Intraclonal variation in RNA viruses: generation, maintenance and consequences

Abstract

This paper explores the evolutionary implications of the enormous variability that characterizes populations of RNA viruses and retroviruses. It begins by examining the magnitude of genetic variation in both natural and experimental populations. In natural populations, differences arise even within individual infected patients, with the per-site nucleotide diversity at this level ranging from < 1% to 6%. In laboratory populations, two viruses sampled from the same clone differed by ∼0.7% in their fitness. Three different mechanisms that may be important in maintaining viral genetic variability were tested: (1) Fisher's fundamental theorem, to compare the observed rate of fitness change with the extent of fitness-related variation within the same experimental populations; (2) magnitude of genomic mutation rate, to assess whether it correlated with fitness-related variation, as predicted by the mutation-selection balance hypothesis; (3) frequency-dependent selection, which affords rare genotypes an advantage. The paper concludes with a discussion of two evolutionary consequences of variability: the fixation of deleterious mutations by drift in small populations and the role of clonal interference in large ones.

fitness, frequency-dependent selection, genetic variation, HIV, mutation rate, nucleotide diversity, retroviruses

INTRODUCTION

RNA viruses are ubiquitous cellular parasites found in nearly all life forms. Major reasons for their success stem from the very high mutation frequencies of their polymerases and from the enormous fluctuations in their population sizes. The error frequencies for several viral RNA polymerases have been estimated by a variety of methods. While estimates vary, most average around 10⁻⁴−10⁻⁵ base substitutions per single base site (Drake & Holland, 1999). The mutation frequencies of RNA viruses exceed by more than a million-fold the average nucleotide substitution frequencies of the DNA replication systems of their eukaryotic hosts (Drake et al., 1998), probably mainly because DNA replication systems contain proofreading mechanisms and mismatch repair systems which most viral RNA polymerases appear to lack (Steinhauer, Domingo & Holland, 1992). Combining these estimates of per-site substitution rate with genome length, it emerges that RNA viruses have a total genomic mutation rate per replication cycle of around 1 (Drake & Holland, 1999). A side-effect of this tremendous variability is the limited efficacy of attempts to control outbreaks. In fact,> 50 new human diseases caused by RNA viruses have emerged in the past two decades, clearly demonstrating how readily some RNA viruses jump from their original hosts to humans (Morse, 1993).

Another feature of RNA viruses is their relatively small genome compared to DNA viruses. A small genome has both limitations and advantages. One limitation is that most mutations will show deleterious fitness effects. Recent estimates of the deleterious mutation rate for fitness for vesicular stomatitis virus (VSV) (Elena & Moya, 1999) showed that most of the mutations produced had deleterious effects on viral replication. (Supporting this notion is the high virion/plaque-forming unit ratio of most riboviruses.) On the other hand, one advantage is that the smaller the genome length, the higher the probability of copying it without errors (Eigen & Biebricher, 1988).

A second important source of genetic diversity is recombination. Although exchange of genome fragments is universal in DNA viruses, it is not in RNA viruses (for a review, see Chao, 1994). Despite intensive searches, recombinants have never been identified in RNA bacteriophages and several eukaryotic RNA viruses (e.g. VSV and Newcastle disease virus of fowl). In many other RNA viruses (e.g. the families Picornaviridae and Coronaviridae), recombination is common. The mechanism involved is believed to be template-switching by replicases during synthesis of a new strand. Lastly, in other RNA viruses (e.g. Reoviridae, Birnaviridae, Orthomyxoviridae, Bunyaviridae, Arenaviridae, Nodaviridae, and Cleviridae) exchange of genomic regions can be achieved by an alternative mechanism: genome segmentation that leads to re-assortment during coinfections.

