3.1. Selection for increased Qß multiplication rate through infectio interruptus
The first objective of this study was to obtain, through selection, a Qß population displaying a multiplication rate greater than that of its ancestral wt population. Multiplication rate (M
) is a measure of phage fecundity calculated as M
= (burst size)/(latent period) scaled as phages/h. M
may be increased by selecting for increased burst size and/or shortened latent period. Previous reports have shown that both burst size and latent period are susceptible to change through selection [8
]. These traits are shaped by the processes that take place during phage lytic growth: (i) adsorption; (ii) infection; (iii) progeny production; and (iv) lysis of the infected host cell. Because there is an indirect connection between progeny maturation and the timing of lysis [11
], we anticipated that the most effective approach for increasing M
would be to select for variants displaying an increased rate of progeny production. In a synchronously growing phage population, such variants would be expected to accumulate their progeny more rapidly than their wt counterparts, so that their numbers of progeny would exceed that of the wild-type at any time before lysis (but keeping in mind that final differences in burst size between the variants and the wild-type are also expected to depend on the length of the latent period).
The growth dynamics of a phage during a single cycle of infection may be discerned through two different types of synchronous-growth experiments: intracellular and one-step. In an intracellular growth experiment, the accumulation of total (intracellular + already released) phages, usually as plaque-forming units (PFUs), is determined throughout infection; in a one-step growth experiment, only the amount of free PFUs (in excess of infected but unlysed cells) is determined over time. Preliminary intracellular and one-step growth assays performed with wt Qß () showed that, under our experimental conditions: (i) Qß infective progeny (PFUs) become detectable 40–45 min after infection (i.e. the ‘eclipse’ period is 40–45 min long); and (ii) Qß requires ≈70 min (maximum latent period) to complete an infection. Based on these values, we decided to select for an increased M through a serial-passage experiment in which synchronously infected E. coli cultures would be artificially lysed between 45 and 70 min post-infection. This procedure would enrich a Qß population with rapidly replicating variants. In order to verify that selection was taking place, the PFU yield would have to be monitored in each passage. In addition, the test cultures would have to be processed in parallel with control cultures in which synchronously infected E. coli cells would be lysed after the typical Qß maximum latent period of 70 min. Therefore, at t = 0, we established six Qß lines: three early-progeny Treatment lines (TA, TB and TC), and three Control lines (CA, CB and CC). In order to (partially) synchronize phage growth during infection, adsorption was confined to an 8-min period. Cell cultures infected with treatment-line phages were artificially lysed at 55 min post-infection (i.e. 55 min after first mixing phages with cells), while in those infected with control-line phages lysis was artificially completed at 80 min post-infection. All lines were serially propagated in fresh E. coli cultures for 26 days (see §4 for additional information).
Figure 1. PFU production dynamics of a wt Qß population during infection of E. coli cells. Circles represent relative free PFU densities. Triangles represent relative total (free + intracellular) PFU densities. Separate symbols at each time represent different (more ...) a
shows the evolution of Qß PFU production in treatments and controls during the selection passages. This figure suggests a small increase in PFU yield in the treatment lines over passages. Indeed, when the data on PFU production collected for each line were independently fitted to a linear trend-line, the slopes of the resulting trend-lines deviated from zero for the three treatments but for none of the controls (data not shown), which indicated that, as a
suggests, only in the treatments were the amounts of PFUs produced in each passage significantly increased over the 26 passages. We also determined in each passage the PFU yields of the control lines at 55 min post-infection, and the corresponding data were fitted to linear trend-lines whose slopes showed no significant deviation from zero (data not shown). Moreover, when the PFU yields of the treatments were expressed as percentages of the corresponding average PFUs observed for the controls at every passage, the resultant values fitted an approximately exponential curve (b
= 0.64, d.f. = 73). Because we started the selection experiment with clonal lines, the relative PFU gains shown in b
most probably resulted from selection. Indeed, the exponential overall PFU gain displayed by the treatments is analogous to the trajectory typically observed for the fitness of RNA viruses and other micro-organisms undergoing adaptation to novel environments [12
]. This trajectory denotes a decreasing fitness gain over time as the populations under selection approach optimality. In RNA viruses, where beneficial mutations may appear in different, competing clones and where the effects of beneficial mutations are usually not independent, this trajectory may also reflect the impacts of clonal interference and epistasis (reviewed in [15
Figure 2. Evolution of Qß PFU production in Treatments (lines TA, TB and TC) and Controls (lines CA, CB and CC) during 26 serial passages. (a) PFU yield obtained for each of the lines at each passage. The bottom half of the graph shows the logarithms of (more ...)
