Annual Review of Virology The Social Life of Viruses

,


INTRODUCTION: WHY SOCIAL VIRUSES?
Consider a single cell coinfected by N identical viral particles, versus N cells infected by one viral particle each.If viral particles performed their infection cycles independently, these two scenarios should yield equally productive infections.Yet, in many cases, deviations from this expectation are observed, revealing virus-virus interactions.Two classical examples are provided by multiplicity reactivation and interference.Reactivation can take place when a noninfectious virus recovers infection capacity in cells containing another functional virus by means of genetic complementation (1,2).This requires a high multiplicity of infection (MOI), defined as the average number of viral genomes entering a cell.Yet, high MOI regimes also favor the emergence of defective interfering particles (DIPs), which function as molecular parasites of the normal virus (2,3).Whereas reactivation is viewed as a positive interaction in the sense that it increases viral fitness, DIPs usually have a strongly negative effect on the mean fitness of the viral population (2).
The fact that opposite outcomes have been reported under the same population regime stresses the need for a conceptual framework that helps us test alternative interpretations of a given observation, as well as to better understand and predict these outcomes.Other well-known virological processes can also be interpreted in opposite ways.For instance, superinfection exclusion takes place when a cell infected with a virus becomes refractory to secondary infections by similar viruses.Intuitively, superinfection exclusion can be viewed in terms of competition for cellular resources because it provides the resident virus exclusive access to the cell.However, exclusion has also been proposed to function as a cooperative process controlled by the resident virus, whereby free viral particles of the same virus are diverted to other cells with more available resources, increasing overall viral population fitness (4).In addition to classical processes such as reactivation, interference, and superinfection exclusion, recent research has revealed new types of viral interactions at the level of viral replication (5), innate immunity evasion (6), counteraction of bacterial defenses (7,8), and regulation of lysis by means of viral communication (9).In some cases, the mechanisms underlying these interactions have been partially elucidated, but their evolutionary basis and fitness implications remain poorly understood.
Social evolution theory provides an established conceptual framework that can be easily adapted to virus-virus interactions (10,11).From an evolutionary standpoint, a given trait is social when it modifies the fitness of the individual expressing the trait but also of other members of the population.This effect on other individuals should not be incidental, but instead should be central for explaining the existence of the trait itself.Social interactions occur in all types of organisms and, although they were first studied in animals, there is now an extensive literature demonstrating the social nature of microorganisms (12,13).Similar to what has been shown in other biological systems, interactions among viruses are mediated by processes such as group formation, the production of public goods, the exploitation of these goods by noncooperative variants, prudent exploitation of resources, and communication.Social evolution theory primarily addresses interactions established among members of the same species.Although interspecies interactions have also been addressed, these fall into a different category and are not discussed here.Hence, this review focuses on interactions involving identical viruses or variants of a given viral population but not on interspecies interactions such as pseudotyping, satellite viruses, or heterologous transactivation (14).
A general principle derived from social evolution theory is that, for cooperative traits to be maintained in populations, cooperators should preferentially interact with other cooperators.This argument has been formalized using different approaches, such as kin selection and group selection (15)(16)(17)(18)(19)(20).A central aspect of cooperative traits is that they benefit other members of the population, whereas selection acts primarily on the individuals expressing the trait (the actor).In some cases, the benefits provided to others come at the cost of reducing the fitness of the cooperator (altruism), as has been suggested for certain viral proteins involved in immunity evasion (6)(7)(8).This direct fitness cost needs to be compensated for selection to favor altruism.For this to happen, altruism should be preferentially directed toward individuals that are also altruistic, such that the overall fitness effect of the altruistic allele becomes positive.These indirect effects can be conceptualized using different definitions of fitness, such as inclusive fitness and neighbor-modulated fitness (20).For instance, inclusive fitness defines indirect effects as the benefits experienced by recipients of the actor's altruistic behavior, and weights them by the genetic relatedness between actors and recipients to calculate the overall fitness effect of the social trait.In mutually beneficial interactions, cooperators do obtain a direct fitness benefit.However, both this form of cooperation and altruism are susceptible to invasion by cheater variants.By reaping benefits of cooperation without reciprocating, cheaters gain a competitive advantage, jeopardizing the maintenance of the cooperative trait in the population.For cooperative interactions such as altruism to be evolutionarily stable, these have to be nonrandom, meaning that they should preferentially involve cooperators and thereby exclude cheaters.Applying these principles to viruses should help us understand, predict, and even manipulate processes such as immune evasion, virulence, lysis regulation, and the emergence of defective viruses, among others.
This article first reviews different forms of cooperation in viruses, which have been summarized in Figure 1.Next, it details how cooperation can be exploited by cheater viruses.Finally, it discusses processes that allow cooperative interactions to be preferentially established among Types of viral cooperation.(a) Collective dispersal.Groups of viral particles or genomes can benefit from being jointly transmitted by increasing environmental stability or resistance to host factors, as well as by promoting cooperative coinfection.Different types of collective infectious units have been described.A baculovirus occlusion body is shown.(b) Cooperative replication.Viral replication involves both template genomes and proteins encoded by these genomes, establishing a positive feedback between the number of replication templates and products.As a result, progeny genome production increases disproportionately with the initial number of templates.This benefit is felt in the short term, before cellular resources become limiting.(c) Cooperative immune evasion.Cells respond to infection by producing certain antiviral factors (yellow circles), but these responses can be overwhelmed in cells infected by multiple particles.This cooperative effect can be achieved in synchronously coinfected cells but also through successive infections, as shown for phage anti-CRISPR proteins.If the antiviral agent is secreted, cooperation can take place at the intercellular level as well, as shown for interferon suppression.(d) Prudent infection.By self-limiting host killing, viruses ensure the availability of resources over the long term.This effect has been demonstrated experimentally in phages and can be mediated by a viral communication system analogous to bacterial quorum sensing.Gray ovals indicate susceptible cells, dashed lines indicate killed cells, and red circles indicate latently integrated prophages.(e) Diversity-based cooperation.Viruses can establish synergistic interactions by sharing different variants of a given protein present in coinfected cells.Genetic complementation provides a classical example.However, this type of cooperation is particularly sensitive to the emergence of cheater viruses.
genetically related members of the viral population, excluding cheaters and making cooperation evolutionarily sustainable.