A central problem in evolutionary biology concerns the nature of the forces that maintain genetic variation among individuals within populations. Three principal processes have been proposed: selection, mutation and drift. Although drift clearly plays a role in maintaining molecular variation and mutation is essential for the origin of variation itself, selection is the only force capable of systematically maintaining variation in fitness-related traits. The field of population genetics has historically paid far more attention to organisms that are diploid, sexual and, of course, with DNA-based genomes, than to those that are haploid, asexual and have RNA genomes. Nonetheless, it has become clear in recent years that viral species harbour tremendous genetic diversity (reviewed in Domingo & Holland, 1997; Elena et al., 2000), although the forces that affect genetic variation may be quite different depending on the organism's genetic system. The most obvious difference between haploids and diploids is that dominance cannot play any role in maintaining genetic variability in haploid populations. A more subtle difference arises from the effect of beneficial mutations sweeping through a population by natural selection. In an asexual organism, all loci are effectively linked so that a selective sweep purges variation throughout the genome (Muller, 1932; Levin, 1981; Gerrish & Lenski, 1998). Therefore, it is especially interesting to examine the population genetic mechanisms that maintain variation in natural and experimental populations of RNA viruses.

In this paper, we present evidence concerning the enormous intraclonal genetic variation characteristic of RNA viruses in both natural and laboratory populations. We examine three fundamental population genetic mechanisms that could theoretically maintain this variation (Elena & Lenski, 1997): (1) it may be transient and reflect the ongoing substitution of beneficial alleles by natural selection; (2) it may be maintained by recurrent deleterious mutations that give rise to a mutation-selection balance; (3) it may be maintained by negative frequency-dependent selection, whereby genotypes are fitter when they are rare relative to when they are common. We also briefly explore two consequences of intraclonal variation in viral populations: (1) the declines in fitness associated with random transmission events of a few particles and the onset of Muller's ratchet (Duarte et al., 1992); and (2), the role played by clonal interference in large populations (Miralles, Moya & Elena, 1999).

GENETIC VARIATION IN NATURAL POPULATIONS

The virological literature contains an enormous number of studies where multiple genotypes are sampled from different organs of the same patient, enabling estimation of the structure of viral populations. For the sake of illustration, we estimated the average per-site nucleotide diversity for several regions of the env (C2-V5), gag (p17), and pol genes of HIV-1 from a suitable set of such studies (Ball et al., 1994; Poss et al., 1995; Zhu et al., 1996; Hughes, Bell & Simmonds, 1997; Wong et al., 1997; Delwart et al., 1998; van't Wout et al., 1998). In order to have a homogeneous sample, we only used data from sero-converted patients who did not receive any drug treatment and only samples taken from the same organs. A total of 917 sequences were compiled for 27 patients and 20 organs or tissues. Across organs, nucleotide diversity (SEM) ranged from 0.0095 ± 0.0028 (t₂ = 3.4139, P = 0.0761) for viruses isolated from blood plasma to 0.0639 ± 0.0074 (t₂ = 8.6060, P= 0.0132) for viruses isolated from the spleen, with an average value of 0.0391 ± 0.0224. These numbers reflect the enormous within-clone genetic diversity that characterizes viral infections and the adaptive radiation experienced by HIV-1 when facing different local conditions imposed by different cellular environments. Table 1 shows a three-level nested analysis of variance that demonstrates how nucleotide diversity differs among organs (P < 0.0001). A simple (although not necessarily the only) explanation for this difference is that, after colonization of a new organ by only a few viral particles, genetic variability is generated de novo. The fate of new variants is determined by their ability to survive in this organ. Therefore, the diversity of the viral population will strongly depend on the selective restrictions imposed by each cell type.

Table 1.

Three-levels nested model II ANOVA testing for differences in the per-site nucleotide diversity of HIV-1. For the scope of our study, the relevant comparison is for organs within patients

Source of variationa	d.f.	SS	MS	F	P
Studies	6	0.3222	0.0537	2.3532b	0.0382
Patients within studies	20	0.4668	0.0233	1.0309	0.4427
Organs within patients	60	1.3586	0.0226	13.5021	<0.0001
Error	830	1.3922	0.0017
Total	916	3.5397

a

All factors were considered as random.

b

MS_studies tested over the pooled sum of squares.

