Convergent evolution toward an improved growth rate and a reduced resistance range in Prochlorococcus strains resistant to phage

Significance

High abundances of the important primary producing cyanobacterium, Prochlorococcus, and its parasitic phages, inhabit vast expanses of the world’s oceans. Their coexistence is facilitated by genetic diversity that has led to an assortment of Prochlorococcus subpopulations with differences in susceptibility and resistance to co-occurring phages. Here, we investigated the fate of recently emerging phage-resistant Prochlorococcus strains. We found that genetic diversification increases, as these strains evolve toward an improved growth rate and reduced resistance range, leading to phenotypes intermediary between the original susceptible and initial resistant strains. These findings suggest a continual increase in the combinatorial interactions between Prochlorococcus and its phages and that the oceans are populated with rapidly growing Prochlorococcus cells with varying degrees of phage resistance.

Abstract

Prochlorococcus is an abundant marine cyanobacterium that grows rapidly in the environment and contributes significantly to global primary production. This cyanobacterium coexists with many cyanophages in the oceans, likely aided by resistance to numerous co-occurring phages. Spontaneous resistance occurs frequently in Prochlorococcus and is often accompanied by a pleiotropic fitness cost manifested as either a reduced growth rate or enhanced infection by other phages. Here, we assessed the fate of a number of phage-resistant Prochlorococcus strains, focusing on those with a high fitness cost. We found that phage-resistant strains continued evolving toward an improved growth rate and a narrower resistance range, resulting in lineages with phenotypes intermediate between those of ancestral susceptible wild-type and initial resistant substrains. Changes in growth rate and resistance range often occurred in independent events, leading to a decoupling of the selection pressures acting on these phenotypes. These changes were largely the result of additional, compensatory mutations in noncore genes located in genomic islands, although genetic reversions were also observed. Additionally, a mutator strain was identified. The similarity of the evolutionary pathway followed by multiple independent resistant cultures and clones suggests they undergo a predictable evolutionary pathway. This process serves to increase both genetic diversity and infection permutations in Prochlorococcus populations, further augmenting the complexity of the interaction network between Prochlorococcus and its phages in nature. Last, our findings provide an explanation for the apparent paradox of a multitude of resistant Prochlorococcus cells in nature that are growing close to their maximal intrinsic growth rates.

Large bacterial populations are present in the oceans, playing important roles in primary production and the biogeochemical cycling of matter. These bacterial communities are highly diverse (1–4) yet form stable and reproducible bacterial assemblages under similar environmental conditions (5–7).

These bacteria are present together with high abundances of viruses (phages) that have the potential to infect and kill them (8–11). Although studied only rarely in marine organisms (12–16), this coexistence is likely to be the result of millions of years of coevolution between these antagonistic interacting partners, as has been well documented for other systems (17–20). From the perspective of the bacteria, survival entails the selection of cells that are resistant to infection, preventing viral production and enabling the continuation of the cell lineage. Resistance mechanisms include passively acquired spontaneous mutations in cell surface molecules that prevent phage entry into the cell and other mechanisms that actively terminate phage infection intracellularly, such as restriction–modification systems and acquired resistance by CRISPR-Cas systems (21, 22). Mutations in the phage can also occur that circumvent these host defenses and enable the phage to infect the recently emerged resistant bacterium (23).

Acquisition of resistance by bacteria is often associated with a fitness cost. This cost is frequently, but not always, manifested as a reduction in growth rate (24–27). Recently, an additional type of cost of resistance was identified, that of enhanced infection whereby resistance to one phage leads to greater susceptibility to other phages (14, 15, 28).

Over the years, a number of models have been developed to explain coexistence in terms of the above coevolutionary processes and their costs (16, 29–32). In the arms race model, repeated cycles of host mutation and virus countermutation occur, leading to increasing breadths of host resistance and viral infectivity. However, experimental evidence generally indicates that such directional arms race dynamics do not continue indefinitely (25, 33, 34). Therefore, models of negative density-dependent fluctuations due to selective trade-offs, such as kill-the-winner, are often invoked (20, 33, 35, 36). In these models, fluctuations are generally considered to occur between rapidly growing competition specialists that are susceptible to infection and more slowly growing resistant strains that are considered defense specialists. Such negative density-dependent fluctuations are also likely to occur between strains that have differences in viral susceptibility ranges, such as those that would result from enhanced infection (30).

The above coevolutionary processes are considered to be among the major mechanisms that have led to and maintain diversity within bacterial communities (32, 35, 37–39). These processes also influence genetic microdiversity within populations of closely related bacteria. This is especially the case for cell surface-related genes that are often localized to genomic islands (14, 40, 41), regions of high gene content, and gene sequence variability among members of a population. As such, populations in nature display an enormous degree of microdiversity in phage susceptibility regions, potentially leading to an assortment of subpopulations with different ranges of susceptibility to coexisting phages (4, 14, 30, 40).

Prochlorococcus is a unicellular cyanobacterium that is the numerically dominant photosynthetic organism in vast oligotrophic expanses of the open oceans, where it contributes significantly to primary production (42, 43). Prochlorococcus consists of a number of distinct ecotypes (44–46) that form stable and reproducible population structures (7). These populations coexist in the oceans with tailed double-stranded DNA phage populations that infect them (47–49).

Previously, we found that resistance to phage infection occurs frequently in two high-light–adapted Prochlorococcus ecotypes through spontaneous mutations in cell surface-related genes (14). These genes are primarily localized to genomic island 4 (ISL4) that displays a high degree of genetic diversity in environmental populations (14, 40). Although about a third of Prochlorococcus-resistant strains had no detectable associated cost, the others came with a cost manifested as either a slower growth rate or enhanced infection by other phages (14). In nature, Prochlorococcus seems to be growing close to its intrinsic maximal growth rate (50–52). This raises the question as to the fate of emergent resistant Prochlorococcus lineages in the environment, especially when resistance is accompanied with a high growth rate fitness cost.

To begin addressing this question, we investigated the phenotype of Prochlorococcus strains with time after the acquisition of resistance. We found that resistant strains evolved toward an improved growth rate and a reduced resistance range. Whole-genome sequencing and PCR screening of many of these strains revealed that these phenotypic changes were largely due to additional, compensatory mutations, leading to increased genetic diversity. These findings suggest that the oceans are populated with rapidly growing Prochlorococcus cells with varying degrees of resistance and provide an explanation for how a multitude of presumably resistant Prochlorococcus cells are growing close to their maximal known growth rate in nature.