Collective Dispersal
Coinfection of cells with multiple viral particles is expected when the viral population density is comparable to or exceeds the density of host cells.This scenario is unlikely in certain stages of the viral infection cycle, such as during interhost transmission.However, coinfection can still occur if viruses disperse as groups, called collective infectious units (21).Examples of collective infectious units include virion aggregates, polyploid capsids, pools of infectious particles inside extracellular vesicles, and other structures such as baculovirus occlusion bodies.Virion aggregates have been observed in certain bodily fluids involved in interhost transmission, such as semen in human immunodeficiency virus type 1 (HIV-1) (22) and saliva in vesicular stomatitis virus (23,24).In turn, polyploidy has been described in filamentous viruses such as bacteriophage M13 and filoviruses, which can easily elongate their capsids to accommodate additional genome copies, but also in spherical enveloped viruses such as measles virus (21,25,26).Occlusion bodies, the interhost transmission structures used by baculoviruses, contain pools of virions enclosed in a proteinaceous matrix.Group transmission of multiple viral particles inside lipid vesicles has been demonstrated in several enteric viruses, including hepatitis A virus (27), enteroviruses (28), noroviruses, and rotaviruses (29).Recent research with the enterovirus coxsackievirus B3 has identified mutations that alter the release of lipid-associated infectious particles, such as substitution N63H in the VP3 capsid protein (30).Vesicle-mediated viral spread has also been shown in nonenteric viruses such as hepatitis C virus (31) and amoebal marseilleviruses (32).
Collective dispersal is a social trait because multiple viral genomes share a given transmission vehicle (capsid, vesicle, occlusion body).Synthesizing this vehicle can be costly for the virus, which might have favored group transmission to reduce per-capita production costs but which may also favor cheating.The fitness advantages of dispersing using these vehicles can be varied.For instance, they could afford increased stability in the extracellular milieu and promote interhost transmission, as is the case for baculovirus occlusion bodies.Transmission in vesicles might help viruses evade circulating antibodies or other antiviral agents such as secreted proteases (33).In addition, the advantage of collective dispersal may reside in promoting coinfection.Possible advantages of coinfection include cooperative replication, cooperative evasion of antiviral immune responses, and synergistic interactions among different virus variants, as detailed below.