GENETIC VARIATION IN EXPERIMENTAL POPULATIONS

The amount of sequence data from experimental populations of RNA viruses is colossal (for a review, see Domingo & Holland, 1997). However, a useful feature of laboratory populations is that they allow, under certain circumstances, estimates of a special type of variability: fitness variability. From an evolutionary standpoint, only those mutations affecting fitness are of interest. Duarte et al. (1994) directly estimated the genetic variance for fitness for a single clone of VSV. To do so, they isolated a well-defined plaque from an infected monolayer of susceptible host cells, diluted it into fresh media to separate all the viral particles, and plated them out onto a new susceptible monolayer. They then picked 100 new plaques and measured the fitness of each one by head-to-head competition against a reference strain (congenic with the original virus but with a visible neutral phenotypic marker). Each of these plaques represented independently sampled components of the original plaque. If all these genotypes were identical, as expected for clonal organisms, we would expect a non-significant estimate for the genetic component of variance for fitness. By contrast, if the high error rate of VSV RNA polymerase incorporates an average of 1.15 mistakes per genome and replication round (Drake & Holland, 1999), and if most of these mutations affect fitness, the expectation is radically different: we should be able to detect a significant among-genotype variance component for fitness. Indeed, the latter was observed; using the ANOVA method to partition the total variance into pure error and among-genotype differences (additive genetic variance, σ²_A) (Elena & Lenski, 1997), we have been able to estimate a significant σ²_A = 4.546 × 10⁻⁵ (95% confidence interval, 3.367 × 10⁻⁵ ≤ σ²_A ≤ 6.376 × 10⁻⁵). This additive variance represents 79.01% of the total observed variance. Taking the square root of this variance component yields a corresponding standard deviation for fitness of 6.742 × 10⁻³. In other words, two viruses chosen at random from the same VSV ‘clone’ differed from one another in their relative fitness by ∼0.7%, on average.

Hence, we conclude that genetic variability within clones is readily generated during error-prone replication of RNA genomes and is maintained in the populations. There are now two questions of interest: (1) How is this genetic variation maintained? (2) How does this enormous variability constrain the evolutionary fate of RNA viruses? We will focus on these two questions in the following sections.

MECHANISMS THAT PROMOTE THE MAINTENANCE OF GENETIC VARIATION

FISHER'S FUNDAMENTAL THEOREM

According to Fisher's fundamental theorem, the rate of change in the mean fitness of a population is equal to the genetic variance for fitness within that population (Fisher, 1930). This theorem is particularly concerned with the selective sweep of beneficial alleles through a population, which simultaneously generates transient genetic heterogeneity and a shift in the population's mean fitness. We sought to determine whether the observed variation in fitness within experimental populations of VSV was simply a manifestation of the ongoing substitution of beneficial alleles carried by different clones. To do so, we looked for the existence of positive correlations between genetic variability and the magnitude of fitness improvements.

To this end, we took advantage of the existence of a considerable body of work that has been done to explore the effect of the initial population size and diversity on the rate and extent of VSV adaptation to in vitro conditions (Novella et al., 1995a, b, 1996). This work demonstrates that the bottleneck size and genetic diversity required to stop Muller's ratchet are strongly dependent on the mean fitness of the initial population. We can use all of the existing data to test the above hypothesis about Fisher's fundamental theorem. To do so, we have to estimate the average per day rate of fitness change () and the clonal diversity of the starting population. The first figure comes directly from Novella et al. (1995a,b, 1996). The second figure is a little more problematic, since Novella et al. did not estimate the genetic variability of their starting populations. Instead, they used a fixed number of viruses to found their populations. Given a genomic mutation rate for VSV of 1.15 (Drake & Holland, 1999), it is not unrealistic to assume that every virus present in the initial set would have, on average, one mutation. Hence, we can assume that the initial samples contained as many different genotypes as viral particles. Of course, this may be an overestimate of the true number, but it at least gives us an idea of the relationship between fitness change and variability. Figure 1 shows the relationship between and Shannon's entropy, a measure of population diversity, for four different experiments carried out with different mAb-resistant genotypes of VSV. A partial correlation coefficient indicates the existence of a significant positive correlation between these two variables (r = 0.9566, 9 d.f., 1-tailed P < 0.0001), as predicted by Fisher's theorem. While it is possible to argue that this correlation was mostly driven by the largest associated with the highest diversity, it should be noted that removal of this data point did not change the conclusion (r = 0.8093, 6 d.f., 1-tailed P = 0.0075).

Figure 1.

Experimental demonstration of the role plaid by Fisher's fundamental theorem in the maintenance of clonal diversity. The per generation change in fitness for several MARM genotypes of VSV, , was estimated for different initial diversities. The population diversity has been measured as Shannon's entropy (, where p_i is the frequency of the ith mutant in the initial population. In our particular case, , where N is the number of viral particles in the initial inoculum). A significant positive correlation gives support to the predictions made by Fisher's fundamental theorem.

