Several remarkable modifications in the mutant spectrum occur when viral populations replicate in the presence of mutagenic agents such as base or nucleoside analogues. The initial experiments by J.J. Holland and colleagues documented very modest (1.1- to 2.8-fold) increases of mutation frequency at single base sites of poliovirus and vesicular stomatitis virus, associated with severe decrease of infectivity, as a consequence of enhanced mutagenesis by a variety of mutagenic agents . Subsequent work showed that replication of different viruses in the presence of mutagenic base or nucleoside analogues could lead to virus extinction accompanied of very modest (from barely measurable up to 6.4-fold) increases of mutation frequency and mutant spectrum complexity ([28, 39, 74]; reviews in [29, 30, 71]). In our previous studies with FMDV and lymphocytic choriomeningitis virus we have observed that in the course of the transition of the viruses towards extinction: i) there is a 102- to 103-fold decrease in specific infectivity (number of infectious units divided by the total amount of viral RNA); ii) low viral loads and low replication capacity (fitness) favour extinction; and iii) internal, interfering interactions among mutagenized components of the mutant spectra play an important role in the extinction [32–36, 38, 39, 75].
The critical participation of the mutant spectrum in viral extinction led to the proposal of "lethal defection" as a model of virus extinction by lethal mutagenesis [34, 35], and motivated the phylogenetic and partition analyses of mutant spectra reported in the present study. The results support our previous conclusions on the general increase of mutant spectrum complexity -here documented by average genetic distances in phylogenetic trees and quantified by the AD parameter in PAQ- as a result of the various mutagenic treatments. Moreover, new insights into the events associated with enhanced mutagenesis have been unveiled by the comparison of phylogenetic and PAQ analyses. The NJ tree topology of a population evolved under enhanced mutagenesis conditions consists of a basal consensus sequence, with the individual components of the population radiating from it (the more mutated the population the bigger the radiation of the branches) and with no apparent population structure in the form of subclusters. Trees constructed with FMDV populations whose 3D (polymerase) included a substitution that decreased the sensitivity to R (RAp45 and RAp60)  displayed a slight reduction of branch length, but maintained the same general topology. This is in sharp contrast with population RA0p35, which, after only 5 passages in the absence of R, presents an internal structure completely rearranged, as shown by the NJ tree, and confirmed with ML and MP methods (Figure 3B, and additional file 1 : neighbour-joining, maximum likelihood and maximum parsimony analysis of populations RAp35 and RA0p35).
The average distance parameter AD of PAQ reflects the fine detail of the intra population structure of diversity better than the other estimators traditionally used, such as mutation frequencies or Shannon entropy. Mutation frequency does not capture the distribution of mutations among the individual components of a mutant spectrum, and the Shannon entropy reaches the maximum possible value of 1 when all the sequences under study are different, independently of the mutational load. This lack of resolution is circumvented by PAQ due to the pairwise comparison of all the clones with respect to the central sequence. In this sense, any subtle variation in the spatial distribution of mutation frequency may be amplified yielding a more sensitive, accurate and non-saturating description of the internal structure of the population. The present analysis has documented striking expansions of the average intrapopulation genetic distance associated with mutagenic treatments (Figure 4). Interestingly the analysis of the RA lineage revealed that, after an initial expansion of intrapopulation diversity, a remarkably fast contraction of the average distance occurred in two situations. First, a 2.15-fold reduction in AD value was observed when the drug treatment was discontinued for only 5 passages. Second, populations with FMDV including in its 3D (polymerase) replacement M296I -that decreases the sensitivity to R  – displayed a 1.90-fold reduction in AD value (RAp45 and RAp60 populations, Figure 4). The ratio of minimum over maximum mutation frequency yielded higher values (indicative of higher diversity), in populations RAp45 and RAp60 than in population RAp35 which is the most diverse. Therefore mutation frequencies (both maximum, minimum or their ratio) and Shannon entropy, are less definitory than PAQ (when analyzing AD) to characterize the internal genetic diversity of mutagenized viral populations.