3.2. Evolution of the multiplication rate and its determinants
The increase in PFU production observed in the treatment lines suggested that their M values might have increased. Therefore, fresh lysates of the founder wt population and of the experimental populations obtained at the 25th selective passage were prepared and used to characterize the corresponding populations: WT, obtained from the founder wt Qß population; TA25, TB25 and TC25, each obtained from a different treatment line; and CA25, CB25 and CC25, each obtained from a different control line.
shows the M
estimates obtained for the examined populations. The M
values of the Treatments25
were 1.2- to 1.8-fold larger than that of WT. M
values were calculated from the corresponding burst size (B
) and latent period (L
) means also shown in . B
values were determined through sigmoid curve-fitting of a pool of one-step growth curves obtained for each experimental population (see the electronic supplementary material, figures S1–S5), as described in §4. Compared with WT, all Treatments25
showed a significantly reduced average L
(Holm Sidak's Multiple Comparisons Test, adjusted p
-values as shown in , d.f. = 128), which indicated that the latent periods of the treatment lines decreased during selection. In addition, all Treatments25
also exhibited an average B
1.2- to 1.7-fold larger than the average B
of WT. Thus, all Treatments25
achieved increased M
values due to combined increases in their B
values and decreases in their L
values. This outcome was unexpected because theoretical and experimental studies on phage latent-period evolution have indicated the existence of a trade-off between latent period and burst size [8
]. Thus, it seems logical that a reduction in the time available for progeny production would reduce the total amount of progeny produced (as long as other phage traits such as eclipse period and maturation rate are kept constant).
Qß life-history traits estimated for the Treatments25, Controls25 and WT. Estimates were obtained as detailed in §4.
We further investigated the process(es) underlying the changes observed in the parameters L
. A reduction in L
coupled with an increase in B
suggested an increase in the rates of infection (sensu stricto
) and/or progeny production. However, the rate at which Qß infects a host cell appeared difficult to measure reliably, and the detailed mechanism of infection remains unknown (reviewed in [18
]). Therefore, we did not attempt to determine the infection rates of the experimental populations. On the other hand, and as indicated earlier, phage progeny production involves phage genome expression, replication and progeny maturation. Because of its easy estimation, we used the rate of PFU maturation (m
) as an indicator of the overall rate of progeny production. Although m
has been shown to be constant during infection for a variety of phages [19
], our intracellular growth assays conducted with Qß revealed that the PFU density follows a sigmoid curve characterized by a gradual decrease in the rate of PFU production (or m
, represented by the slope of the curve), probably due to the termination of the latent period (). Therefore, a series of intracellular growth curves obtained for each experimental population were fitted to a sigmoid curve (see the electronic supplementary material, figures S1–S5), whose slopes were used as estimates of m
. The three Treatments25
displayed average m
values higher than for WT () but the overall differences were not significant (Holm Sidak's Multiple Comparisons Test, adjusted p
-values > 0.8, d.f. = 100). This result hinted that the expected trade-off between L
may have been partly prevented by an increase in m
Because the reduction in L
observed for the three Treatments25
might also have resulted from an increase in their adsorption rates, we determined the adsorption rate constant (k
, ml min−1
) of each of the experimental populations. Altogether, the k
values estimated for the Treatments25
were significantly higher than those obtained for WT and Controls25
(; Mann–Whitney test, two-tailed exact p
< 0.0001). Yet, for this difference to be meaningful, such an increase in k
should have been paralleled by a gain in infectivity under our typical conditions, where only a minority of the particles has time to adsorb. Phage adsorption to the host cell is commonly described by Nt
, where C
is the host cell density in the adsorption mixture and Nt
is the density of phages that remain free after adsorption time t
. Note that, under these circumstances, k
only indicates how rapidly free phages disappear from the adsorption mixture but says nothing about the efficiency of adsorption or, more precisely, whether the adsorbed phages actually infect their host cells [23
] (although more intricate protocols to determinate phage adsorption rates, based on increases in numbers of infected bacteria over time, could provide this information if needed). Thus, one may enquire whether the increase observed in k
was accompanied by a parallel increase in infectivity. We investigated whether such a relation existed through correlation analysis between the infectivity values (determined as detailed in §4) and the k
values estimated for all of the experimental populations (). Results from such analysis (Pearson's r
= 0.6468, R2
= 0.4183, two-tailed p
= 0.0001) indicate that, as suggests, in Qß, infectivity under our conditions is proportional to k
and that the increase in k
observed for the Treatments25
was paralleled by an increase in infectivity.
