INTRODUCTION

During a full-blown virus infection, huge numbers of virions are often generated within the host. It is these huge numbers—in combination with high mutation rates—that generate the high levels of variability that facilitate the exploration of genotypic space, and thereby provide the raw materials that fuel the rapid adaptation often observed in virus populations in the clinic (1, 2), in the field (3), and in laboratory experiments (4). Whereas virus infections eventually lead to the production of large numbers of virions, the kinetics of incipient infections are more elusive (5); key events often cannot be observed directly, even with our contemporary arsenal of techniques (6). Although large numbers of virions are typically applied in challenge experiments, the notion that not all of them will effectively contribute to infection is intuitively appealing. Virions could fail to contribute to initiating infection because of (a) degradation by the host environment, (b) failure to bind to and enter host cells, due to stochastic spatial processes or physical barriers such as passive host defenses, (c) clearance by active host immune defenses, which also extends to initial infection sites, and (d) the fact that a considerable fraction of virions produced are noninfectious to begin with. But how many virions then effectively initiate the infection of a multicellular host? Classical studies on plant-virus local lesions—published over 75 years ago—first suggested the possibility of infections by a few infectious units, or even just one (7). Since then, new approaches for estimating the number of effectively infecting virions have been developed, and a range of new experimental techniques have become available. In many cases the number of virions that actually initiate infection appears to be very small (5, 8–19).

If virus infections are typically initiated by a small number of virions, the associated genetic bottlenecks could have profound effects on the development of disease in individual hosts as well as the evolution of virus populations over multiple rounds of host infection. A small number of infecting virions will increase the strength with which genetic drift acts, thereby increasing the strength of stochastic effects during evolution (20). Moreover, the number of infecting virions will also determine the extent to which mixed-genotype coinfections occur (5). The occurrence of mixed-genotype infections can limit the rate of pathogenesis-inducing interactions between virus genotypes (21, 22), while also limiting evolutionarily relevant interactions such as recombination between different virus genotypes (23). Although these two effects are in themselves straightforward, their evolutionary consequences may not always be intuitive.

In this review, we first consider the available approaches for estimating the size of population bottlenecks, which have links to models of virus infection kinetics. Next, we consider empirical estimates of virus population bottlenecks that occur during cellular infection, colonization of new host organs, horizontal transmission (nonparent to offspring), and vertical transmission (parent to offspring). Finally, we consider the potential effects of genetic bottlenecks on virus evolution suggested by theory and experimental observations. There is widespread evidence for the occurrence of genetic bottlenecks at different stages of infection. Although these bottlenecks will often have detrimental effects on virus populations, we also emphasize their potential benefits for viruses.

Genetic bottlenecks occur when the size of a population is strongly reduced, or when a small number of individuals from a large population colonize a new region (24) or, in the case of viruses, a new host or organ. The individuals that survive or colonize a new region found a new population, and hence are referred to as founders. We refer to the total number of founders as the size of the genetic bottleneck, to avoid confusion in talking about any specific founder. The founder effect occurs when the number of individuals that pass through the bottleneck is small enough that the frequency of genotypes in the new population of founders can be appreciably altered compared with that in the main population (Figure 1). From a conceptual perspective, a genetic bottleneck is a stochastic process: Individuals do or do not pass through the bottleneck at random, and regardless of their fitness or any phenotypic traits. Changes in genotype frequencies due to genetic bottlenecks—which are therefore entirely stochastic—are referred to as genetic drift. The number of individuals in a population at a given time point, determined by counting the number of actual individuals, is the census population size, N. The size of an idealized population that would behave—in terms of genetic drift—just as the population under study, is the effective population size, N_e. For populations that fluctuate in population size, N_e can be approximated as the size of the smallest bottleneck typically encountered (24).