Results

In our previous study, we isolated Prochlorococcus substrains that were resistant to various T7-like podoviruses. These resistant substrains were derived from four susceptible Prochlorococcus strains belonging to two high-light–adapted ecotypes: MED4, MIT9515, MIT9215, and MIT9312 (14). In most cases, a single mutation conferred resistance to between two and five phages. Approximately one-half (48%) of the resistant substrains displayed a reduction in growth rate and another 22% had a greater susceptibility to other phages, whereas there were no detectable costs associated with 30% of the resistant substrains (14).

Here, we followed the evolution of 12 of the above resistant strains (Table S1) at the phenotypic and genotypic levels. The results were compared with those for paired susceptible control substrains (wild type) that were isolated at the same time as the initial resistant substrains and grown under the same conditions throughout the period of the study. To assess whether our findings constitute reproducible evolutionary pathways or less predictable stochastic events, we also investigated 68 independently evolving clones that were derived from three of these strains (Fig. 1). We use the term “initial resistant substrain” (iR) to refer to strains immediately after their selection for resistance. These substrains were derived from single resistant cells that formed colonies on plates (Fig. 1). They were then grown in liquid cultures and transferred serially over time, giving rise to what we refer to as “evolved cultures” (Ev). Seven months after isolation of the resistant substrains, three of the evolved cultures (R7, R8, and R9) were plated to produce clones, each of which was then grown independently in liquid (Fig. 1). We refer to these as “evolved clones” (Ev-cl). We use the term “strain” in a more generic fashion to refer to evolved cultures and/or clones. A timeline of the phenotypic and genotypic analyses performed on these evolved cultures and clones is found in Fig. S1.

Fig. 1.

Derivation of resistant substrains and evolved cultures and clones. Isogenic susceptible colonies of Prochlorococcus were used to isolate susceptible control colonies in the absence of phage (Top) and to select for phage-resistant colonies in the presence of selecting phage (Middle), as previously described (14) (Table S1). Resistant colonies were then transferred and grown in liquid medium and are termed “initial resistant substrains” (iR). These substrains were then transferred serially over time, giving rise to the “evolved resistant cultures” (Ev). Susceptible colonies were also transferred and grown in liquid medium for the duration of the experiment and are termed “susceptible wild-type controls” (WT). Seven months after isolation of the initial resistant substrains, three of the evolved cultures (R7, R8, and R9) were plated to produce single colonies (Bottom). More than 20 colonies were selected randomly from the plating of each culture. These latter colonies were grown independently of each other and transferred serially in liquid medium, giving rise to the “evolved resistant clones” (Ev-cl). The number of generations in the timescale refers to the estimated range for the R7 and R8 evolved resistant strains.

Phenotypic Change of Evolved Resistant Strains.

Following 9 mo of semibatch culturing after isolation of the initial resistant substrains (110–135 generations later), we noticed that the growth rate of two of them (MED4_R7 and MED4_R8, referred to as R7 and R8 from here on) had significantly improved (Fig. 2B). To assess whether this was a common phenomenon, we compared the growth rate of 12 evolved cultures to that of their initial resistant substrains (via comparisons of growth rates to the susceptible controls; Methods). This was done at 24, 27, or 40 mo after initial isolation, by which time the selecting phage had become extinct in all tested cultures except for R9 (Methods). Six out of seven of the evolved cultures with an initial growth rate cost, exhibited an improved growth rate, even reaching the growth rate of the wild type in some cases (Fig. 2 B and C). In contrast, the growth rate of one evolved culture, MED4_R9 (R9), had declined further (Fig. 2B). No growth rate change was observed for the five evolved cultures for which no initial growth rate cost was found (Fig. 2A). Furthermore, no growth rate improvement was observed for the susceptible control strains over the same period (Fig. S2). Thus, the growth rate of the evolved cultures improved over time but only in cases in which they had an initial growth rate cost.

Fig. 2.

Phenotypes of evolved resistant cultures. Growth rate (Top), resistance range (Middle), and enhanced infection (Bottom) of evolved cultures (Ev) are shown in comparison with the initial resistant substrains (iR). Resistant strains R1 to R9 were derived from Prochlorococcus MED4 (A and B), whereas R69 and R70 were derived from Prochlorococcus MIT9312, and R74 and R77 were derived from Prochlorococcus MIT9215 (C). The growth rates of the evolved cultures were normalized to those of the susceptible control cultures measured at the same point in time. Resistance ranges were tested against the following T7-like cyanophages: P-GSP1, P-TIP38, P-SSP7, P-TIP1, and P-TIP2 for strains derived from MED4; P-SSP2 and P-SSP3b for MIT9312-derived strains; and P-SSP1, P-RSP1, and P-TIP39 for strains derived from MIT9215. Dark blue indicates resistance, and black indicates susceptibility. Enhanced infection costs were tested for the MED4 strains against the P-GSP1 and P-TIP38 phages (+, has enhanced infection cost; −, no enhanced infection cost). Strains for which enhanced infection could not be tested because they were resistant to all phages in our collection are marked “nd” for no data. The growth rates and resistance ranges of R7 and R8 evolved cultures were first tested 9 mo after initial isolation of the resistant substrains (t1) and tested again 24–27 mo after isolation (t2), at the same time as the rest of the evolved cultures in B and C. Growth rates and resistance ranges in A as well as enhanced infection in A and B were tested 40 mo after initial isolation. Data shown are average and SD of three to six biological replicates. Statistically significant growth rate differences (t test) between Ev and iR are marked above the bars for each strain (*P < 0.05; **P < 0.01; ***P < 0.001). The iR values reported here were taken from ref. 14.

To assess whether there was also a reduction in the enhanced infection cost of resistance over time, we compared the rate of population decline of the evolved cultures to that of their initial resistant substrains when infected with other phages. Of the five evolved cultures tested (corresponding to those that had an initial enhanced infection cost), only MED4_R1 (R1) lost the enhanced infection cost, as shown by similar infection dynamics to that of the wild-type susceptible strain rather than the more rapid decline of the revived initial resistant substrain (Fig. 3A). In contrast, the four other evolved substrains retained their enhanced infection cost (Figs. 2A and 3B). Thus, a reduction in the enhanced infection cost of resistance occurred but less frequently than the reduction in the growth rate cost. This may be due to a greater selection pressure being exerted on the growth rate cost relative to the enhanced infection cost, as a growth rate cost is manifested under all conditions and at all times whereas the cost of enhanced infection is local in time and place, being expressed only when phages capable of infecting it more efficiently are present.

Fig. 3.