THE MUTATION-SELECTION BALANCE AND THE QUASISPECIES CONCEPT

A polymorphism of deleterious mutations can be steadily maintained by a balance between newly created mutations and the purging effect of selection. As far as we know, no experiments done with RNA viruses directly test whether this mechanism works in viral populations. However, we can explore this possibility by looking at the level of non-synonymous variability in certain fitness-related genes in order to establish a positive correlation between this type of variability and total genomic mutation rate. The rationale behind this hypothesis is as follows. Different RNA viruses and retroviruses have total genomic mutation rates that vary by as much as two orders of magnitude (Drake, 1993; Drake & Holland, 1999). For instance, the Moloney murine leukaemia retrovirus (MoMuLV) genomic mutation rate was estimated to be as low as 0.003. At the other extreme, the genomic mutation rate for the bacteriophage Qβ was approximately 6.5. Hence, we should expect that the degree of population heterogeneity for the first virus at fitness-related loci would be far less important than for the second, simply because it is producing fewer mutant genotypes. A second difficulty involves deciding which genes are of special relevance for viral fitness. Obviously, in a compacted genome, all units are important. However, a priori, we can argue that RNA polymerase and coat proteins should be of extreme importance. The former because it is the responsible for the replication of the RNA genome; the latter because they are generally involved in viral transmission, stability outside the cell, interaction with cellular receptors, and entrance into host cells. Indeed, a survey of all the different types of gene functions present in RNA viruses confirms that these two functions are universally present, whereas other functions are specific to certain groups and absent in others. Lastly, non-synonymous changes are usually expected to have a greater impact on fitness than synonymous changes.

We built our analysis upon previous studies by Drake (1993) and Drake & Holland (1999). These authors computed genomic mutation rates, among others, for MoMuLV (geometric mean of several determinations = 0.0031), poliovirus (geometric mean = 0.3090), Rous sarcoma virus (RSV, 0.43), VSV (geometric mean = 1.1093), and bacteriophage Qβ (6.5). For these five viruses, we did an exhaustive survey in the databases for protein sequences both of polymerases (RNA polymerase, or in the case of retroviruses MoMuLV and RSV, reverse transcriptase) and analogous coat proteins. For each virus and gene, we obtained the corresponding multiple alignment (The aligned data sets are available upon request). From each alignment, we computed the average distance (Poisson-corrected) for all possible pairs of amino acid sequences. Figure 2 shows the result of our analysis. We obtained a positive partial-correlation coefficient among total genomic mutation rate and non-synonymous genetic variability at fitness-related loci (r = 0.9722, 5 d.f., 1-tailed P = 0.0001), thereby giving support to the hypothesis of genetic variability being maintained at the population level by a mutation-selection balance. However, this correlation could be entirely attributed to the fact that bacteriophage Qβ had both the highest non-synonymous diversity and the highest genomic mutation rate. Excluding this virus, however, does not eliminate the positive correlation between non-synonymous diversity at fitness-related loci and mutation rate (r = 0.8557, 4 d.f., 1-tailed P = 0.0149). Hence, our above conclusion is not spurious but well founded from observation.

Figure 2.

Mutation-selection balance in natural viral populations. We obtained a measure of non-synonymous genetic variability for fitness-related genes for several viruses (RNA polymerases or RT represented with black dots; analogous coat proteins with white dots). Each one of the virus studied (see main text) had a different total genomic mutation rate. Horizontal error bars for mutation rates correspond with the minimum and maximum values reported. A significant positive correlation gives support to the hypothesis of mutation-selection balance.