A possible interpretation of intrapopulation contractions is that in the course of the mutagenic treatment subsets of genomes with better than average replication efficiency are produced. However, their potential replicative advantage can not be expressed due to the continuous mutational input due to the presence of R. When R pressure is removed, specific subsets of genomes showing higher than average replication capacity, replenish the population. A similar, albeit less pronounced, contraction would occur as a consequence of the dominance of viral mutants with decreased sensitivity to R.
The MARLS-derived populations that replicated in the presence of R (the RA lineage, Figure 1B), showed unexpected high resistance to extinction at passage 35 and subsequent passages, associated with a mutation that decreased the sensitivity to R . Despite such mutation, the number of total mutations fixed in the consensus sequence of population RAp35 was very high (1.13 mutations/passage, Figure 6) when compared with its ancestor (0.2 mutation/passage). Also the molecular clones derived from RAp35 presented a 12-fold increase in mutation frequency with respect to the control population, CAp35 (Table 1). The specific pattern of mutations of FU-treated, AZC-treated and R-treated populations (Figure 5), and the increase in mutant frequency with respect control populations (Table 1) suggest that the increase in mutation frequencies was effectively produced by the action of mutagens, in agreement with the mutational bias induced in the course of polymerization assays in the presence of R-triphosphate or FU-triphosphate using purified FMDV 3D (polymerase) (, Agudo et al submitted for publication).
Despite the high error frequencies induced by mutagen, mutations did not accumulate at random along the viral genome, but rather they accumulated at preferred genomic regions (Figure 7). The non-random accumulation of mutations and the excess of corrected synonymous mutations (Table 1) are consistent with purifying selection operating in the evolution of mutagenized populations. Viruses harbouring less deleterious mutations might have a selective advantage in a landscape of highly damaged viruses. In this view, the quasispecies would display a mutagenesis buffering activity, accepting those mutations that affect the more permissive regions of the viral genome. This dynamics is strikingly parallel to that observed in the course of bottleneck transfers carried out with the same FMDV clones C-S8c1 ([47, 48, 76–78] reviewed in ). In bottleneck transfers individual clones are selected for replication. In this situation Müller ratchet effect operates on the clones . In clones that resisted extinction, mutations accumulated preferentially at some genomic regions and, again, an excess of synonymous mutations was found [48, 77].
The present study supports the "lethal defection" model of viral extinction in that a limited number of mutations per genome can be sufficient to drive a virus to extinction [28, 34, 74]. Again, the comparison with FMDV clones subjected up to 409 bottleneck transfers is highly significant. In such clones extinction was not achieved despite the genomes reaching average mutation frequencies which are 5- to 37-fold higher than those associated with viral extinction by lethal mutagenesis [32, 39, 41, 48]. It has been proposed that in successive bottleneck transfers, the kinetics of mutation accumulation allows the fixation of compensatory beneficial mutations that counteract the Müller Ratchet effect [76–79].
Several theoretical models of lethal mutagenesis of viruses have been proposed, either as a direct consequence of quasispecies dynamics and its corollary concept of error catastrophe, or independently of error catastrophe [70, 81–84]. All models converge in that lethal mutagenesis is a feasible strategy to achieve virus extinction by mutagenic nucleotide analogues. What the models did not take into consideration is the key influence on extinction of internal interactions exerted among components of the mutant spectra. Lethal and interfering mutations impede a substantial "evaporation" (or "diffusion") of genomic sequences in sequence space , as it could not be otherwise, considering that we are dealing with loss of multiple biological functions compactly integrated in a viral genome . The phylogenetic and PAQ approaches used here should be extremely helpful to monitor in a quantitative fashion the evolution of mutagenized viral populations both in cell culture and in vivo, as they increase their mutational load and either succumb or escape extinction.