In this work we analyze how beneficial mutations spread in a virus population when replication takes place at increased error rate. To carry out this study we have chosen an RNA virus, the bacteriophage Qβ, which as a consequence of its high error rate
[40, 41] constitutes extremely heterogeneous populations composed by a complex mutant spectrum
. The presence of AZC in the growing medium increased the mutation frequency in viable viruses about an order of magnitude relative to that observed in virus populations evolved in the absence of AZC
. The mutation excess (Table
1 and Additional file
1) probably distributes among deleterious, beneficial and neutral mutations, although the much higher frequency of errors having a negative effect on fitness suggests that many of them must be deleterious. Thus, it is expected that beneficial mutations generated in the presence of AZC arise in unfavourable genomic contexts, which can reduce their selective advantages
[14–16]. In addition, when the error rate is high enough several beneficial mutations can expand simultaneously, which constitutes another important factor delaying their spreading in the population
[12, 13]. Thus, although replication at high error rate provides important adaptive advantages to RNA viruses
[43, 44], it is also true that when subjected to increased mutagenesis, they are frequently pushed to the edge of extinction
[45–47], as a result of both the increase in the number of deleterious mutations and the difficulties for the expansion of the beneficial ones.
The analysis of the consensus sequences and the mutant spectra of the AZC-evolved populations showed the fixation of two mutations and the presence of a number of polymorphisms (Table
1) higher than that found during the evolution of the same virus in the absence of AZC (see the two first subsections of Results). It should also be noted that, given the relatively low number of genomes analyzed in each population, the frequency of polymorphic mutations could be even higher than we have reported. Adaptation of bacteriophage MS2 to different selective pressures in the absence of mutagenic conditions also led to the presence of polymorphisms, although to a lower extent to that reported in the current study
[26, 29]. The findings obtained in studies performed with DNA viruses, which replicate with lower error rate than RNA viruses, are more difficult to compare to our results
. Nevertheless, evolution of bacteriophage T7 in the presence of a mutagen also showed a dramatic increase in the number of polymorphic substitutions
. Therefore, the high frequency of polymorphisms observed during the evolution of bacteriophage Qβ in the presence of AZC seems to be due, at least in part, to the increase of the error rate.
The high population sizes used for the transmission of the virus (above 107 pfu) together with the absence of clear evidences of hitchhiking suggests that most polymorphic mutations provide a selective advantage, at least in the selective environment provided by the mutagen. The lower values of the ratios ds/dn obtained for the polymorphic mutations than for the non polymorphic ones (Table
3) also supports that selection can be responsible, at least in part, of the high frequency reached by some substitutions. In addition, synonymous changes can also have a fitness effect mediated through cis interactions that may be significant in the case of highly compacted genomes, as it is the one of bacteriophage Qβ
[49–51]. In contrast to our assumption that most polymorphic mutations are beneficial, there are some studies showing that in natural populations of RNA viruses many high frequency mutations are deleterious, and will be later purged by natural selection
. The easiest way for these deleterious mutations to reach high frequency is their linkage with a beneficial mutation. In the absence of clear associations among mutations, this explanation could still be possible if mutations recur frequently as it seems to happen in bacteriophage Qβ (see below).
We have demonstrated that one of the mutations fixed (A2187C) and at least one of the polymorphic substitutions (G3945A) are beneficial in both the presence and the absence of AZC (Table
, whereas others (A2982G, and U3582C) only provide an advantage in the selective medium, having a fitness cost under standard replication conditions
. The results obtained with mutant QβA1746U deserve particular attention. This mutant has lower fitness than the virus Qβwt, when they grow independently. However, QβA1746U fixes in only 5 transfers when both viruses propagate together in the presence of AZC (Table
4). Fitness is a complex parameter that involves many traits, among which some of the most relevant are the lysis time, the adsorption rate, the burst size, and the stability of the particles outside of the host
. It is possible that some of these features influence differently the performance of a virus when it grows isolated or in competition. In good agreement with these considerations Springman et al.
 reported fitness gains in a population of bacteriophage T7 that, however, showed a clear decline in the burst size.
Our results also showed that the fitness advantages provided by the substitutions assayed were of different magnitude (Tables
5), which is in disagreement with the proposal of the multiple mutations model which posits a single characteristic value for the coefficient of selection of all beneficial mutations
[17, 18]. The situation becomes more complex due to mutations whose effects are context dependent. This is the case of substitutions A2982G and U3582C which were beneficial when present in separate genomes, and, however, were lethal or highly deleterious when present in the same genome, providing a clear example of antagonistic epistasis. Another substitution whose effect varied depending on the mutational context and on the competitor genomes was U3989C. This substitution behaved as deleterious in both the presence and the absence of AZC when it was present in the mutational context of the wild type virus (Table
4), and, however, it could be propagated when it was present in the mutational context of substitution A1746U (Figure
2b). These findings support the notion that epistatic interactions among mutations are very frequent in RNA viruses
[6, 54–58] and can influence the adaptive trajectories followed and the intensity of interference. The increase of the error rate may exacerbate the frequency of this type of interactions, which are not included in any of the current models developed to explain the fixation of beneficial mutations in asexual populations.