In the virological literature, the notion of mutation-selection balance (along with a component of supra-individual selection) is known as the ‘quasispecies theory’ (Domingo, 2002). This concept has been widely used by virologists in describing the polymorphic nature of viral populations. However, the use of quasispecies theory to describe the dynamics of viral populations has recently been criticized by some virologists (Smith et al., 1997) as well as by population biologists (Jenkins et al., 2001; Holmes & Moya, 2002). These authors argue that no experimental evidence unambiguously demonstrates the quasispecies nature of viral populations. At best, classical population genetics provides an equivalent predictive and powerful theory. For us, perhaps the most convincing evidence supporting the quasispecies model is that reported by Duarte et al. (1994) and by Burch & Chao (2000). Duarte et al. (1994) demonstrated that the fitness of a viral population cannot be explained simply by adding up the fitness of all its components (Fig. 3). As we have already described above, these authors isolated a large number of individuals (n = 100) from a given clonal population of VSV and measured their fitness relative to a common competitor. Then they measured the fitness of the clonal population as a whole. They found that the average fitness estimated from the 100 individuals (1.0275 ± 0.0007) was significantly smaller than the fitness of the whole population (1.0405 ± 0.0010) (Mann–Whitney U= 136, P < 0.0001). Certainly, this observation does not undercut classical population genetic models, since models of kin selection make similar predictions too. At face value, it only demonstrates that the unit of selection is not necessarily the single virion but the whole population. Burch & Chao (2000) demonstrated that the ‘evolvability’ of RNA bacteriophage φ6 was constrained by the distribution of mutational neighbours. As predicted by quasispecies theory, they observed that a high-fitness member of the genotypic distribution may evolve to a lower mean fitness as the distribution of deleterious mutations is regenerated.

Figure 3.

The average fitness of a viral population cannot be explained simply adding up the fitness of all its components; the fitness of the entire population has emerging properties. Duarte et al. (1994) measured multiple times the fitness of a given population of VSV. They then isolated 100 individuals from the same population and measured the fitness of each one. As predicted by kin selection models (or by quasispecies theory), the fitness of the population as a whole is greater than the average fitness estimated from all its components.

FREQUENCY-DEPENDENT SELECTION

In an environment that affords ecological opportunity and where selection has favoured the evolution of niche specialists, the maintenance of coexisting genotypes is assured through the operation of density-dependent processes (Dempster, 1955). Assuming a constant primary resource, the fitness of a niche-specialist genotype will be a function of the availability of the primary resource. Thus, selection will operate in a negative frequency-dependent manner, favouring genotypes when they are rare (because resources will be abundant), but not when they are common (because resources will be scarce due to intense competition). Elena, Miralles & Moya (1997) demonstrated for the first time that negative frequency-dependent selection acts in laboratory populations of VSV (Fig. 4). They mixed two different genotypes (Fpop40 and a surrogate wild-type) at different frequencies and used these mixtures to infect a monolayer of susceptible cells. Daily, for the next five days, they sampled each lineage and, after convenient dilutions, infected a new monolayer. From the temporal variation in the frequency of each competitor, they estimated the fitness of Fpop40 relative to the surrogated wild-type at each initial frequency. As Figure 4 shows, a significant negative correlation was observed (R = 0.9100, F_1,71 = 342.1697, P < 0.0001). The point where both genotypes are equally fit (relative fitness of 1.0) corresponds to the equilibrium frequency in the population. For this experiment, the equilibrium frequency was estimated to be 0.44. This value represented a stable equilibrium at which both genotypes could coexist indefinitely, in the absence of genetic changes induced by mutation in any genotype (Elena et al., 1997).

Figure 4.

Negative frequency-dependent selection operates whenever a genotypes has a selective advantage when rare, that disappears when the genotype becomes common. Here, we show the results obtained by Elena et al. (1997) for VSV. They mixed two genotypes at initial frequencies ranging from 0.05 to 0.95 and determined the fitness of one genotype relative to the other (n = 6 replicates).

The reason for this negative frequency-dependent selection was not experimentally demonstrated, although the authors suggested two possible scenarios. First, it is quite possible that heterogeneity exists among host cells in terms of age, density, stage in the cellular cycle, or physiology, and that each viral genotype was using this variability in different ways. A second possibility is that there were within-cell interactions among viral genotypes. For instance, one genotype might be better when infecting alone, while the other may be better when coinfecting because it uses some functions in which it is defective or less efficient.

Recently, the role played by negative frequency-dependent selection in modulating viral diversity has been extended to HIV-1 (Yuste, Moya & López-Galíndez, 2002).