Cooperative Replication
In principle, N particles coinfecting a cell should overall produce fewer progeny viruses than N particles infecting one cell each because the coinfecting particles compete for limited cellular resources.This is generally the case when the cellular infection cycle has been completed.However, as shown recently, coinfecting viruses tend to produce progeny faster because viral replication functions as a biochemically cooperative process (5).Cooperativity stems from the fact that virally encoded products are required for making viral progeny, including polymerases, replication cofactors, and other proteins, establishing a positive feedback between the number of replication templates and virus-encoded products.This should be a general feature of viruses but not of other replicons, such as plasmids, which do not encode their own replication machinery.By accelerating replication, coinfection can lead to earlier release of mature viral particles as long as the coinfecting viruses share their gene products.Moreover, initiating infection with multiple particles disproportionately reduces the chances that the infection is abortive.These expected Annu.Rev. Virol.2021.8:183-199.Downloaded from www.annualreviews.orgAccess provided by CSIC -Consejo Superior de Investigaciones Cientificas on 03/01/22.See copyright for approved use.benefits are supported by the analysis of mathematical models describing the infection cycles of well-characterized viruses such as HIV-1, influenza viruses, and hepatitis C virus (5,(34)(35)(36).
Experimental data in widely different systems have suggested the short-term fitness benefits of coinfecting cells with multiple particles.In one study, microfluidics was used for administering a controlled number of vaccinia viral particles to cells (37).Whereas most cells receiving single viral particles remained uninfected, infection probability increased disproportionately with the number of viral particles deposited per cell.In HIV-1, it was found that the infection cycle proceeded faster in situations promoting cellular coinfection, such as high local virion-to-cell ratios and cellto-cell viral spread (38).In the large amoebal marseilleviruses, pools of vesicle-encapsulated viral particles delivered to the same cell completed their infection cycles faster than individual particles (32).In some of these studies, though, viral progeny production was not calculated relative to the number of initial particles.This was addressed in another study in which collective infectious units constituted by aggregates of vesicular stomatitis particles were compared with the same number of particles in nonaggregated form.It was found that aggregated infection promoted faster release of viral progeny, which resulted in an accelerated infection progression in the short term (39).Recently, the cooperative replication model was tested by inoculating cells at different MOIs and measuring early mature viral particle release.The viral progeny production per initial particle increased with the MOI until a certain point in which competition for cellular resources reduced the per-capita yield.Studies in widely different viruses including influenza virus, vesicular stomatitis virus, respiratory syncytial virus, coxsackievirus, and human adenovirus have recapitulated these findings (5).

Cooperative Suppression of Host Immune Responses
Several studies have found that the relationship between the MOI and short-term viral progeny production rate is cell type dependent.In general, the benefits of coinfection are observed more clearly in less permissive cells.For instance, when comparing the yields of free HIV-1 particles versus pools of particles transmitted by the cell-to-cell route, no clear differences were found in highly permissive cells such as HEK293 or MT4 leukemia T cells, but the cell-to-cell spread mode was superior in primary CD4 cells (40).In vesicular stomatitis virus, infection with virion aggregates markedly increased viral fitness in cells displaying robust innate immunity responses, whereas the benefits of aggregation were lost in tumoral cells displaying poor or no innate immunity (39).These findings suggest that coinfection helps viruses overcome certain antiviral responses, for instance by allowing them to complete the infection cycle before these responses are deployed, or by overwhelming the cell with antiviral pathway antagonists.However, in most cases, the molecular details of these interactions have not been elucidated.
Certain innate immunity responses are deployed after the first infection cycle is completed, as shown by the dynamics of interferon secretion in infected cells for different viruses (41)(42)(43).The benefits of blocking these responses should thus not be felt by the individual virus exerting the blockade but by viruses infecting other cells in subsequent infection cycles.Interferon diffuses locally and signals entry of cells into an antiviral state (44).By preventing interferon release by a given cell, the virus creates a local environment favorable to viral spread.This should benefit the progeny of this specific virus if that progeny infects neighbor cells, but also other viruses present in the same neighborhood.This defines interferon suppression as a cooperative trait that can be modeled using classical kin selection theory (15, 45) (Figure 2).Specifically, let c be the direct fitness effect of interferon blockade in the infected cell (autocrine signal) and b the indirect benefit effect experienced by other members of the viral population resulting from the blockade of paracrine signaling.The direct effect c could be detrimental if suppressing interferon release is Assortment (spatial structure)