The fixation of substitution A1746U was not delayed by the presence in different genomes of either A2982G or U3582C (Table
5). As a consequence, this substitution was probably little affected by interference and could get fixed in population Qβ-AZC(t40). By contrast, substitutions A2982G and U3582C that fixed in 5 transfers when they competed with the wild type virus remained as polymorphisms during at least 20 transfers when they competed with each other (Table
5). A similar competition could be established among the different mutants present in the populations analyzed in this work, supporting that interference among mutations is one of the reasons underlying the sustained presence of polymorphisms.
A phylogenetic analysis carried out with the complete set of genomes obtained at different stages of the evolutionary series allowed us to group them into several clusters that represent independent evolutionary lines that may compete among them (Figure
3). In addition to the mutations representatives of each line, genomes can also contain different mutations among those represented at high frequency, and others that are exclusive of each virus. These mutations can modify the fitness of the genomes where they appear, adding further complexity to the process of interference, and making it difficult to fix any beneficial mutation and the emergence of a defined consensus sequence. The generation of new beneficial mutations in genomes carrying others still not fixed, together with the different magnitude of the effects of beneficial mutations, is more in line with the model of complete interference
 than with the models of clonal interference
[13, 14] or multiple mutations
Another important finding is that most polymorphic mutations were present in different genomic contexts, suggesting that they were generated repeated times. Results obtained in two adaptation experiments carried out with the bacteriophage MS2 also showed a high number of beneficial mutations in different mutational contexts
[26, 29]. A theoretical analysis demonstrated that this repeated presence of mutations was more probably due to their generation multiple times than to recombination
. The low capability of Qβ replicase to switch between templates does not allow for homologous recombination to be observed in most of the systems assayed
, as it would be masked by the much higher frequency of punctual mutations
. Therefore, we think that the presence of polymorphic mutations in different mutational contexts is more likely due to a high recurrence of punctual mutations than to recombination. This fact would allow combining several beneficial mutations in the same genome, alleviating in this way the costs of interference
. However, our results showing that two of the mutations having high selective value in the presence of AZC (A2982G and U3582C) cannot coexist in the same genome suggest that the fitness landscape for bacteriophage Qβ is highly rugged, with the existence of several local maxima where the virus could be trapped without reaching the best adaptive solution.
The high incidence of recurrent mutations could also account for the few hitchhiking mutations found in this study. Given the high mutation frequencies of the bacteriophage Qβ populations evolved in the presence of AZC, it would seem reasonable that almost each genome with a beneficial mutation in the process to fixation will also carry a set of hitchhiking mutations. However, the repeated occurrence of the beneficial mutations on different backgrounds also means that no single set of accompanying mutations gets fixed. A similar result was reported for bacteriophage MS2 adapting to cold temperatures
, and was also supported by theoretical calculations
Although at a first glance the interference among mutations can be seen as a negative feature that delays adaptation, there are also positive consequences that deserve to be pointed out. Maybe one of the most relevant is the coexistence in the population of multiple adaptive possibilities that can provide selective advantages by different molecular mechanisms. Population bottlenecks occurring during the propagation of the virus can lead to the fixation of different adaptive solutions. In this way, interference could be seen as a positive factor contributing to the diversification of populations, and permitting the exploration of the different local maxima that exist in rugged fitness landscapes, such as those described for RNA viruses
Interference among mutations can also play a relevant role in the extinction of viruses through lethal mutagenesis, a new antiviral strategy that derives from theoretical considerations
[65–67] and that consists in the treatment of virus infections through the artificial increase of the virus error rate
[46, 47, 68, 69]. Our results show that virus replication under mutagenic conditions can lead to the simultaneous presence in the mutant spectrum of multiple mutations conferring different advantages in the presence of the mutagen. The fixation of these mutations in particular individuals upon transmission of the virus through population bottlenecks, as indicated above, can lead to the co-circulation of viruses differing in their adaptive properties, jeopardizing in this way the efficacy of further treatments. Given the high error rates of RNA viruses, similar situations could also occur during the treatment with some replication inhibitors. Therefore, it is expected that future research on the evolutionary consequences of the interference among mutations also provides significant benefits to clinic and epidemiologic virology.