Population decline of evolved resistant cultures during phage infection. Growth of the R1 (A), R5 (B), and R8 (C) evolved cultures (Ev; red) were compared with susceptible wild-type control strains (WT; black) and initial resistant substrains (iR; blue) when infected with the P-TIP38 phage (filled circles). They were also compared with growth of noninfected cultures (open circles). Average and SD of three biological replicates. The R1 evolved culture (A) responded differently to the initial resistant strain (P < 0.001) but not to the susceptible wild-type control (P = 0.304), indicating a loss of the enhanced infection cost of resistance. Similar results were obtained for the R1 evolved culture during infection with the P-GSP1 phage. The R5 evolved substrain (B) did not respond differently to the initial resistant substrain (P = 0.243) but responded differently to the susceptible wild-type control (P < 0.001), indicating that it retained the enhanced infection cost. These results are representative of infection of the R5 culture with the P-GSP1 phage, as well as for additional evolved cultures (R1, R4, and R6) during infection with the same two phages. The R8 evolved culture (C) was infected more slowly than the susceptible wild-type control (P < 0.001). These are representative of 6 out of 10 host–phage interactions with the R1, R7, and R8 evolved cultures. The initial resistant substrains used in this assay (A and B) were revived from frozen stocks. AU, arbitrary units.

We then wondered whether the resistance range of these strains had also changed over time. Nine months after isolation, the R7 and R8 cultures had maintained their resistance to all five phages (Fig. 2B). However, 24 mo after initial isolation, we observed a reduction in resistance range for these evolved cultures, and after 40 mo, for the R1 culture (Fig. 2 A and B). Intriguingly, in most cases, the population decline of these evolved cultures was slower than when the same phages infected the wild-type susceptible strains (Fig. 3C). None of the three evolved cultures lost resistance to all phages, maintaining resistance to at least the phage used for isolation of the initial resistant substrain. In contrast, the other nine evolved cultures retained their full range of resistance (Fig. 2).

Grouping the evolved cultures according to changes in both cost of resistance and resistance range showed all possible phenotypic permutations: two displayed both a reduced cost of resistance and a narrower resistance range, four displayed a reduced growth rate cost without a change in resistance range, one showed a further increase in growth rate cost but no change in resistance range, one had no growth rate change but had a reduced resistance range, and four strains displayed neither a change in their cost of resistance nor in their resistance range (Fig. 2). These findings indicate that changes in these two phenotypes were not coupled in the evolved cultures. This is further supported by the finding that two of the evolved cultures (R7 and R8) developed an improved growth rate (at t1) before the reduction in their resistance range (Fig. 2B).

Initially, we were surprised that the manifestation of these two phenotypes did not appear to be coupled, especially because the initial acquisition of resistance was often coupled to a pleiotropic fitness cost (14). This prompted us to make use of clones that were plated 7 mo after initial isolation of the MED4 R7, R8, and R9 substrains (Fig. 1). This enabled us to assess whether (i) reproducible evolutionary trajectories were followed multiple times as these clones were undergoing evolution independently of each other and of the cultures they were derived from, or (ii) whether the process was somewhat random. We investigated over 20 evolved clones originating from each of the three substrains (for a total of 68 clones), tracking their parallel evolution. We placed emphasis on whether the improved growth rates and reduced resistant ranges were indeed noncoupled phenomena and, below, also investigated the mutations responsible for the different phenotypes.

We first report on the phenotype of these multiple independently evolving clones relative to the evolved cultures and then present pairwise phenotypic comparisons to assess the extent of the correlation between changes in growth rate and resistance range. Similar to that of their respective evolved cultures, practically all R7 and R8 clones had evolved an improved growth rate during the 27 mo since isolation of the initial resistant substrain (Fig. 4 B and C, and Tables S2 and S3). Despite the increase, growth rates remained below that of the wild-type MED4 strain in ∼70% of the cases. In contrast, the growth rate of most R9 clones remained the same as the initial resistant substrain, one showed an increase in growth rate, and only two displayed a similar decrease in growth rate as that found in the evolved R9 substrain culture (Fig. 4D and Table S4). In addition, 87% and 96% of the evolved clones from all three substrains displayed a reduced resistance range 27 and 33 mo after initial isolation, respectively (Fig. 4A and Tables S2–S4). The overall phenotype of a narrowing resistance range in the R7 and R8 clones was similar to that for the R7 and R8 evolved cultures, whereas in the R9 clones this contrasted with the unchanged resistance range in the R9 evolved culture (Fig. 2B).

Fig. 4.

Resistance range and its relationship to growth rate in evolved clones. A total of 68 evolved clones originating from R7 (n = 21), R8 (n = 23), and R9 (n = 24) resistant substrains were tested for resistance to three cyanophages: P-GSP1, P-TIP38, and P-SSP7. (A) Resistance ranges are presented as the percentage of clones resistant to all three (green), two (blue), one (purple), or no (black) phages at 27 and 33 mo after initial isolation. The 4-mo bar represents resistance ranges of the three initial resistant substrains, as reported in ref. 14. (B–D) Growth rates of evolved clones (Ev-cl), normalized to that of the susceptible controls, are shown as a function of resistance ranges tested at 27 mo. The number of evolved clones in each category is shown in parentheses on the x axis. Averages and SDs are for all evolved clones in a particular resistance range category. The wild type (WT) and initial resistant strains (iR) are shown for comparison. Statistically significant differences (P < 0.001, ANOVA or Kruskal–Wallis one-way ANOVA) are marked above the bars with the letters A, B, C, or D. The growth rate of each evolved clone was determined from three biological replicates except for three cases where n = 2. The resistance range was determined from two biological replicates for each clone. The iR values reported here were taken from ref. 14.

In most cases, the loss of resistance was partial, as the majority of clones (83%) remained resistant to at least one phage 33 mo after initial isolation (Fig. 4A). In the R7 and R8 clones, this was always to the phage used for initial isolation, whereas in the R9 clones it was sometimes to an additional or different phage. Furthermore, phages that could now infect these evolved clones led to a slower population decline than when infecting the susceptible wild type (Fig. S3A and Tables S2–S4). These phenomena of partial loss of resistance are similar to those we found for the R7 and R8 cultures. The remaining 16% of the clones (11 in total) were phenotypic revertants as they had lost resistance to all three phages and the dynamics of their infection was the same as for the susceptible wild-type host (Fig. S3B).