CONSEQUENCES OF EXTREME GENETIC VARIATION

Enormous clonal variability is a doubled-edged sword, with its positive or negative effects depending strongly on viral demographic conditions. Under conditions where competition among clones is minimized (small population sizes or genetic bottlenecks) and natural selection cannot purge mutations that have negative effects, genetic drift will randomly fix these deleterious mutations in the population. In contrast, increasing population sizes will lead to improvements in fitness, although the relationship between the rate of fitness improvement and population size has not been proven to be linear; above a certain size, competition among clones is so strong that the likelihood of fixing new variants decreases, thus slowing down the rate of adaptation.

MUTATIONAL LOAD, RANDOM DRIFT AND THE ONSET OF MULLER'S RATCHET

If the mutation rate is high in small populations, mutation-free individuals become rare and can be lost due to genetic drift. In clonal populations, as Muller (1964) noted, the losses is irreversible and the number of mutations increases in a ratchet-like fashion with successive loss of the least-mutated genotypes. Chao (1990) provided the first experimental evidence for the action of Muller's ratchet in RNA viruses, observing that in the segmented RNA bacteriophage φ6, consecutive bottlenecks of size one reduced population fitness by 22%. These observations were later generalized to include several other RNA viruses. In general, the data showed a common pattern of fitness declines. But the magnitude of the decline depended strongly on the virus studied. For instance, VSV losses ranged from extreme (≥99%) to minimal (∼0.1%), with an average decline of 18% after 20 transfers (Duarte et al., 1992, 1993, 1994; Clarke et al., 1993). The non-segmented RNA bacteriophage MS2 (de la Peña, Elena & Moya, 2000) suffered fitness declines similar to VSV (17%). Foot-and-mouth disease virus (FMDV) fitness declines were also extremely variable, ranging from complete extinction to ∼14% (Escarmís et al., 1996). Perhaps HIV-1 losses after 15 bottleneck passages differ a bit from this general pattern because the average effect is greater (from 99% to 89%) and the variance among lines is smaller (Yuste et al., 1999).

COMPETITION AMONG BENEFICIAL VARIANTS, CLONAL INTERFERENCE AND THE SPEED LIMIT OF ADAPTATION

One principle in evolutionary ecology describes how two species competing for a limited resource may experience an ‘arms-race’ effect: their performance improves relative to their ancestor, but the two remain in the same position with respect to each other (the Red Queen, as described by van Valen, 1973). Two neutral viral clones may coexist for long periods without being displaced by each other (i.e. they are equally fit), but when compared to their corresponding ancestors, they show fitness improvements. Beneficial mutations can arise in both genotypes, improving their respective fitnesses. If the fitness effects associated with beneficial mutations that fix are similar in magnitude, neither of the genotypes will show a clear advantage relative to each other. However, after long periods of coexistence, a beneficial mutation of strong effect eventually will arise in one genotype and not the other. When this happens, the first genotype will have a clear advantage over the second and will displace it from the population. This obligatory displacement by the newly generated ‘better-fit’ genotype is a reflection of another general principle of evolutionary ecology, the ‘Principle of Competitive Exclusion’ (Hardin, 1960). Both principles have been demonstrated to operate in VSV (Clarke et al., 1994; Quer et al., 1996).

Despite all of the above evidence for rapid adaptation attributable to a virus's ability to generate beneficial genetic variability, recent discoveries indicate that the latter is also a double-edged sword that can reduce the rate of adaptation. In a large asexual population, two or more coexisting lineages may be created by different beneficial mutations. When this happens, the lineage carrying the largest beneficial mutation eventually displaces all other lineages. Such interference reduces the fixation probability of any given beneficial mutation, augments the overall population fitness increase because of adaptive substitution, and increases the time between substitution events (Gerrish & Lenski, 1998). These predictions have been demonstrated to hold in experimental populations of VSV (Miralles et al., 1999; Miralles, Moya & Elena, 2000). The larger the population size, the larger the effect associated with a beneficial mutation that becomes fixed in the population. In other words, increasing the competition that takes place between genetic variants ensures that only the best possible genotype will become fixed. More importantly, population size has a diminishing-returns effect on the rate of adaptation as a consequence of the longer time required for fixation of beneficial mutations (Gerrish & Lenski, 1998; de Visser et al., 1999). On its way to fixation, the winning clone in these conditions must out-compete more and more genotypes that are carrying smaller-beneficial mutations.