In applying these concepts to viruses, we limit our discussion to infections of multicellular hosts initiated by a single, instantaneous exposure to a virus inoculum, or to the colonization of distal organs within the host by viruses spreading from the initial inoculation site. We do not consider host exposures to inocula at multiple time points: Whereas superinfections are biologically highly relevant, they are only beginning to be quantitatively studied and are a complex topic in their own right (25–27). Second, the number of virions that actually contribute genetically to infection is then the bottleneck size. We understand a genetic contribution to infection to be that a virus genotype has initiated a bona fide infection of a host and can be detected in the virus population found in that host, thereby acknowledging that sampling procedures and the sensitivity of detection techniques will undoubtedly affect estimates of bottleneck size. Third, many of the approaches used to study genetic bottlenecks require multiple, distinguishable virus “genotypes.” The ideal approach is to have two isogenic variants differing only in some neutral marker, such as a fluorescent protein or a noncoding sequence that can be specifically amplified. In practice, naturally occurring genotypes have been used in many studies, and even supposedly neutral markers are sometimes not entirely neutral. For some approaches the differences in the probability of infecting the host can be corrected for; the model then assumes a different probability of sampling each genotype, while the sampling itself is still random (5). Finally, in a virological context, genetic drift should also be carefully contrasted with antigenic drift; the latter is a phenomenological description of changes in the antigenic composition of virus populations, which appears to be mainly driven by selection for immune escape variants.

APPROACHES TO ESTIMATING BOTTLENECK SIZES

In attempting to estimate bottleneck sizes, the use of mathematical models is essential. For many model systems, it is difficult to directly observe early infection events without destructive sampling, necessitating the inference of bottleneck sizes from the virus populations present in systemically infected tissues. Moreover, even when early infection events can be observed quantitatively and tracked over time—for example, with plant viruses expressing marker proteins (16, 28)—whether and to what extent each site of primary infection contributes to systemic infection need to be inferred from the population present in systemically infected hosts (i.e., is it certain that each primary infection site actually contributed genetically to systemic infection?). Moreover, an important advantage of some of the mathematical models used to estimate the number of founders is that they take virion dose in the inoculum into account (5). In these cases, predictions of bottleneck size can be made over a range of inoculum doses, and inferences can be made about whether dose-dependent interactions (29) occur between virions.

Models of Infection Kinetics: Dose-Independent and Dose-Dependent Infection Probability Models

The simplest model of virus infection kinetics is the independent action hypothesis (IAH) (5, 30). This model assumes that each virion in the inoculum can be assigned a nonzero probability of infection that is independent of the inoculum dose, and that virions act independently of one another throughout the infection process. The number of effectively infecting virions, which is of course the genetic bottleneck size, is then simply the number of virions multiplied by the infection probability. If the model holds, the genetic bottleneck can be estimated for any dose. The model moreover makes other predictions, such as the overall rate of infection and the rate of mixed-genotype infections (5). In practice it is hard to measure the probability of infection per virion, and it is therefore usually estimated from the data. If a trait predicted by the model (e.g., genetic bottleneck size) can be measured at different doses, it can also be determined whether this simple model satisfactorily accounts for the data, or whether more complex models must be considered (29, 31–33). The IAH model is a gross simplification of the infection process, and for this reason, such a model may ring hollow for describing the intricacies of basic virus-host interactions. However, it must be remembered that—from a conceptual perspective—parsimony makes a model attractive, if indeed the model can account for observations. Although the IAH model clearly falls short in many instances, what is most surprising is the number of cases in which this model is supported by the data (5, 16). Moreover, when the IAH model is rejected, it is often due to simplifying assumptions made, and not to the principle of independent action itself. For example, for one case in which the simple IAH model with a fixed probability of infection was poorly supported by the data, allowing host susceptibility to vary over hosts resulted in better model support than alternative infection models without independent action (33). The IAH model is therefore a good starting point for understanding virus infection kinetics.

When the IAH model is not supported by the data, alternative infection models must be considered. First, the probability of infection per virion can be virion dose dependent (29). If it becomes harder for each virion to infect as the dose is increased, there is antagonistic dose-dependent action. If it becomes easier for each virion to infect as the dose increases, there is synergistic dose-dependent action. Second, more complex models that suppose the infection process consists of two or more stages could be considered (18, 33). For example, the virus can cause localized primary infections, and these primary infections interact with each other to establish systemic infection (34). These interactions—undoubtedly modulated by the host immune system—can also be synergistic or antagonistic. Finally, in some cases, completely different models have to be considered. For example, for some multipartite viruses, there is an extreme form of dependent action: Infection proceeds only when all required genome segments are present (35).