Pairwise comparisons of the growth rate and resistance range of these clones (measured at 27 mo) showed that improved growth rates and reduced resistance ranges occurred independently in 38% of the cases. This was found for 20 of the R9 clones, which had a reduced resistance range without a corresponding change in growth rate (Fig. 4D and Table S4) and for six R8 clones that had an increased growth rate without a parallel reduction in resistance range (green bars in Fig. 4C, and Table S3). Six months later, a reduced resistance range was observed for four of these five R8 clones (Table S3). It is interesting to note that the R8 clones with both an improved growth rate and reduced resistance range had, on average, a 13% greater growth rate than those with just an increased growth rate (compare purple and green bars in Fig. 4C), suggesting that some part of the increase in growth rate was coupled to the resistance phenotype. As such, independent and sequential events led to an increased growth rate and reduced resistance range in many cases and some degree of growth rate improvement appears to be linked to the reduced resistance range. However, there were many other cases for which we cannot ascertain whether phenotypic changes occurred independently or concurrently as they displayed changes in both phenotypes by the first time of measurement (Fig. 4 B and C).

Genotype of the Evolved MED4 Resistant Strains.

To determine the genetic basis for the above phenotypic changes, we sequenced the entire genomes of four evolved MED4 cultures (R1, R7, R8, and R9) and eight clones that had evolved from the R7, R8, and R9 resistant substrains, of which 55% of the mutations identified were experimentally verified by PCR. We further screened an additional 59 evolved clones by PCR for mutant genes that were identified by whole-genome sequencing, and tested susceptible control strains for mutations in these same genes. Overall, mutations were found in 34 different genes as well as in five intergenic regions (Tables S5 and S6). Fifteen of the genes (44%) were found in genomic islands (13 in ISL4), and 18 (53%) were noncore genes (Tables S5 and S6), indicating that mutations relating to host–phage interactions preferentially accumulated in noncore genes localized in genomic islands (3.2-fold enrichment for their position in genomic islands, P = 1.1E-05; and 1.6-fold enrichment for noncore versus core genes, P = 0.01), as did the mutations that conferred initial resistance to phage infection (14).

The evolved R7 and R8 cultures, which displayed both an improved growth rate and a narrower resistance range, had three mutations that differentiated them from their respective initial resistant ancestors. One of these mutations was at the original resistance locus in each evolved substrain (PMM0278 in R7 and PMM1209 in R8), both of which are predicted to be involved in cell surface biosynthesis. In each, a leucine was introduced in place of the stop codon found in the initial resistant substrain, thus reinstating full-length proteins, but with an amino acid different from the tryptophan found in the susceptible wild-type strains (Table S5 and Fig. 5). In addition, both evolved cultures had mutations in the PMM1124 gene, which codes for a predicted autotransporter (Tables S5 and S6 and Fig. 5D), homologs of which act as virulence factors (53–55). The third mutation in each of the two evolved cultures was in neighboring genes within the same gene cluster (PMM1200–PMM1205) that are predicted to be involved in cell surface biosynthesis: PMM1201 in R7 and PMM1200 in R8 (Fig. 5 and Tables S5 and S6). Mutations in none of these genes were found in the susceptible wild-type control strains. These findings provide clear evidence that additional mutations rather than genetic reversals were responsible for the evolved phenotypes and that these mutations were specific to the evolution of the resistance substrains.

Fig. 5.

Mutations in the most common mutant genes in evolved strains. (A) PMM1209, (B) PMM0278, (C) PMM0076, (D) PMM1124, and (E) the PMM1200–PMM1205 gene cluster. The PMM0278 and PMM0076 genes are homologs. Mutations are found in initial resistant substrains (red), evolved cultures (blue), evolved clones (green), and strains revived from frozen stocks (orange). Blue bars represent the coding region of the genes, and the vertical lines show the relative position of the mutations within the genes (not drawn to scale). Different types of mutations are shown by different shapes: stop codon or frame shift (Stop/fs, octagon), amino acid substitution (Subst, square), reversion to wild-type amino acid (Reverse, triangle), regain of frame through loss of stop codon or frameshift without reversion (Regain, right block arrow). Mutations upstream of the PMM0076 gene indicate nucleotide substitutions. See Tables S2–S5 for the exact mutations. The number of different clones displaying the same mutation type at the same locus is shown by X#. The iR data reported here were taken from ref. 14.

The presence of multiple mutations in the evolved cultures raised the question as to which mutation or combination of mutations is responsible for each of the different phenotypes observed. Genetic analysis of evolved R8 clones allowed us to address this question because the clones displayed a separation between the two phenotypes: although all clones had acquired an improved growth rate, not all had a reduced resistance range. PCR screening of numerous R8 clones revealed a lack of correlation between mutations in the PMM1124 and PMM1200 genes and resistance range, as even clones with no change in resistance had mutations in these latter two genes (Table S3). In contrast, there was a highly significant degree of correlation between mutations in the PMM1209 gene and the reduction in resistance range (88.5% correlation; P < 0.001, Fisher’s exact test): first, only clones with a change in resistance range had replaced the resistance stop codon with one of three other amino acids (eight with leucine as found for the evolved culture, and three each with serine or cysteine). These clones were infected by the same two phages irrespective of which amino acid restored the protein to full length. However, they caused a slower population decline than when they infected the wild-type strain (Table S3 and Fig. S3A). Second, six clones with no change in resistance range had retained the stop codon. Third, two clones that had lost resistance to all three phages had reverted to the wild-type tryptophan. However, this correlation was not absolute, as two clones with the wild-type tryptophan maintained resistance to one of the phages and three other clones that retained the initial resistance stop codon had a reduced resistance range (Table S3). Whole-genome sequencing of one of the latter clones revealed additional mutations (Table S5) that could well be responsible for the reduced resistance range. These data strongly suggest that PMM1209 is the gene responsible for the reduced resistance range in most R8 clones, but that mutations in other genes may also have led to changes in resistance.

The R7 clones are less suited for determining which mutation is responsible for changes in resistance range because all clones displayed both phenotypes. Nonetheless, we wanted to assess whether mutations in the original resistance gene, PMM0278, could be impacting resistance range. Different from our initial expectation, a PCR screen of R7 clones with reduced resistance ranges revealed that only 5 of the 21 clones had a different mutation in this gene: the original resistance stop codon was exchanged with either a tyrosine or glutamine (Fig. 5 and Table S2). Another five clones had reverted to the wild-type tryptophan at this locus, four of which were also phenotypic revertants (Table S2). The 11 remaining clones, however, retained the initial resistance stop codon. Whole-genome sequencing of two of the latter clones revealed a mutation in or upstream of the PMM0076 gene, which is a homolog of PMM0278. Both are core genes and have 24% amino acid identity to each other. Further PCR screening revealed that all 11 clones had mutations in or upstream of the PMM0076 gene. In fact, all evolved R7 clones had a mutation in one but not both of these two homologous genes compared with the initial resistant substrain. These data are consistent with the possibility that mutations in either the original resistance gene or its homolog are responsible for the observed changes in resistance range in the R7 clones.