Figure 3. Relationship between infectivity and adsorption in Qß. Different symbols represent observations obtained for different populations: the clear circles, squares and rhomboids stand for TA25, TB25 and TC25, respectively, while the black upright and (more ...)
Overall, the degree of phenotypic evolution varied among the experimental populations. Owing to the randomness of mutagenesis, the occurrence of phenotypic differences among parallel populations in evolution experiments is expected. Here, however, some phenotypic variability might have also resulted from small differences in the manipulation of the parallel experimental lines during the selection experiment (see §4 for details). Thus, TC25 displayed the greatest values of M, B and m, while TB25 showed the smallest average L. Controls25 also showed some degree of phenotypic evolution, probably as a result of an adaptation to grow in E. coli NR16205 cells under our experimental conditions. It is worth noting here that, during the serial-passage experiment, artificially lysing the experimental cell cultures infected with control lines 80 min after infection might have terminated longer latent periods, hence selecting for phage variants with a progeny production higher than that of wt Qß, as in the case of the treatment lines. Thus, the phenotypic evolution observed for the Controls25 was in the same direction as the evolution observed for the Treatments25, namely increased M, B, m and k and decreased L, but all these changes were subtle and, at least for L and m, which could be statistically analysed, did not render values significantly different from those observed for WT.
3.3. Evolution of Qß plaque-forming unit virion stability
PFU stability, which represents the virion's capability to remain infectious over time, was estimated from the residuum of free PFUs in a phage suspension after 6 h of incubation in LB medium at 37°C with gentle shaking. An endpoint measure was preferred over a rate because the evolved populations were anticipated to be genetically heterogeneous and their PFU decay rates were thus expected to vary over time and to be difficult to compare, as previously observed with phage T7 [24
]. The results showed that the three Treatments25
have low PFU stability compared with WT (). When the stability values of the Treatments25
were combined and compared with the corresponding values of WT and Controls25
(also combined), the difference was significant (Mann–Whitney test, two-tailed exact p
PFU stability (expressed as percentage) of the wt and evolved Qß populations. Bars represent means ± s.e.m. (n = 6).
These results, together with the changes observed in M
, imply a trade-off between Qß fecundity and lifespan. What may be the mechanism(s) underlying this trade-off? In general, PFU decay may be due to the degradation of any of the virus components and consequent loss of virion integrity. For instance, in the ssRNA virus causing foot-and-mouth disease, genome insertions decreased virion thermal stability [25
], while large deletions (greater than 400 bases) increased virion stability and lifespan [26
]. Similarly, across a wide range of double-stranded DNA (dsDNA) coliphages, PFU decay rate was positively correlated with the density of the packaged genome, although there was no correlation between this trait and the multiplication rate [6
]. Among the phage traits analysed, only the capsid surface mass (the molecular weight of the capsid shell divided by the surface of the capsid) was negatively correlated with both mortality and multiplication rates, suggesting that the lesser the relative amounts of time and energy invested in the production of each phage particle, the higher the mortality rate [6
Although our data do not lend themselves to statistical inferences on the possible mechanistic causes for the fecundity/lifespan trade-off reported here for Qß, it is notable that the experimental population showing the lowest PFU stability values, TB25
, was the one showing the shortest latent period. Although the experimental population displaying the highest average B
, the difference in average B
is probably not significant considering the variability in burst size usually observed for Qß [27
]. Overall, these results seem to suggest that the less the amount of time that Qß needs to invest in progeny production (due, for instance, to a faster but less stable folding of the progeny genome), the shorter is the lifespan of the progeny. Likewise, as shows, PFU stability and latent period showed a strong positive relationship among the experimental populations (nonlinear lines regression, R2
= 0.6463, d.f. = 28).
Figure 5. Relationship between PFU stability and latent period in Qß. Different symbols represent PFU stability and latent period means obtained for different populations: the clear circles, squares and rhomboids stand for TA25, TB25 and TC25, respectively, (more ...)