Mixed-Variant Infections

A common approach to estimating the number of infection founders has been to consider the frequency of mixed-variant infections (10, 11, 36). Hosts are challenged with a virus inoculum composed of one or more variants, of which the frequencies are usually known. In an ideal situation, these variants are genetically identical other than an easy-to-distinguish neutral marker, although this approach can still be used if there are small infectivity or fitness differences. However, any situation in which different genotypes are used is undesirable, because interactions between genotypes could mask the interactions between conspecific individuals that we are actually interested in when making first estimates of founder numbers. The host population is then screened for the presence of the virus genotypes, and the frequencies of single- and mixed-variant infection are determined. A model is then required to estimate from these frequency data the number of founders.

A range of approaches have been used to estimate the number of founders. Numerous mathematical models can be used to estimate the number of founders from the frequency of mixed-variant infections. The simplest mathematical model assumes a fixed number of founders, and then assumes the variants follow a binomial distribution (36). In other words, a coin is flipped a fixed number of times to determine how often a virus variant is found in infected hosts. A slightly more complex model assumes the total number of founders follows a Poisson distribution, and that the distribution of variants in turn follows a binomial distribution (5). (Alternatively, the distribution of founders for each variant follows a Poisson distribution.) In other words, a coin is flipped a variable number of times. Both of these approaches assume no interactions between virions or virus variants, but such interactions can be included in two-stage infection models if there are reasons to suspect them (18).

If there are no or extremely few mixed-variant infections, then the number of infection founders must be approximately 1, and hence no mathematical model is needed. This approach is used in classical bacteriological (37, 38) and virological (8) studies, and it is often linked to the idea of independent action. One must be careful with semantics here: Such a result shows that a single infectious unit founded the infection in the host, but it does not tell us much about infection kinetics and whether infectious units act independently. That is, strong antagonistic dependent action will always lead to single-variant infections, whereas in a low-dose range, synergistic dependent action can also lead to single-variant infections. On the other hand, if there are only mixed-variant infections, other approaches must be used to estimate the number of founders, because, for example, only the lower bound of a confidence interval for the founder number can be estimated using the frequency of mixed-variant infections.

If all hosts display mixed-variant infections, estimates of the number of founders can be made using two methods if (a) the frequency of variants can be determined in individual hosts and (b) there are no fitness differences between the virus variants. The first method makes use of Wright's fixation index (F_ST), which allows for the partitioning of genetic variance within and between populations (39, 40). The number of founders can then be estimated based on data from any two time points or tissues. An alternative method that tends to give similar results uses changes in the variance of the frequencies of variants to estimate the number of founders, because there is a simple relationship between the number of founders and the increase in variance due to genetic drift (41).

Estimating Genetic Bottleneck Size from the Number of Primary Infection Sites

The first estimates of the number of founders of virus infection were reported in the 1930s by Bald (7), who exploited the fact that—in some nonpermissive plant hosts—primary sites of virus infection result in a strong, highly localized necrotic host response: a local lesion. Bald counted the number of local lesions produced on leaves inoculated with different dilutions of a viral inoculum, a method that has since been widely used to titer plant virus preparations and to study infection kinetics (42, 43). For this approach, the number of founders of infection is simply the number of local lesions observed. Although this method is very straightforward and robust (44), it has two key disadvantages. Only plant viruses typically produce quantifiable local lesions and—given the effectiveness of this host response in preventing viral spread and systemic infection—local lesion numbers may have little or no bearing on what happens in a permissive host.

However, with currently available technologies such as the expression of marker proteins by the viruses, these limitations can be mitigated to some extent. Primary infection sites have been quantified by using plant viruses marked with fluorescent markers, allowing for studies in permissive hosts (16). Furthermore, this approach can also be applied in other hosts, such as Venezuelan equine encephalitis virus (VEEV) infection of midgut cells (15), although sampling is then destructive. Nevertheless, one shortcoming of this approach remains: It does not address to what extent each primary infection site actually contributes genetically to systemic infection. Such limitations could, in principle, be overcome by tracking infection from cell to cell, which is feasible to some extent in plants (45) and may very well also work in other hosts.