Two important evolutionary conclusions can be drawn from the clonal interference model. (1) Adaptive substitutions appear as discrete events. They do not occur simply as the result of a single mutational event but instead represent the best possible candidate. (2) Because the rate of adaptation is not positively affected by increases in mutation availability, it is questionable whether the high mutation rate shown by RNA viruses has evolved because of the adaptive capacity it confers, as has been postulated by some authors. Furthermore, a decrease in mutation rate would benefit the population by slowing the accumulation of deleterious mutations. Therefore, we think it is more likely that high mutation rates are the result of a trade-off between keeping a compact genome and the costs of maintaining the enzymatic system required for error detection and correction.

CONCLUSION

We have reviewed three population genetic mechanisms that may explain the stable maintenance of genetic diversity in viral populations. Viral populations are characterized by enormous polymorphisms, generated as a consequence of the lack of proofreading mechanisms in viral RNA polymerases. We have shown evidence that this polymorphism can be partly explained by the transient existence of different beneficial genotypes on their way to fixation. In addition, the continuous input of mutations can be balanced by purifying selection, being the frequency of each new mutant inversely proportional to the fitness disadvantage associated to it. Lastly, the existence of negative frequency-dependent mechanisms can also account for the maintenance of a certain level of polymorphism. These three mechanisms are not mutually exclusive. Genetic diversity can be maintained by all three along with several others. For example, another mechanism that has recently received special attention from microbial population geneticists is adaptive radiation in heterogeneous environments (Rainey & Travisano, 1998; Buckling et al., 2000; Rainey et al., 2000; Kassen, 2002). Adaptive radiation in heterogeneous environments implies the existence of specialization to different ecological opportunities. Whenever vacant niches exist and selection is capable of generating specialist types that trade-off fitness in different environments, then polymorphisms are likely to arise and be stably maintained. Adaptive radiation can be easily explored with data from real infections, such as those above shown for AIDS patients, and with data from laboratory experiments where viral populations adapt to different cell tissues. As we have shown above, HIV-1 shows different levels of population heterogeneity on different cell tissues. In laboratory studies (Novella et al., 1999; Weaver et al., 1999; Crill, Whichman & Bull, 2000; Turner & Elena, 2000; Cooper & Scott, 2001), populations propagated in heterogeneous environments diversified more than control populations evolved in a homogeneous environment, in an equivalent time period. Furthermore, adaptation to certain cell types was specific and had negative pleiotropic fitness effects on alternative environments.

Selection experiments alone can be seen as powerful tools for demonstrating the plausibility of different mechanisms on the outcome of selection. However, they can tell us nothing about the importance of these mechanisms in natural viral populations; we still need field data. Here we have combined comparative analysis of data from natural viral populations together with results from experimental populations to explore whether certain population genetic mechanisms can explain the high viral population diversity observed both in the field and in the laboratory. We believe this will be the way to successfully advance the study of viral evolutionary biology.

Finally, we do not want to finish this article without making a comparison between the results reported here and similar ones reported for the bacterium Escherichia coli. Elena & Lenski (1997) also explored which mechanisms maintained population diversity in 12 experimental lines evolved during 10 000 generations in a simple and well-defined environment. As we have done here, they tested for the roles played by Fisher's fundamental theorem, the mutation selection balance, and frequency-dependent selection. Their results suggest that whereas the two former mechanisms played a minor role in maintaining fitness diversity, frequency-dependent selection was of greater importance. Putting all of these results together, we conclude that despite the huge differences between RNA viruses and bacteria in terms of genetic material, genome size and complexity, physiology and ecology, the same fundamental mechanisms of population genetics theory operate in both types of organisms, validating its generality. When Fisher, Wright, Haldane and others laid the foundations of population genetics, they were probably not thinking of microbes. Today, microbes turn out to be the perfect model for testing the predictions of their theories.

ACKNOWLEDGEMENTS

This research was supported by grant GV01-65 from the Generalitat Valenciana. F.M.C and R.S. are in receipt of fellowships from the Universitat de València and the Ministerio de Educación, Cultura y Deporte, Spain, respectively. We thank R.E. Lenski, E. Ostrowski and an anonymous reviewer for valuable comments and critical reading of the manuscript.