Complete
Intermediate Null Kin selection theory applied to viruses.Cooperation benefits other members of the population and hence is promoted by natural selection in an indirect way.If c is the direct fitness effect of a cooperative trait on the bearer of the trait, and b is the benefit afforded to others, Hamilton's rule establishes that the total fitness effect of the trait equals rb − c, where r is a relatedness term that measures the level of assortment between cooperators.In viruses and many other biological entities, r is strongly determined by spatial population structure.Two viral genotypes are shown in a grid of host cells.Red indicates cells infected with a virus that blocks innate immunity, and green indicates cells infected with a virus that does not block innate immunity.Cells infected with the green variant respond to infection by secreting signals (e.g., interferon) that allow other cells to become resistant to infection (blue cells).By blocking this signaling, the red virus creates an environment that is more favorable to infection.In strongly structured populations (left), the benefits of this blockade are experienced exclusively by the progeny of the red virus (r = 1).In contrast, in nonstructured populations (right), the progenies of the red and green viruses experience the same antiviral response (r = 0).If exerting this blockade has direct costs (c > 0), the red virus could be outcompeted by the green virus in nonstructured populations.The social nature of interferon suppression has been experimentally shown in vesicular stomatitis virus.costly, which would make interferon suppression an altruistic trait.As defined by Hamilton's rule, the condition for cooperation to be favored by natural selection is rb − c > 0, where r (relatedness) can vary between 0 and 1 and measures to what extent the benefits of cooperation (here, blocking interferon secretion) are preferentially experienced by other cooperators (higher r values indicate higher cooperator assortment).Therefore, despite the obvious benefits of suppressing innate immunity, natural selection may or may not favor this trait depending on the spatial structure of the viral population and of the interferon-mediated response, which could provide a rational explanation for seemingly disparate and counterintuitive findings between different viruses (46).
Cooperative antagonism of antiviral immunity has also been suggested in phages.The CRISPR-Cas system evolved as a mechanism for preventing phage infection (47).Immunity is acquired by the incorporation of small phage sequences in the form of spacers within CRISPR loci, which are transcribed as small RNAs that guide the digestion of the phage genome.In turn, phages have evolved anti-CRISPR proteins (Acrs) that block this form of immunity and that were first described in Pseudomonas but are also present in phages infecting Vibrio, Listeria, Neisseria, and other bacteria (48,49).Interestingly, as shown for the Pseudomonas aeruginosa phage DMS3m, a single phage particle often fails to produce a sufficiently high Acr activity, such that infection with multiple viral particles is required to successfully counteract CRISPR-Cas immunity (7,8).Interestingly, this form of cooperation does not require synchronous infection with multiple particles.Although a first incoming phage typically fails to block CRISPR-Cas and is degraded, this leaves the cell in a temporary immunosuppressed status, which can be exploited by a second phage particle to achieve a productive infection.Because the first phage increases the fitness of other members of the population at the cost of being degraded, this interaction may be viewed as a form of altruism.However, for this trait to be confirmed as altruistic, Acr production should per se entail a fitness cost such that, under conditions of low relatedness, nonproducers should be favored by selection.

Prudent Infection
A reduction in the virulence or infectivity of a virus tends to limit the chances of producing viral progeny in the short term, but by keeping the host alive for a longer period or allowing hosts to thrive, this also increases the chances that the viral population survives over the long term.As such, prudent resource exploitation obeys the definition of a cooperative trait.This rate versus yield trade-off underlies a large body of theoretical work aimed at explaining the evolution of virulence (50).The evolution of prudent infection has also been investigated experimentally (51,52).For instance, in one study, Escherichia coli phage ID11 was passaged under laboratory conditions of high genetic relatedness among interacting phages.This selected for an amino acid substitution in the major capsid protein that reduced adsorption efficiency without altering lysis time or burst size.Overall, this change increased the duration of the infection cycle, which afforded more time for cells to reproduce before being infected and allowed the phage to increase the total number of infections (53).
Recent research has revealed that, in certain temperate phages such as SPbeta Bacillus phages, prudent host exploitation can be mediated by a novel communication system called arbitrium (9).Similar to bacterial quorum sensing, the system is based on secretion of a small molecule that signals population density.Bacteria internalize this signal (AimP peptide), which is recognized by phage receptors and suppresses a negative lysogeny regulator.When a bacterial population contains a high density of prophages, the peptide concentration increases and lysogeny is derepressed, whereas at low prophage density lysis is allowed.Therefore, the arbitrium system allows phage to lyse cells and release progeny particles only when the expectation of finding susceptible hosts is high.Phylogenetic analysis suggests that the system is common among different types of Bacillus phages in addition to the SPbeta group (54).It was initially proposed that avoidance of cross signaling between different phage species has selected for rapidly diverging peptide sequences (9) and that specificity would also be promoted by changes in the peptide-receptor mechanism of interaction (55).However, structural analysis of the Aim receptor suggests that the system might be relatively flexible, allowing for a certain degree of cross signaling (56).A social evolutionary theory of this system that considers relevant molecular details is an exciting avenue of future research.