Dose Response

Although the dose-response relationship is usually not used to estimate the number of founders of infection, it is still highly pertinent to this discussion. First, dose-response data have been used to test the IAH model (29, 33), which if supported implies that the minimum number of founders is 1 at low doses. Druett (30) recognized that the IAH model with a fixed probability of infection leads to a fixed sigmoid-shaped response, which shifts its position only as the probability of infection is changed (Figure 1). This property has led to many tests of the IAH model over the years—in both virus and bacterial systems—by comparing observed infection levels or mortality in experiments to model predictions (5, 9, 14, 16, 29, 32, 33, 46, 47). This approach has gained considerable traction with experimentalists probably because of the straightforward experiments and wide applicability to any virus system in which the infection status of individual hosts can be determined. A range of approaches have been developed for comparing data to IAH predictions, and most tests rely on model selection (32): Does the IAH model adequately describe the data, or is a less parsimonious alternative model required? Second, founder estimates can be made directly from response data under IAH assumptions: If the zero term of a Poisson distribution describes the fraction of noninfected hosts, the corresponding mean is the average number of infecting virions (5). But how much weight can be given to these tests of IAH and the possibility of making founder estimates?

A considerable number of theoretical studies have highlighted the serious limitations and conundrums when analyzing dose-response data (9, 29, 31–33, 35, 44). IAH predicts a sigmoid-shaped dose response, but many models of infection kinetics can predict similar curves. For example, consider the combined effect of synergistic dependent action, which results in a steeper dose response, and variability in host susceptibility, which results in a more gradual dose response. The resulting response can be practically indistinguishable from IAH predictions (Figure 1). Moreover, dose response is also a very insensitive test to small deviations from IAH (44).

Concluding Remarks and Outlook on Approaches to Estimate Founder Numbers

Methods based on mixed infection of virus variants provide the most convincing and relevant estimates of the number of viral founders of infection. Whereas methods based on foci counting are direct and in principle robust, the implications of such estimates for founder numbers in full-blown systemic infections are not obvious. Dose-response-based methods are perhaps interesting as a “quick-and-dirty” method, but given the indirect nature of founder estimates and problems with identifiability of different dose-response models, their value is limited. Methods based on mixed infections are rather direct and robust when model assumptions are met, and by using marked isogenic virus variants, the possibility for interactions between these variants can be reduced practically to nil. Interactions between conspecific virions could still occur, and hence a modeling framework can be useful to detect them and consider their effects (18).

EMPIRICAL ESTIMATES OF VIRUS BOTTLENECK SIZES

Now that we have reviewed different approaches for estimating founder numbers, we consider empirical estimates of bottleneck sizes reported in the literature. We first consider bottlenecks that occur within the host at the cell and tissue levels, followed by bottlenecks that occur during horizontal transmission.

Cell-Level Within-Host Bottlenecks

Viruses are obligate intracellular parasites, and hence a bottleneck can occur during the infection of an individual cell. Bottleneck size in individual cells is usually referred to as the cellular multiplicity of infection (MOI). We give only a synopsis of the different results obtained in this field for the sake of completeness (Table 1), while noting that these studies have overcome a number of major technical hurdles (45, 48–51). Note that here we report the MOI in infected cells only, values of which can range from 1—because cells considered are by definition infected—to infinity (52). Bull et al. (50, 51) provided the first estimate of cellular MOI for a multicellular host: infection of insect larvae by the alphabaculovirus Autographa californica multiple nucleopolyhedrovirus (AcMNPV). An innovative approach was used that exploits the fact that multiple alphabaculovirus genotypes can be embedded in the same occlusion body (53)—the horizontal transmission vehicle—leading to an estimate of MOI in the final round of infection of 4.3 (50, 51). We also made an estimate of MOI for a second insect virus pathosystem, VEEV infection of mosquitoes, using data on primary infection of midgut cells reported by Smith et al. (15). Using Model 2 (52), we estimate MOI in this case to be very low, with a value of 1.094.

Table 1

Overview of cellular bottleneck size estimates

For plant viruses, a number of estimates of MOI have been made using different methods, including (a) detection of fluorescently labeled virus variants in protoplasts by microscopy (48) or flow cytometry (54), (b) observation of fluorescently labeled viruses in situ (27, 45), and (c) PCR-based detection of virus variants in protoplasts (49). This has led to low MOI estimates for tobacco mosaic virus (TMV) and tobacco etch virus (TEV), although both appear to increase over time (34, 48, 55). For citrus tristeza virus (CTV), very low MOI values were found during full-blown infection, suggesting the existence of a second mechanism of superinfection exclusion in CTV (27). Cauliflower mosaic virus (CaMV) MOI appears to increase and subsequently decrease over time, showing a wide range of values, ranging from 2 to 13 (49). The MOI for RNA2 of soil-borne wheat mosaic virus (SBWMV) is between 4 and 5 during the second and third round of cellular infection (45). Although MOI estimates are variable for these different systems, there are two trends in the data: (a) MOI is initially low, often near the minimum value of 1, and (b) MOI appears to increase following initial infection. The high initial MOI values for SBWMV may be due to its multipartite nature, as multipartite viruses are predicted to require high MOI values (56). Taken together, the MOI data also suggest that the high levels of infection may explain the high MOI for CaMV, as the rate of cellular infection is not as high in any other model system studied (Table 1). In general, we can therefore probably expect bottlenecks at the cellular level, especially during early infection.