3.4. Evolution of Qß growth rate
Empirical evidence of a trade-off between virion fecundity and stability in a challenging environment was recently shown for the dsDNA coliphage T7 [24
], wherein T7 virions adapted to survive in 6 M urea displayed a reduced growth rate in cell culture. Similarly, a trade-off between virion survival and reproduction was analysed for the dsRNA phage
6 in another recent study [28
], wherein phage populations adapted to periodic heat shocks displayed decreased fitness. A confounding aspect of these two reports is that, contrary to multiplication rate, both phage growth rate and fitness involve survival per se
, and hence their estimation depends strongly on local environmental conditions. For example, all other things being equal, increased virion stability might or might not be coupled to decreased growth rate or fitness depending on the host-cell density: if this density is low enough to render a competition advantage to those virions showing higher stability, increased stability would be coupled to an apparent increased growth rate or fitness. To illustrate this example, summarizes the results of a set of assays in which we determined the growth rates of our experimental populations under conditions of low and high host-cell density (LHD and HHD, respectively). Under LHD conditions, Treatments25
showed significantly lower growth rates than WT and Controls25
(Mann–Whitney test, two-tailed exact p
= 0.0110). This result suggests that, under these conditions, the increased M
values displayed by the Treatments25
were insufficient to compensate for their decreased PFU stabilities between consecutive infections. Conversely, under HHD conditions, in which the dispersal period (the time between two consecutive infections) was reduced to a minimum, Treatments25
showed growth rates significantly higher than those of WT and Controls25
(Mann–Whitney test, two-tailed exact p
= 0.0004), in agreement with their higher M
values. These results show that, when a trade-off between fecundity and any other component of fitness is analysed, special care must be taken in the parameter used to determine changes in the reproduction ability.
Growth rates of the wt and evolved Qß populations under conditions of high and low host-cell density (HHD and LHD, respectively). Growth rates are in PFU doublings per hour. Symbols represent medians and ranges (n = 3).
When the constraints that govern phage life-history evolution are analysed within the framework of the optimal foraging theory (concerning the adaptive evolution of animal foraging behaviour), the latent period equals the residence period, i.e. the time spent feeding on a particular source of energy (in this analogy, a bacterial cell; reviewed in [29
]). According to the marginal value theorem of the optimal foraging theory, when energy sources are found in patches, the residence period is expected to evolve towards an optimum that ensures the highest fitness available [30
]. When applied to phage evolutionary biology, this theorem predicts that phages would evolve shorter latent periods in environments containing high densities of susceptible host cells [17
]. This prediction has been empirically supported in a number of experiments [8
], although attempts to reach an optimal latent period have failed, probably because of unidentified constraints [10
]. Our results also seem to endorse such a prediction because the evolution of shorter Qß latent periods produced higher growth rates under HHD conditions, and vice-versa.
3.5. Genetic evolution
The Qß genome contains 4217 bases organized in three cistrons that encode a total of four proteins whose functions are well determined: (from 5′ to 3′) A2, which mediates both adsorption to the host cell and post-replicative host lysis; the Coat protein and its elongated A1 or Read-Through version, which is required for progeny maturation; and the catalytic ß subunit of the Qß replicase. With the aim of determining which specific changes in the genes that code for these proteins produced the observed phenotypic changes, the consensus genome sequences of WT and of the populations resulting from the 25th and 26th selective passages were obtained and analysed. The genetic changes, all of them polymorphisms, detected for each of the Treatments25 and Controls25 relative to the founder wt population, are shown in . The same polymorphisms (with slight modifications in the corresponding percentages) were observed for both the Treatments26 and the Controls26 (data not shown).
Table 2. Genetic changes in the consensus genome sequence of each of the Treatment25 and Control25 populations relative to the ancestral wt Qß population. Percentages represent the proportion of the observed change. The function of the proteins encoded (more ...)
