rather than ironic) and think that this outlook does not help studies of early evolution. To be more explicit, we do not accept that "evolutionary scenarios are an artform" but rather contend that this is a distinct and important area of research within the general domain of historical sciences, such as evolutionary biology and cosmology. These scenarios do not have to be true, i. e., they do not have to be and never can be precise, proven accounts of the events that actually happened, but have to be earnest and defendable attempts on attaining an approximation of the truth that is, at least in some aspects, closer than previously available approximations. That is, we believe, the justification of research into such scenarios rather than the benefits of intellectual workout that accompanies these efforts.

Presumably, the notion that evolutionary scenarios are not a form of science stems, primarily, from the apparent lack of Popperian falsifiability for these concepts. There is, however, a lot to say about the status of such scenarios vis-à-vis the Popperian model of science and about the validity of that model and its applicability, especially, in the domain of historical sciences. First of all, the notion that evolutionary scenarios are unfalsifiable needs to be clarified. There are specific, falsifiable predictions in any evolutionary scenario worth its salt. To use an obvious example from the present work, Omnis virus e virus is an important part of our general concept of viral evolution, and it can be falsified by the discovery of a clear case of the origin of a virus from escaped genes. It is true that the scenarios are not falsifiable in their entirety, and neither is any histor