Organ- and Systemic-Level Within-Host Bottlenecks

After infection has been established in a host organism, genetic bottlenecks can still occur during the colonization of new host tissues. Possible reasons for such bottlenecks include (a) the clearance of most virions by the host immune system, (b) few virions breaching a barrier or establishing infection in a particular tissue, or (c) few virions reaching certain host organs due to the host's physiology. An overview of quantitative estimates of within-host bottlenecks is given in Table 2.

Table 2

Overview of within-host bottleneck size estimates

Due to the planar anatomy of leaves and the ease with which they can be observed and cleanly removed from the host, plants have proved to be excellent model systems for studying within-host bottlenecks occurring during virus movement from one leaf to the next. A large number of studies have estimated the number of founders as viruses move from the inoculated leaf to other leaves. Hall et al. (10) pioneered this approach, simply using two natural strains of wheat streak mosaic virus (WSMV), which could be distinguished by restriction digests of RT-PCR products. It was estimated that there were 4 viral founders of infection during virus spread to a new tiller in wheat plants (36). Sacristán et al. (11) estimated genetic bottlenecks during TMV spread to systemically infected leaves—using different combinations of nearly isogenic strains—and also found a low number of founders. In contrast, studies have found much larger founder numbers during the colonization of leaves for CaMV (41) and pea seed-borne mosaic virus (PSbMV) (18). Other studies have shown that genetic bottlenecks occur during the colonization of systemic leaves, without actually making estimates of bottleneck sizes (12, 13, 57). In these cases, however, the number of genotypes found can give a rough estimate, because the experimental populations used were composed of a large number of genotypes.

What could explain differences in bottleneck sizes in these different studies? Of course different host-virus pathosystems are being used, but a couple of studies have proposed more general explanations. First of all, Gutiérrez et al. (58) have shown a correlation between the virion levels in the phloem—measured using an innovative approach employing aphids—and the size of the bottleneck in a leaf. Second, Tromas et al. (54) estimated TEV bottleneck sizes during spread to different leaves in tobacco and consistently found differences between differently positioned leaves. These differences could be explained by the host physiology; sink-source transitions, which determine where virions will be carried in the host plant; and host anatomy, that is, how the phloem is wired in the host plant and which sources are likely to contribute to which sinks. Hence, differences in the number of viruses being transported to leaves and their ability to be unloaded in these leaves may be important determinants of within-host bottleneck sizes in plant viruses.

For RNA animal viruses, many studies show good evidence for bottlenecks during within-host viral spread, but there are few quantitative estimates of founder numbers. Arbovirus infection of mosquitoes is thought to lead to numerous within-host bottlenecks, as shown by reductions in genetic diversity (59), although in one instance such bottlenecks were not observed for West Nile virus (WNV) (60). Forrester et al. (17) showed that a bottleneck exists between primary VEEV infection of midgut cells and subsequent infection in the hemocoel. Moreover, founder numbers depend on the inoculum dose, varying between 1 for low doses and 50 for high doses. For poliovirus, Kuss et al. (61) have shown the existence of multiple within-host bottlenecks in mice expressing the human poliovirus receptor. Interestingly, upon the downregulation of the host interferon response, within-host bottlenecks were no longer observed. For animal DNA viruses, the consistent high frequency of mixed infections of murine cytomegalovirus during spread from the peritoneum to the salivary glands of mice suggests there is not a strong bottleneck (62). For alphabaculovirus infection of insect larvae, however, the high variation between virus-variant frequencies suggests there may be further bottlenecks downstream of primary infection sites (33).