The Diversity-Based Cooperation Hypothesis
It has been suggested that high genetic diversity increases viral fitness by promoting synergistic interactions among different virus variants, particularly in fast-mutating RNA viruses (2,57,58).Genetic complementation between variants carrying otherwise deleterious mutations is the most straightforward mechanism underlying this type of cooperation.Genetic complementation has been demonstrated in segmented viruses such as influenza virus, in which noninfectious viral particles missing one or more segments regain infectivity in cells coinfected with other particles that provide these segments in trans (59).Subtler interactions between particles containing full segment sets have also been described in influenza virus.In one study, cells were coinfected with two isolates, one containing a hemagglutinin segment that improved cell attachment and another containing a neuraminidase variant that enhanced viral release from infected cells (60).It was found that viral fitness increased in coinfected cells as a result of protein sharing between the two isolates.
Complementation can be taken to the extreme in multipartite viruses, which encapsidate segments into different viral particles.Ensuring that at least one segment copy is present in every cell would require puzzlingly high MOIs (61,62).A solution to this problem is distributed infection, whereby complementation functions at the intercellular level.Even if most cells are not infected by a full segment set, trafficking of virally encoded products across cells can provide the functions required for completing the infection cycle, as shown for Faba bean necrotic stunt virus (63).This might explain why multipartite viruses are frequent in plants, where plasmodesmata allow for efficient intercellular channeling of proteins across cells, but rare in animals, which lack a similar intercellular trafficking system.
Positive interactions between different virus variants have also been suggested in nonsegmented viruses.For instance, in measles virus, coinfection with two virus variants led to the formation of cellular syncytia, despite the fact that neither variant alone had such fusogenic capacity (26).In this case, coinfection was promoted by the formation of diploid capsids carrying a genome copy of each variant.Generally speaking, in compact genomes such as those of RNA viruses, a potential advantage of coinfection is that genetic variants of a given multifunctional protein could become specialized in different subtasks such as, for instance, replication and transcription in viral RNA polymerases.Nevertheless, diversity-based cooperation has also been suggested in baculoviruses, which are slow-mutating, large DNA viruses.Baculovirus occlusion bodies often contain a mixture of different virus variants, including large deletion mutants (64).It was found that occlusion bodies containing such mutants were more infectious than genetically homogeneous occlusion bodies, suggesting synergistic interactions among these variants (65), although the underlying mechanisms remain uncharacterized.
However, whether natural selection has favored these positive interactions, for instance, by promoting coinfection mechanisms, remains to be shown.It should also be noted that interactions between different genetic variants of a virus need not increase viral fitness.A well-known counterexample is negative dominance, a process whereby a deleterious mutant reduces the fitness of the nonmutated virus in coinfected cells.Negative dominance might be the norm rather than the exception in oligomeric structures such as viral capsids, in which the deleterious variant mixes with the fully functional protein and produces a suboptimal structure.Negative dominance can even reduce viral adaptability, as demonstrated in poliovirus using a drug that blocks uncoating by binding to capsid proteins (66).Whereas drug-resistant mutants rapidly emerged, their expansion in the population was hindered at high MOIs by the presence of the wild-type drug-sensitive virus, which exerted a negative dominant effect over the resistant mutant in coinfected cells.A systematic measurement of the fitness of spontaneous mutants alone and in coinfection with other mutants would be needed to clarify whether interactions established in trans are predominantly synergistic or antagonistic.