Next, we wanted to assess whether mutations in the PMM1124 gene and/or a gene in the PMM1200–PMM1205 gene cluster might be responsible for the improved growth rate in the evolved cultures and clones. Indeed, whole-genome sequences of five clones and PCR screening of another 17 clones, revealed that all R7 and R8 clones had a mutation in the PMM1124 gene (Tables S2 and S3). Furthermore, all but one clone had an additional mutation in a gene within the PMM1200–PMM1205 gene cluster, the majority of which were in PMM1200 (Fig. 5 and Tables S2 and S3). It should be noted that the three evolved control strains (i.e., that had evolved from the susceptible wild-type control) had no mutations in these two sets of genes. Therefore, it is quite feasible that a mutation in either of these sets of genes, or a combination of these mutations, is responsible for the improved growth rate.

A close look at the specific mutations showed that all R8 clones had the exact same mutation as the R8 culture in the PMM1124 gene, a stop codon in position 201 instead of a tryptophan (Fig. 5 and Table S3), suggesting that this mutation was fixed in the culture before the plating of the clones. The R7 clones, on the other hand, displayed a variety of mutations in this gene that were also different from that in R8 (Fig. 5). However, all R7 mutations also introduced either a stop codon or a frameshift, most likely leading to a loss of function of the PMM1124 protein. The mutations in the PMM1200–PMM1205 gene cluster were different among both the R7 and R8 clones, and caused a mix of amino acid substitutions as well as a number of premature stops and frameshifts. These findings indicate that multiple independent mutations led to parallel evolution in these clones, whether they occurred before plating but had not become fixed in the population or occurred after plating.

It should be noted that 16% of the evolved clones that were derived from the R7, R8, and R9 resistant cultures and displayed the same phenotype as wild-type strains with respect to growth rate and susceptibility were genetic revertants (Tables S2–S4 and Fig. S3), similar to our previous findings for R5 (14). Thus, although continued evolution included reversion to previous genotypes, this was at a relatively low level, with new mutations and intermediate phenotypes being far more common.

Genetic investigation of the R1 and R9 evolved cultures were less informative due to the presence of a high number of mutations. The R1 strain, which lost the enhanced infection cost and had a reduced resistance range, retained one of its original mutations, lost the second, and gained five more mutations. Three of these mutations were in genes of unknown function, one was in an amino acid carboxylase and the last was in the same PMM1124 autotransporter gene found for the R7 and R8 evolved cultures (Tables S5 and S6).

The evolved R9 culture had by far the highest number of mutations in any of the evolved cultures or clones, with a total of 18 mutations (Table S5). One of these was in the mutS gene, which is part of the DNA mismatch repair machinery. Mutations in this gene are known to lead to a high frequency of mutations in other bacteria (56–58). This mutator strain lost both of the mutations found in the initial resistant strain yet maintained resistance to all five phages, presumably due to the other mutations acquired, and even though only the phage used for initial isolation was present in the culture. The loss of one of the initial resistance mutations, that in the predicted PMM1124 autotransporter, was most intriguing, because the growth rate of this evolved culture declined relative to the initial resistant substrain. In fact, the initial resistant R9 substrain may have had a higher growth rate than other initial resistant substrains due to the early acquisition of a mutation in the PMM1124 gene. However, with so many other mutations, we cannot determine whether this is simply coincidence or whether the restoration of this gene was responsible for the decline in growth rate.

The three evolved R9 clones that were fully sequenced had mutations in only two to three genes relative to the initial resistant substrain. This included a mutation either in the original resistance gene (PMM1259) or a gene in the same gene cluster (PMM1257), which may be responsible for their reduced resistance range (Table S5). Interestingly, all retained the original mutation in PMM1124 and also gained a mutation in the PMM1200–PMM1205 gene cluster. Different from the evolved culture, two of these clones displayed a growth rate improvement (Table S5). These clear differences in mutations in the evolved R9 culture relative to the evolved R9 clones explains the large differences in phenotype observed.

Genotype of an Evolved MIT9312 Resistant Strain.

The evolved culture of MIT9312_R70 (R70), which displayed an improved growth rate but maintained its resistance range (Fig. 2C), retained the mutation found in the original resistant substrain, that in the pstA phosphate transport gene. Three additional mutations were also found, all in noncore genes in close proximity to each other in ISL4 of the genome (Tables S5 and S6). Two of these genes are predicted to be involved in cell wall biosynthesis, and the third codes for an ABC transporter with potential export and substrate-processing activities (Table S6). Clearly, we cannot tell which mutation, or combination of mutations, are involved in the improved growth rate phenotype of this evolved culture. Nonetheless, we consider the mutation in the putative CMP-N-acetylneuraminic acid synthetase gene to be a candidate based on the fact that this gene is a homolog of two MIT9215 genes, mutations in which were identified previously to have led to phage resistance as well as a reduced growth rate (14). Similar to the autotransporter in MED4, it is also possible that the mutation in the ABC export transporter caused an increase in growth rate (Discussion).

Discussion

Here, we report on convergent evolution toward an improved growth rate and a reduced resistance range among phage-resistant Prochlorococcus substrains. This improved growth rate was observed for all but one of the evolved cultures that had an initial growth rate cost, whereas the reduced resistance range was apparent for about a quarter of them (Fig. 2 B and C). The similarity of the phenomena among the different independent cultures, and clones derived thereof, suggests that the improvement in growth rate is a general and predictable evolutionary trajectory for resistant substrains that incurred a growth rate cost initially. Furthermore, the largely parallel evolutionary route adhered to by multiple clones derived from the same evolved culture suggests a nonrandom and an almost inevitable evolutionary pathway for a particular resistant strain. Assuming that our findings are indicative of the processes occurring in nature, such a predictable pattern of evolution would lead to Prochlorococcus populations growing at, or close to, their intrinsic maximum, but with constantly shifting ranges in resistance to co-occurring phages.