Finally, it is worth noting that bottlenecks can in principle occur during sustained infection of a particular tissue, if virus levels fluctuate over time. Such fluctuations could be due to the host immune system, but they might be induced by other environmental factors or—in the case of defective interfering viruses—the virus population itself (63, 64). If MOI is low and few cells are infected, we can expect some founder effects in such situations. Nevertheless, to the best of our knowledge, no estimates of such temporal bottlenecks have been made, although for hepatitis C virus infections one may very well occur (65). However, there is likely a large supply of potentially beneficial mutations during an established infection, due to the large N and high mutation rates. Coupled to the strong selection imposed by, for example, immune responses, there is likely a predominant role for selection under such a scenario. Although challenging to estimate from a computational perspective, bottlenecks during sustained infection may be of importance to understanding within-host infection dynamics.

Estimates of Bottleneck Size During Horizontal Transmission

Next we consider empirical estimates of bottleneck size during primary infection only or for complete host-to-host transmission, which have been determined for many systems (Table 3). For plant viruses, Hall et al. (10) showed that a bottleneck occurs during aphid transmission of WSMV between wheat plants, and based on the combined data reported, we use Model 2 (52) to estimate the number of founders to be 1.30. Moury et al. (19) estimated founder numbers upon aphid-borne transmission of potato virus Y (PVY) between pepper plants—using mixtures of infectious and noninfectious virus variants—and estimated a narrow bottleneck: 0.5 to 3 virions per insect. Betancourt et al. (66) then made an estimate for aphid-borne transmission of the multipartite cucumber mosaic virus (CMV) between tomato plants and also estimated 1–2 founders per transmission event. Evidence for narrow bottlenecks has been found in other studies on transmission by different aphid species to different plant hosts, although estimates of bottleneck size were not made. Sacristán et al. (67) showed there was a narrow bottleneck upon transmission by brushing of leaves, without estimating bottleneck size. We recently estimated founder size upon mechanical inoculation with different virion doses and found dose-dependent founder numbers—in accordance with IAH predictions—between 1 and 50 virions (16). Taken together, these results show that for plant viruses there can be narrow bottlenecks during horizontal transmission, although at least for mechanical transmission this depends on virion dose.

Table 3

Overview of between-host transmission bottlenecks

For animal viruses, estimates of bottleneck size during horizontal transmission have also been made. Smith et al. (15) showed that even when the number of primary VEEV infection sites approaches 1, a considerable proportion of mosquitoes still become systemically infected, a finding further detailed in follow-up work by the same group. These results show that founder number for systemic infection can approach 1, and furthermore that it is also dose dependent (17). Turning to animal DNA viruses, there are a number of studies on horizontal-transmission founder numbers for baculoviruses. Using natural virus isolates administered at low doses, Smith & Crook (8) argued that for both alpha- and betabaculoviruses, narrow bottlenecks approaching 1 could occur. This idea was revisited for baculoviruses later on, with an experimental test of IAH (5). No estimates of founder numbers were reported in the study, but they can be easily made using Model 2 (52). For example, for AcMNPV infection of Spodoptera exigua third-instar larvae, this procedure renders bottleneck size estimates ranging from 1.3 at low doses to 5.3 at higher doses. In other hosts, the frequency of mixed-variant infections was much higher than expected under IAH (5), which a follow-up study showed is probably due to variation in host susceptibility (33).

What is striking when all the estimates for bottleneck sizes for horizontal transmission are considered together is the relationship with infection level (Figure 2). This relationship appears to be roughly similar to IAH predictions, suggesting that infection levels can be used to make a rough first estimate of bottleneck size. However, cases in which the simple IAH model is not supported (33) can for some doses clearly lead to predictions that clash with the model, urging caution, especially if highly variable real-world host populations are to be considered.

Estimates of Bottleneck Size During Vertical Transmission

Many viruses are transmitted vertically, and as such, it would be valuable to better understand the evolutionary dynamics of vertical transmission. This area has, however, long been neglected and has only recently received more attention. Three experimental evolution studies have confirmed that vertical transmission can lead to reductions in virulence and fitness, as compared with horizontal transmission (68–70). In one study, the evolution of increased rates of horizontal transmission was paired with reductions in virus titer (69), which raises the question of how low-accumulation variants can outcompete those with higher accumulation. Genetic bottlenecks could clearly play an important role, providing for the segregation of virus variants. Moreover, clinical studies on human immunodeficiency virus type 1 strongly suggest that there is a narrow bottleneck during transmission in utero or during parturition (71).