Unrestricted Coinfection Promotes Cheating
As outlined above, cooperation ultimately requires that virus-virus interactions do not involve random members of the viral population.Otherwise, cheaters are expected to take over.Therefore, diversity-based cooperation faces a major conceptual challenge.For this type of cooperation to be evolutionarily stable, viruses should display mechanisms that allow synergistically interacting genetic variants to associate preferentially (a sort of recognition system).Alternatively, there should be mechanisms allowing virus-virus interactions to take place only episodically, such that there is no sufficient time for cheaters to accumulate.It should also be noted that, by buffering the effects of deleterious mutations, genetic complementation promotes mutation accumulation, potentially offsetting the initial benefits of cooperation and promoting error catastrophe (67,68).The emergence of cheaters when different virus variants are allowed to mix is best exemplified by DIPs, which have been reported in many types of viruses (69)(70)(71).DIPs typically evolve when the MOI is kept above one virus per cell for several generations.This scenario can take place when viral particles reach high population sizes but also when a virus spreads collectively even if the overall virus-to-cell ratio is small, as shown for baculovirus occlusion bodies (72) and vesicular stomatitis virus aggregates (73).
DIPs are deleted for genes encoding proteins that are sharable among viral genomes infecting the same cell.As such, these deleted genes define intracellular public goods in viruses.In general, if producing the public good is costly, cheaters that benefit from but do not produce the public good are favored by natural selection, ultimately leading to extinction of the public good (the tragedy of the commons).In principle, not all viral proteins need to be equally sharable among viruses coinfecting a cell.This depends on whether proteins diffuse relatively freely in the cytoplasm or remain associated to the genomes encoding them.In both RNA and DNA viruses, DIPs can lack many genes, showing that a large fraction of virally encoded products indeed function as intracellular public goods.Capsid proteins probably represent the most easily sharable good, as shown by pseudotype formation (74).DIPs are rapidly favored by selection because their much-shorter genomes are replicated faster.Additional fitness advantages result from the fact that DIPs do not have to trade off replication and transcription (75).DIPs can interfere with the normal infection cycle at different levels including entry, replication, packaging, and immune evasion (76,77).
Because DIPs are noninfectious in the absence of a functional virus coinfecting the same cell, in principle they should not drive the wild-type virus to deterministic extinction, although stochastic extinction can happen in finite populations.In contrast, deterministic wild-type extinction can be achieved by replication-competent cheaters.Game theory provides a useful framework for predicting the outcomes of competition at the molecular level (78) (Figure 3).For simplicity, let us consider pairwise interactions between viruses coinfecting a cell.As a first approximation, this scenario can be described by a two-by-two payoff matrix defining the fitness of a cooperator in cells coinfected with another cooperator (fA|A), a cooperator in cells coinfected with a cheater (fA|C), a cheater in cells coinfected with a cooperator (fC|A), and a cheater in cells coinfected with another cheater (fC|C).If fC|A > fA|A > fA|C > fC|C, natural selection produces an evolutionarily stable coexistence between the two variants, usually with fluctuating frequencies, because cheaters are fitter when most cells contain cooperators, but cooperators become fitter when most cells contain cheaters (frequency-dependent selection).DIPs are an extreme case of this configuration, in which fC|C = 0.In contrast, if fC|A > fA|A > fC|C > fA|C, selection can lead to the deterministic extinction of the cooperator because the cheater is still superior when most cells contain cheaters, despite the fact that mean population would be higher if all viruses were cooperators than if all viruses are cheaters (fA|A > fC|C).This paradoxical outcome in which selection does not lead to the highest-fitness outcome is known as the prisoner's dilemma and was originally demonstrated in the Pseudomonas bacteriophage Ø6 (79).

Cheating Beyond Coinfection
Cooperation can involve viruses infecting different cells, and so does cheating.In the context of innate immunity evasion, mutants that fail to suppress interferon production will be selected if suppression is costly for the virus in the infected cell (c > 0) and the benefits it affords (promoting infection of neighbor cells) are experienced equally by these mutants and the normal interferonsuppressive virus (r = 0) (Figure 2).This was suggested for a vesicular stomatitis mutant carrying a point deletion in residue 51 of the viral matrix protein M (6).This mutant failed to block .In well-mixed populations (no assortment, r ≈ 0), the prisoner's dilemma configuration leads to the deterministic extinction of A, whereas DIPs can prevail over but cannot drive the wild type to extinction.In spatially structured populations, cooperator assortment allows A to outcompete C. interferon secretion, yet was favored by selection under low-relatedness conditions, leading to a strong reduction in mean viral population fitness.
In the context of CRISPR-Cas immunity, mutants that do not produce Acr proteins have been described in P. aeruginosa phage DMS3vir.These mutants can exploit Acr proteins produced by other phages present in the same population by taking advantage of transiently immunosuppressed cells.However, they do not gain a net fitness advantage over Acr-positive phages and, hence, are not expected to take over the population (80).Interestingly, the interaction between Acr producers and nonproducers may lead to the evolution of Acr proteins of variable strength.On one hand, phages with strong Acrs have greater chances of suppressing CRISPR-Cas immunity, but, on the other hand, strong Acrs are more easily exploited by Acr-negative phages because they create more immunosuppressed cells.These trade-offs might explain why phages with no Acr, weak Acr, and strong Acr proteins are found in nature, despite the fact that, intuitively, one may expect selection to favor exclusively the evolution of the strongest possible Acrs.
Concerning prudent host exploitation, the long-term benefits of not exhausting the host population prematurely can be jeopardized by the emergence of rapacious viruses.These are cheaters because they benefit from rapid replication at the expense of exhausting host resources (they favor rate over yield) and can outcompete prudent viruses in the short term.The outcome depends on the level of mixing between the two virus variants.In populations exhibiting limited virus dispersal, interactions take place preferentially among similar viruses, meaning that the benefits of prudent exploitation tend to be shared by other prudent viruses (high relatedness).In contrast, in populations with high virus dispersal rates, prudent viruses will be outcompeted by rapacious counterparts even if this reduces long-term viral population fitness (52).Work with Pseudomonas fluorescens phage Ø2 showed that an additional condition for prudent phages to be selected is that hosts should not be a limiting resource (81).If bacterial density is low relative to phage density, rapacious mutants can rapidly evolve in prudent subpopulations.Therefore, the evolution of prudent exploitation depends on the strength of local competition relative to global competition for resources.