Based on our findings, we propose a model of this process, whereby continued evolution subsequent to the initial acquisition of resistance occurs in two sequential steps when an initial growth rate cost to resistance was incurred (Fig. 6). The first step comprises the selection for a mutation that leads to a significant increase in growth rate. The second step involves the emergence of cells with a reduction in resistance range that may or may not also be accompanied by a further increase in growth rate. The reproducibility of the phenomenon argues against this latter step being due to genetic drift, even in cases where no significant increase in growth rate was observed. The most plausible explanation is that an improvement in growth rate occurred in these latter cases as well but was too small to be detected in our assays, or that some other unknown associated competitive advantage was obtained. The improved growth rate likely occurs before the reduced resistance range due to the greater competitive advantage conferred by the former. However, it is also feasible that the first mutation primes the cell for the second mutation, as was found for evolution of phage lambda (59). Unfortunately, the lack of a genetic system for Prochlorococcus prevents us from introducing the different mutations in a controlled manner to test this latter hypothesis.

Fig. 6.

Model of genotypic and phenotypic evolutionary steps. Growth rate (GR) and resistance range (RR) of the R7 and R8 evolved cultures are shown relative to the susceptible wild-type control (WT) and the initial resistant (iR) strains after the first (Ev1) and second (Ev2) proposed evolutionary steps. Mutations likely causing these phenotypic changes are shown in the Middle. The additional improvement in growth rate associated with R8 in Ev2 was not detected for R7.

The continued evolution of resistant Prochlorococcus clones, in the vast majority of cases, yielded cell lineages that were phenotype intermediates between that of the wild-type susceptible ancestor and the initial resistant substrain. This intermediate phenotype is expressed in the various strains in one or more ways: an increase in growth rate that remained below that of the susceptible wild-type strain; resistance that was retained against some, but not all phages; and clones that had lost resistance to some phages but whose population decline was slower when infected by these phages than when they infect the susceptible wild-type strain. This lack of complete reversal to the susceptible phenotype serves to create new combinations of infectivity.

These intermediate phenotypes resulted from multiple additional mutations in the original resistance genes as well as in additional genes, leading to further genetic diversification. The fact that most of these additional mutations were located in genomic islands, provide further evidence that these genomic regions and their diversification play a central role in host–phage interactions and coevolution, and are in agreement with previous findings (14, 40, 60, 61). Furthermore, the growth rate-related genes in both MED4 and MIT9312 were in noncore genes, providing further evidence (14, 62) that important cellular functions that impact cell fitness are carried out by noncore genes localized in genomic islands.

A single mutation in the initial resistance strain often led to pleiotropic effects: both resistance and an associated cost (14). In contrast, the change back toward the original phenotype often occurred in individual steps and was caused by different mutations. This decoupling of phenotypic changes would allow selective forces to work separately on growth rate and resistance. Thus, improvements in growth rate can occur even in the presence of strong phage pressure, such that slow-growing cells are not necessarily maintained in the population to retain resistance to co-occurring phages. In addition, the degree to which loss of resistance will occur in nature, will depend on the level of phage selection pressure present in the specific environment inhabited and, over time, will be less subjected to negative selection due to lower growth rates.

Compensatory mutations that ameliorate the growth rate cost (this study), together with resistance mutations that have no initial associated growth rate cost (14), likely lead to rapidly growing Prochlorococcus-resistant strains in nature that outcompete slower growing resistant strains. As such, rapidly growing resistant cells will come to dominate the population, and slow-growing resistant strains will be lost. These phenomena explain the apparent paradox of a multitude of presumably resistant Prochlorococcus cells in nature (14, 30) that are growing at or close to their maximal known growth rates (50–52).

A comparison of our findings for this oligotrophic marine host–phage system to enteric, nutrient-rich Escherichia coli systems, shows a number of similarities. First, Lenski (63) and Kashiwagi and Yomo (64) found that the growth rate of several phage-resistant E. coli strains increased over time without a corresponding change in resistance range. In addition, the mutations responsible for such increases in growth rate were at a different locus from that found for the original resistance mutation (63, 64). A different study of long-term evolution in phage-resistant E. coli reported an increase in susceptibility to a single phage over time, as well as a parallel pattern of evolution in multiple E. coli lineages (65). Therefore, decoupling of these two phenotypes, the resultant increase in genetic diversity during evolution subsequent to the acquisition of resistance, and the predictable pathway followed by multiple lineages, suggests that these phenomena are common in host–phage coevolution.

Parallels can also be drawn between phage resistance and resistance to antibiotics. First, resistant strains with clear growth rate costs and those with no observable growth rate costs are often found (14, 25, 27, 66, 67). Second, a mutation that confers resistance to one phage or antibiotic can lead to hypersensitivity to other phages or antibiotics (14, 15, 27, 28, 68). Third, just as a different mutation in the same locus can lead to different rates of infection (this study), different mutations can lead to variable levels of resistance to the same antibiotic (69). Fourth, compensatory mutations are found over time that ameliorate the growth rate cost of both phage and antibiotic resistance, and these are often in genes other than the ones that conferred the resistance (refs. 63, 64, 70, and 71; this study). Last, similar to phage resistance (refs. 63 and 64; this study), isolation of cells after continued evolution from the human ecosystem (human patient), shows that the growth rate cost of antibiotic resistance is ameliorated while resistance itself is maintained (70, 71). These parallels suggest common themes in the evolutionary process, not only for resistance to phage infection of diverse bacterial types inhabiting vastly different environments, but also across selective forces leading to resistance.

Mutations in similar gene types are likely to be responsible for the improvement in growth rate in both the MED4 and MIT9312 evolved strains. Such mutations in both cell surface biosynthesis and transporter genes were repeatedly found. However, at this stage, we cannot ascertain whether mutations in both gene types are required or whether a mutation in just one of these genes is sufficient to cause the improved growth rate phenotype.

The mutations in the transporter genes are particularly thought-provoking in the context of a potential improvement in growth. The MED4 gene, PMM1124, and the MIT9312 gene, PMT9312_1332, belong to two different families of transporters that secrete virulence factors and toxins in other bacteria (53–55, 72), some of which negatively impact the growth of their eukaryotic hosts (53–55). How then may mutations in such proteins lead to a growth rate improvement in Prochlorococcus? One feasible explanation is that the lack of transport of the substrate provides an energetic saving to the cell. However, an improved growth rate of up to 20% seems enormous for such an energetic saving. An alternative hypothesis is that, when the transporter is functioning, the secretion of the substrate negatively impacts Prochlorococcus growth and that mutations preventing their secretion enable an increase in the growth rate. Further research into the function of these transporters and the impact of their substrates on Prochlorococcus is required to ascertain the extent to which such conjectures reflect their mode of action and whether they indeed serve as negative regulators of growth.