An experimental estimate of bottleneck size during vertical transmission is therefore most welcome and was recently provided—for the first time—by Fabré et al. (18). The authors considered the seed-borne transmission by peas of PSbMV and estimated a low founder number of 0.84. Interestingly, the authors' analysis strongly suggests that the occurrence of infections in seedlings is positively density dependent, whereas the different virus variants might inhibit each other. A narrow transmission bottleneck in an experimental system in which vertical transmission is readily achieved suggests such bottlenecks might also occur in other systems in which vertical transmission is scarcer. Further studies will hopefully shed light on the generality of this phenomenon.

IMPACT OF BOTTLENECKS ON VIRUS EVOLUTION

We have reviewed how experimental studies show that narrow genetic bottlenecks are common during many stages in the life history of a virus: between-host transmission, initial infection, spread within the host, and cellular infection. Such bottlenecking events are generally perceived to be disadvantageous for evolving populations, as they lead to random changes in genotype frequencies—counteracting selection—and they can reduce advantageous genetic variation in a virus population. Indeed, a range of studies in cell culture and in host organisms have shown that continuously bottlenecking a virus population unleashes Muller's ratchet (72–77): In each bottleneck new mutations are fixed that are likely to be deleterious, and reductions in fitness may in turn lead to smaller bottleneck sizes, thus driving the population into a mutational meltdown (78). Moreover, a number of studies have shown that genetic variation in virus populations may be advantageous for a virus: Different genotypes have propensities to invade different host tissues (79, 80). Bottlenecks could remove these variants from the population, although it should be mentioned that in these cases error-prone viral polymerases would normally rapidly generate the required variants. Moreover, beneficial de novo variation will also need to reach a given frequency in a virus population before it is likely to carry over a bottleneck, meaning that narrow bottlenecks, or frequent wider bottlenecks, can hamper adaptation (81). Understanding how viruses manage to thrive despite these bottlenecks is therefore highly interesting and relevant.

However, we feel another perspective on virus population bottlenecks could be given more consideration: They can also have important advantages for viruses (Figure 3). First, population bottlenecks can very effectively remove cheaters—such as defective interfering viruses—from virus populations (82). Widespread occurrence of bottlenecks is probably one important reason why defective interfering viruses are not common in natural pathosystems. Second, Miyashita & Kishino (45) generalized this point by convincingly arguing that population bottlenecks can make selection more effective, if beneficial alleles also act in trans. For example, a mutation might improve the activity of an inhibitor of apoptosis, while a high cellular MOI confers the benefit to all viruses in cells infected by a variant carrying the mutation. Selection will be stronger when MOI is low, and the difference in apoptosis inhibition between virus genotypes becomes apparent. This argument works at the cellular level, but it might also be applicable at the organ or host level.

Another possible advantage of genetic bottlenecks could be their effects on evolutionary trajectories in rugged fitness landscapes (83). A number of studies have considered what the interactions are between different mutations in the viral genome (i.e., epistasis). These studies have typically found strong epistatic interactions between mutations—indicative of a rugged fitness landscape—and in particular many cases of positive-magnitude epistasis (84). In some cases reciprocal-sign epistasis has been observed (85, 86): The fitness of two strains with individual mutations (Ab or aB) is lower or higher than the fitness of wild-type (ab) or double-mutant (AB) strains. Reciprocal-sign epistasis is characteristic and necessary for rugged fitness landscapes with multiple fitness peaks (87). On such a landscape, it becomes readily possible for viruses to become trapped on suboptimal fitness peaks (88). In some cases relatively low and smooth peaks can be more favorable than high but steep peaks (the “survival of the flattest” effect) (89). Given the typical distribution of mutational fitness effects for a virus and of epistatic interactions, the fixation of mutations by drift will on average lower fitness. However, the fixation of these mutations may be necessary to traverse a valley in the fitness landscape, depending on the exact topography of the fitness landscape, mutation rates, and bottleneck sizes. Whereas a large N between bottleneck events—during infection of an organ or host—leads to the effective reconnoitering of the local genotypic space and strong selection, intermittent genetic bottlenecks between infections can benefit populations by allowing for the exploration of rugged genotypic spaces with multiple fitness peaks. Narrow genetic bottlenecks during the virus infection cycle are probably caused by limitations imposed by the host environment. But in some cases these bottlenecks could have some advantages, and there may not be strong selection—on the between-cell, between-organ, or between-host level—to eliminate them.

disclosure statement

The authors are not aware of any affiliations, memberships, funding, or financial holdings that might be perceived as affecting the objectivity of this review.