A total of 11 single-base-substitution polymorphisms were detected: two in the 5′ untranslated region (5′-UTR), three in the A2-coding gene, five in the A1-coding gene and one in the gene encoding the ß subunit of the Qß replicase. Two of the polymorphisms (at sites 47 and 2016) exhibited two different non-wt alleles each. In the 5′-UTR, one of the two detected polymorphisms was present only in the TA lineage while the other was also present in TC (). These two polymorphisms mapped at genome positions 47 and 52. Both positions are located in a sequence of the Qß genome (bases 47–57) that forms a hypothetical long-distance structural interaction with another sequence of the Qß genome (bases 459–474) [31
]. In addition, position 52 maps in the binding-site sequence (bases 50–78) of the ribosomes during A2-translation initiation [32
]. Thus, we may reasonably venture that the 5′-UTR polymorphisms affect Qß genome folding and/or translation. Interestingly, changes at the same two positions of the Qß 5′-UTR have previously been observed in Qß populations evolved in E. coli
], which further suggests that these positions are involved in one or more important functions of the Qß life cycle.
In the gene that encodes the A2 protein, we detected three polymorphisms, located at genome positions 143, 149 and 1294 (). The first two positions map in the amino-terminal half of A2, where at least the lytic function of this protein resides [34
]. The polymorphism detected at position 143 is the only one present in all treatments and controls. This polymorphism comprises a U → C mutation that reached a substantially higher frequency in the treatments than in the controls and was almost fixed in TC. These observations, combined with the phenotypic results, suggest that this change is related to the increased adsorption rate constant (k
) observed for Treatments25
. Thus, there is some parallelism across the experimental populations between the frequency shown by this particular mutation and the changes observed in k
. This mutation, which translates as L28P in A2, was previously observed among Qß A2 mutants able to lyse E. coli
cells resistant to the lytic function of wt A2 [34
], which indicates that it may modify the lytic activity of A2. Our results further suggest that this mutation may also modify the attachment activity of A2 and thus that both functions of A2 (lysis and attachment) may map to close regions of A2. The polymorphism detected at position 149 of the A2-coding gene, which also comprised a U → C mutation, was present in all controls but in none of the treatments (). During the selection experiment, only the control lines were given sufficient time in each passage to lyse their host cells, so that this specific change might be related to the lysis function of A2.
Interestingly, among treatments, the gene encoding the Coat/A1 protein accumulated the greatest number of detectable polymorphisms, mostly in the part of the gene exclusively involved in coding A1. (A1 is translated when a ribosome incorporates tryptophan at the natural UGA stop codon of the Coat-coding gene located at position 1743 of the Qß genome.) The changes observed in the A1-coding gene might be linked to the increase in the rate of PFU maturation (m
) observed for all the treatments if this increase were actually due to an enhanced efficiency for Qß assembly, which, unfortunately, we cannot verify with the available data. All the non-synonymous polymorphisms detected in the A1-coding gene were located in a sequence of about 100 bases (genomic positions 1915–2016) of the total 993 bases coding for A1, which indicates that residues important for the function of A1 map in this region. One additional polymorphism (at position 1649), observed only in TB, involved a synonymous change, suggesting that this polymorphism may serve a structural function [35
Finally, we detected a polymorphism in the ß-subunit gene that maps in the bridge domain of the Qß replicase [36
]. This domain is suggested to interact with the Qß replicase cofactors EF-Tu and S1 and is probably also involved in unwinding the duplex between the template and the nascent portion of the product when the duplex reaches a maximum size of six to seven base pairs [36
]. Thus, the observed change in the ß subunit may have modified Qß RNA synthesis initiation and/or extension. Interestingly, this change was observed only in TC, which also displayed the highest overall multiplication rate.
All the analysed populations showed genetic evolution, although the 10 polymorphisms that the treatments accumulated exceeded the three in the controls. Among the treatments, TA acquired the most polymorphisms (eight), followed by TC (with six) and TB (with four). In these populations, convergent evolution (the same polymorphisms showing similar frequencies) was observed at three genome positions (143, 2001 and 2016), and two other convergent changes were observed between TA and TB (at genome positions 52 and 1930). Among the controls, the same three changes were detected in the three populations, two of these polymorphisms also being observed among the treatments ().
Convergent evolution is a common result of evolution experiments carried out with RNA viruses (reviewed in [37
]). Apart from the obvious reason that, in this type of experiment, parallel virus lines are challenged with a set of identical environmental conditions, there are some unique characteristics of RNA viruses that favour convergent evolution. Notable among these are the shortness, compactness and structural constraint of their genomes, which offer a limited array of adaptive responses to a given environmental change [38
]. In addition, antagonistic epistasis and clonal interference, which are thought to be common in RNA virus populations (reviewed in [15
]), would allow the same beneficial mutations with strong effects to be preferentially fixed in most, if not all, parallel evolving populations.