The narrowing of resistance ranges observed here during the continued evolution of resistant Prochlorococcus strains provides an additional means to that of enhanced infection for the emergence of strains with altered resistance ranges. These altered resistance ranges emphasize the nondirectional nature of the battle between hosts and viruses (30). Indeed, our current observations provide a further explanation for why directional arms race dynamics are unlikely to explain long-term coexistence in nature.

In a recent theoretical study, Thingstad et al. (73) suggested that numerical abundance of certain bacterial species present in the oceans with their phages, may be related to subtle differences in defenses at the molecular level among strains within that species that do not lead to large competitive trade-offs. Our findings provide empirical support for their model: first, some resistant strains display no cost of resistance (14, 25, 27, 66). Second, although emerging resistant strains often exhibit a large cost and broad range of resistance initially (14), additional mutations occur over time that serve to reduce the cost and narrow the range of resistance such that the trade-off differential diminishes (this study). As such, the relative contribution of fluctuations between competition and defense specialists, and their amplitude, likely declines with time after the emergence of resistant host strains. Furthermore, fluctuations between hosts with different susceptibility ranges (30) may well increase in the community. Thus, it appears that large Prochlorococcus populations belonging to the same ecotype coexist in the oceans with their phages as these populations are made up of diverse Prochlorococcus subpopulations with differences in cell surface-related genes and thus susceptibility to co-occurring viruses (14, 40), and that, in turn, the coexistence of these diverse Prochlorococcus subpopulations is enabled due to only marginal differences in their competitive trade-offs (ref. 73; this study).

The continual evolution of resistant Prochlorococcus strains reported here generates new lineages with growth rate and resistance range phenotypes that are more similar to the ancestral susceptible strain than the initial resistant substrain. This process, however, is driven by the accumulation of additional mutations, such that these same strains become genetically more distant from the ancestral susceptible strain. Thus, genetic diversity continues to grow in the population, creating a never-ending increase in the combinatorial interactions among closely related hosts and their viruses. This constantly shifting interaction landscape has likely resulted in a highly branched and dynamic network of interactions between Prochlorococcus and its phages that will keep on changing as time goes on.

Methods

Strains and Culturing Conditions.

Resistant Prochlorococcus substrains used in this study (Table S1) were initially selected for resistance to one of the following T7-like cyanopodoviruses: P-SSP7, P-TIP2, or P-GSP1 for MED4-derived strains, P-SSP2 for MIT9312-derived strains, and P-SSP1 or P-RSP1 for MIT9215-derived strains, as described by Avrani et al. (14). These and the following additional T7-like cyanopodoviruses were used in the present study to assess resistance range and host population decline: P-TIP1 and P-TIP38 for MED4 strains, P-SSP3b for MIT9312-derived strains, and P-TIP39 for MIT9215-derived strains (14, 74). The susceptible control cultures used in this study were isolated at the same time as the initial resistance substrains and were derived from the same ancestral colonies as their respective resistant substrains (Fig. 1). Three such control cultures were isolated from MED4: one for R7 and R8, another for R9, and a third for R1 to R6 (Table S1). Single susceptible controls were isolated for each of the MIT9312 and MIT9215 resistant cultures used in this study (Table S1).

Prochlorococcus strains derived from MED4, MIT9312, and MIT9215 were grown in the seawater-based medium Pro99 (75) at 21 °C under a regime of 14-h light and 10-h darkness with 7.5–10 μmol photons⋅m⁻²⋅s⁻¹ during the light period. Growth of cultures in liquid medium was followed using in vivo chlorophyll a autofluorescence as a proxy for biomass in arbitrary fluorescence units (AU) using a Synergy Mx plate reader (BioTek) with excitation and emission wavelengths of 440 ± 20 and 680 ± 20 nm, respectively, or a 10-AU fluorometer (Turner Designs) with excitation and emission wavelengths of 340–500 and >665 nm, respectively. The growth rates of the different susceptible wild-type control strains ranged from 0.5 to 0.75 doublings per day under these growth conditions (Fig. S2). Examples of growth curves can be seen in Fig. 3 and Fig. S3 for the uninfected cultures and clones, respectively.

Growth of Prochlorococcus on plates to produce lawns or single colonies was done by pour plating using Invitrogen Ultra-Pure low–melting-point agarose at a final concentration of 0.28% containing Pro99 medium (45, 75). A heterotrophic helper strain, Alteromonas sp. strain EZ55, was added to the cells to increase plating efficiency to 100% (76).

Derivation of Evolved Strains.

Two types of evolved strains are referred to in this study: evolved cultures and evolved clones, all of which were derived from the initial resistant substrains isolated by Avrani et al. (14). The evolved cultures were serially transferred for up to 40 mo (∼750 generations). Serial transfers were carried out every 2 wk into fresh medium so that the inoculum contained ∼10⁶ cells per mL. Prochlorococcus strains reach ∼2 × 10⁸ cells per mL toward the end of the exponential phase of growth. Initially, the selecting phage used for initial isolation was added to the cultures (10⁵ phage per mL) at each transfer, but this was stopped at least 23 mo before testing growth rate and resistance range at the 24- or 40-mo time point (as per Fig. S1). Some cultures continued to produce phages for a time, but phages became extinct in six of the seven cultures tested by the time of assessment. The tested cultures were as follows: R1, R5, R6, R7, R8, R9, and R69. Only the R9 culture continued to produce phages throughout the experimental period. The evolved clones were produced 7 mo after the initial isolation of the resistant substrains by plating a portion of three of the evolved cultures (R7, R8, and R9) in pour plates. Over 20 colonies from each resistant strain (21, 23, and 24 clones from R7, R8, and R9 cultures, respectively) were transferred into fresh liquid medium and were grown in separate vessels for an additional 26 mo.

Revived Strains.

Initial resistant substrains were revived from cryopreservation by transferring the contents of the frozen tube of concentrated cells to fresh medium in low light (45). These strains were used to compare infection dynamics of the evolved cultures and clones to the initial resistant substrains to assess the enhanced infection cost. Revived strains with a growth rate cost either failed to grow (e.g., R8), or appeared to go through a bottleneck when they did grow, as both their phenotype and genotype changed (Table S5; R7 def and R9 def). Therefore, revived initial resistant substrains that had a growth rate cost could not be used in this study.

Phenotypic Characterization of the Evolved Strains.

Growth rates of evolved cultures and clones (three to six biological replicates) were compared with those of the appropriate paired susceptible control substrain grown in separate vessels under conditions described above. Growth rate results of the evolved strains are presented as a percentage of the growth rate of the wild-type controls measured at the same point in time. Exponentially growing cells were used after they had been growing steadily at the same growth rate for a minimum of 10 generations. Note that the evolved cultures and susceptible control wild-type substrains were grown under identical conditions throughout the period of the study.