Spatially Structured Infections Promote Cheater Avoidance
As discussed, mechanisms of cheater avoidance are required for cooperation to be evolutionarily stable.Viral populations actually display features that increase genetic relatedness among interacting viruses (Figure 4).Most viruses undergo stringent population bottlenecks during interhost transmission (82)(83)(84).For instance, it has been estimated that sexual transmission of HIV-1 often involves a unique or very few founder sequences (85).This implies that the viral progeny produced within an individual host often derives from a common parental virus, hence ensuring high levels of genetic relatedness, at least during early infection.However, relatedness may decay as the infection progresses, being lower in chronic infections than in acute infections, particularly for fast-mutating RNA viruses.
In addition to the fact that each individual host constitutes a relatively well-isolated viral deme, spatial population structure can be found at the intra-host level.This has long been investigated in HIV-1, which often displays organ compartmentalization. HIV-1 subpopulations infecting different tissues are genetically distinct, particularly when comparing isolates from the central nervous system and blood (86).The eco-evolutionary implications of within-host spatial structure have been recently modeled in well-studied systems such as influenza virus (87).Most viral infections, particularly in solid tissues, progress in the form of delimited foci, either through direct cell-tocell viral spread or as a result of local virion diffusion (87)(88)(89)(90).These infection foci are typically founded by a single particle and thus initially constitute clonal subpopulations.Spatial clustering is further promoted in viruses that use cell-to-cell spread, such as many plant viruses, HIV-1, human T cell leukemia virus, measles virus, vaccinia virus, and herpes virus (91).Cell-to-cell spread can indeed be viewed as a form of collective spread mode in which the viral progeny from one cell is massively transferred to another cell.Because the density of viral particles can be high within foci, high MOIs can be achieved locally, allowing viruses to experience some of the benefits of coinfection, such as cooperative replication and counteraction of innate immune responses.DIPs may emerge within certain foci, reducing viral fitness locally, but these subpopulations should be outcompeted by DIP-free subpopulations, keeping global DIP frequency low.Therefore, spatial structuring at multiple levels promotes interactions among genetically related viruses, increasing the chances that cooperators remain segregated from cheaters.Cheater avoidance can also take place at the individual cell level by means of superinfection exclusion.Indeed, superinfection exclusion could be considered as cooperative in two different ways.First, it prevents other members of the population from infecting already occupied cells and diverting them toward free cells.This requires that exclusion takes place at the level of viral attachment (i.e., before entry), as shown for vaccinia virus (4).Second, it prevents DIPs and other cheater viruses from entering cells.Here, exclusion could operate at any level, including entry, replication, and transcription.
By promoting simultaneous coinfection, collective viral spread can favor the emergence of DIPs and other types of cheater viruses, as shown for vesicular stomatitis virus serially propagated in aggregates (73).However, collective infectious units do not necessarily support unrestricted mixing of viral particles.In several types of such units, such as polyploid capsids, virion-containing vesicles, and occlusion bodies, all viral genomes originate from the same cell and hence should exhibit relatively high levels of relatedness (33).Indeed, serial passage of coxsackievirus B3 vesicles for 20 transfers did not apparently lead to DIP emergence (92).Furthermore, it has recently been shown that the progenies of two coxsackievirus variants coinfecting the same cell do not mix freely because the pools of lipid-associated virions released from these cells tended to contain one variant or the other, but not both (92).
The above findings reveal that spatial segregation may even occur at the intracellular level in cells coinfected with multiple viruses.Most plus-strand RNA viruses replicate in membranous structures, known as viral factories or replication centers, in which certain viral components are entrapped (93)(94)(95).This could potentially limit the extent to which two coinfecting viruses share proteins.In contrast, other viral products tend to traffic across different compartments and hence be more easily used as public goods by the coinfecting viruses, particularly capsid and envelope proteins, as evidenced by pseudotyping.