Phage challenge tests were performed to determine the resistance or susceptibility of mutant substrains to a set of podoviruses. Exponentially growing cells were challenged with phages at a multiplicity of infection (MOI) of 0.1–1 (ratio of phage to host) in liquid medium. The paired susceptible control strains were used to verify phage infectivity. When an evolved strain was susceptible to one of the phages, the rate of population decline was compared with that of the paired susceptible control when challenged with the same phage using the same concentrations of host and phage and thus maintaining the same MOI.

It should be noted that the revived initial resistant R1 strain that displayed an enhanced infection cost when infected with P-GSP1 and P-TIP38 (Fig. 3A) is different from our previous report for R1 and similar to that for strains R2–R6 (14). It thus appears that the R1 resistant strain had already evolved and lost the enhanced infection cost by the time of those previous analyses.

High-Throughput Sequencing and Mutation Calling.

Whole-genome sequencing was carried our for five evolved cultures (R1, R7, R8, R9, and R70), eight evolved clones derived from R7, R8, and R9 (two, three, and three clones, respectively), and two resistant cultures revived from frozen stocks (R7 and R9). Genomic DNA was extracted using the DNeasy Blood and Tissue Kit (Qiagen). Complete genomes of the evolved cultures and clones were sequenced using the Illumina HiSeq 2000 genome analyzer using the manufacturer’s protocols at the genome sequencing unit at the Technion Genome Center. The sequence quality per base was evaluated with FASTQC, version 0.10.0. All bases used were with quality higher than a phred score of 30. Depth of coverage ranged from an average of 119-fold to 836-fold for the different strains, with read lengths of 50 nt. The whole-genome sequence reads of the 15 cultures and clones have been submitted to the Sequence Read Archive of the National Center for Biotechnology Information (NCBI) under project accession no. SRP054991.

The reference genomes of Prochlorococcus strains MED4 (CCMP1986) and MIT9312 (GenBank accession numbers NC_005072 and NC_008817) were downloaded from NCBI. Unique reads of each isolate were mapped to their corresponding reference genome using the Burrows–Wheeler Alignment tool, BWA 0.6.1 (77), using two parameter sets: The first set allowed up to three mismatches per 50-base read and one gap as long as it was not within the last five bases of the read. The second set, allowed up to two mismatches within a 32-base seed sequence. This latter set allows mapping of reads that contain large indels. In both cases, only unique mappings were used in this analysis. Samtools 0.1.18 (78) was then used to call mutations when a locus had coverage of at least five reads with a frequency of difference of at least 80% of these reads (14). Samtools parameters were set as follows: –L and –d to 10,000 and –E to recalculate extended base alignment quality computation. Mutations in the evolved strains were considered to be those changes not present in the either the susceptible wild-type control strains (Tables S7 and S8) or the initial resistant substrains (Table S5), as reported by Avrani et al. (14).

Although the lower bound of coverage for mutation calling was set at 5 reads, there were only two mutations with coverages of less than 50 reads (19 and 47 reads each). Fifty-five percent of all called mutations were experimentally tested by PCR (see section below). In all cases, the mutations were verified, including the two mutations with coverage of less than 50 reads. Furthermore, the genotype of the control susceptible cultures was assessed for all of the newly identified mutant genes discussed in the manuscript. None of the mutations were present in any of the control cultures.

It should be noted that a genetic system is not available yet for Prochlorococcus. Thus, it was not possible to experimentally verify the phenotype of any of the observed mutations.

PCR Verification and Screening of Mutations.

Fifty-five percent of the mutations identified by genome sequencing were verified by PCR amplification and direct Sanger sequencing of four pooled PCR amplicons. The identification of the mutations in strains whose genomes were fully sequenced was used to select genes for subsequent PCR screening in other evolved strains and in the susceptible control strains. In these cases, the entire gene was amplified and sequenced, because various mutations were found in these genes. This screening was achieved by PCR and Sanger sequencing of four pooled PCR amplicons. Primers used for PCR amplification and sequencing appear in Table S9. PCR amplification was done as described by Avrani et al. (14).

Functional Gene Annotation.

Functional annotations were initially based on information from GenBank or Cyanobase (genome.microbedb.jp/cyanobase) and Avrani et al. (14). These functional predictions were augmented by protein family (Pfam) information (79) and by information from the clusters of orthologous genes database (80) where possible.

Gene Classification.

Genes were classified as core genes when orthologous genes were found in all 13 sequenced Prochlorococcus strains as defined in ProPortal (81). All others were considered noncore genes. Genes were also classified as being located in syntenic or island regions of the genome based on the genomic island borders described by Avrani et al. (14).

Statistical Analyses.

T tests for independent samples were carried out to test for significant differences between initial resistant and evolved strains for growth rates (two-tailed); the data were normally distributed, as determined from P > 0.05 using the Kolmogorov–Smirnov or Shapiro–Wilk tests. For R7 and R8 strains that had data from three time points, we used the one-way ANOVA and Tukey and the Bonferroni post hoc tests to test whether the differences between the time points were significant. Similarly, t tests were also carried out to test for significant differences in the growth rate of control strains at the beginning and end of our experiment. A repeated-measure ANOVA was used to assess the significance of differences in population decline. One-way ANOVA was carried out to test for significant differences between the growth rates of initial resistant substrains and evolved clones with different ranges of resistance when the data were normally distributed (R8 and R9 strains). The Tukey and Bonferroni post hoc tests allowed us to determine the mean values that were significantly different from each other. In one case, the data were not normally distributed (R7), and therefore the nonparametric Kruskal–Wallis one-way ANOVA test was used instead. In this case, because no suitable post hoc test is available, the nonparametric Mann–Whitney test was used to distinguish between the three groups (between two each time). Fisher’s exact test was used to correlate between the mutations in PMM1209 and the reduction in resistance range in R8 clones. Enrichment for mutant genes in genomic islands or in noncore genes was tested by calculating the hypergeometric probability for finding this number of genes in genomic islands or in noncore genes at random, relative to the number of these genes in the MED4 genome. Enrichment for mutations in genomic islands was also assessed from the perspective of the number of nucleotides in these regions relative to the entire genome length. The same results were obtained and the results for enrichment of mutant genes are presented.

The PASW Statistics 17 package was used for these analyses (release 17.0.3, September 2009; SPSS). In the Mann–Whitney test applied on the growth rates, the α was calculated according to Bonferroni adjustment and was reduced to 0.025. This correction did not change the significance of any of the growth differences between resistant and control substrains.