SUMMARY POINTS
1. Cooperative interactions have been demonstrated in viruses at different levels, including transmission, replication, immunity evasion, and prudent exploitation of host resources.Viral cooperation can take place at the intra-and intercellular levels, and can involve host-or virus-encoded communication systems.
2. A general principle derived from social evolution theory is that, for natural selection to favor cooperation, social interactions should be nonrandom.Specifically, cooperators should preferentially interact with other cooperators.Otherwise, cheater genotypes are expected to invade populations.
3. Cheater viruses can evolve readily, as best exemplified by defective interfering particles, which typically take over viral populations under high multiplicities of infection.Other forms of cheating include hypervirulent viruses that undermine prudent host exploitation and mutants that give up immunity evasion.
4. Viral populations display features that promote cooperator assortment at multiple levels, such as transmission bottlenecks, organ compartmentalization, localized viral spread of infection foci, superinfection exclusion, and replication within discrete intracellular compartments.

5.
A special type of cooperation consisting of synergistic interactions between different viral genotypes has been postulated, but the conditions under which such cooperation is expected to occur also favor the emergence of cheater viruses, posing a major conceptual issue from an evolutionary standpoint.
6. Social evolution approaches including kin selection, group selection, and game theory should help us better comprehend, predict, and manipulate virus-virus interactions.

FUTURE ISSUES
1. Coinfection is a well-known driver of virus-virus interactions.Yet, recent research has revealed that cooperation and cheating can also be established among viruses infecting different cells, as shown by the arbitrium viral communication system controlling lysislysogeny decisions in phages.Whereas fundamental mechanistic aspects of this system have been elucidated, its evolution as a social trait remains to be explored.This offers an excellent opportunity for better understanding virus-virus interactions (cooperative or not), how these interactions depend on many factors (type of virus, transmission and spread mode, etc.), and the implications for pathogenesis.
4. The notion that cooperative interactions are established among different variants clashes with central principles derived from the social evolution field.The interpretation of experimental data supporting this tenet should be reevaluated in light of social evolution theory.It is possible that these interactions are not truly cooperative in the evolutionary sense but are instead by-product benefits of noncooperative traits.

DISCLOSURE STATEMENT
The author is not aware of any affiliations, memberships, funding, or financial holdings that might be perceived as affecting the objectivity of this review.

3
Figure 3 Game theory applied to viral cooperation and cheating.A cooperator (A) and a cheater (C) virus are considered.Examples of cooperative traits can include any sharable viral protein.Cheaters do not encode the cooperative trait but can benefit from it.By avoiding production costs, cheaters gain a competitive advantage.Evolutionary outcomes depend on the two-by-two payoff matrix defining the fitness of each virus alone (fA|A, fC|C) and in combination with the other virus (fA|C, fC|A).Two examples are shown: prisoner's dilemma (fC|A > fA|A > fC|C > fA|C) and defective interfering particles (DIPs) (fC|A > fA|A > fA|C > fC|C).In well-mixed populations (no assortment, r ≈ 0), the prisoner's dilemma configuration leads to the deterministic extinction of A, whereas DIPs can prevail over but cannot drive the wild type to extinction.In spatially structured populations, cooperator assortment allows A to outcompete C.

aFigure 4
Figure 4Multilevel spatial population structuring in viruses.Restrictions to viral migration among host populations, host-to-host transmission bottlenecks, limited viral trafficking across body compartments, localized viral spread (due to limited particle diffusion or to direct cell-to-cell spread), the discrete nature of cells, and superinfection exclusion, as well as intracellular viral replication centers, are all sources of spatial population structuring, which promotes cooperator assortment and hence limits the emergence of viral cheaters.

2 . 3 .
By favoring coinfection, collective viral spread should promote certain types of cooperation but may also set the stage for the emergence of cheater viruses.How collectively spreading viruses keep cheaters in check is an open research avenue.For instance, it has been shown that pools of viruses enclosed in extracellular vesicles tend to be siblings, which should help prevent cheater invasion.Spatial structure pervades viral infections from host populations to intracellular organelles.Recent advances have allowed tracking viral spread with unprecedented detail.