Ancestral presence of capsule drives phenotypic evolution
To analyse how the capsule affects phenotypic change, we measured an adaptive trait (population yield after 24 hours), and two traits associated to virulence (biofilm formation and capsule production) in all end-point populations in their respective evolutionary environments. We also measured the hypermucoidy phenotype, HMP, commonly associated with hypervirulence [34, 35]. This was measured in M02 for all populations, as the growth media strongly alters HMP measurements. Due to experimental constraints, CFU could not be analysed for populations with high HMP, mostly capsulated populations in LB and ASM. Biofilms and capsule production could not be measured in soil (below the limit of detection).
For each of the four traits, we tested the direction and magnitude of change of each population, relative to its ancestor (Table S1). The changes observed in all four traits were dependent on the ancestral strain and capsule genotype (Multifactorial ANOVA, P<0.001) (Table S2). Biofilm production and CFU were also dependent on the environment, but changes in capsule production or HMP were not. Most interestingly, all four traits were also dependent on the presence/absence of the capsule. Out of the 41possible trait-environment comparisons between capsulated and non-capsulated populations, populations evolved in opposite directions 16 times (~40%), of which 12 significantly so (Table S1), suggesting that there is a significant effect of the capsule in driving the direction of phenotypic evolution. We analysed this further in the two environments for which all traits could be measured, namely AUM and M02 (Figure 1). We observed that non-capsulated populations, but not those descending from capsulated ancestors, have increased yield (CFU), production of surface-associated polysaccharides and biofilm formation. In capsulated populations, biofilm formation is reduced and capsule production either remains near ancestral levels or is strongly diminished.
We then analysed whether adaptive changes and traits associated with virulence may coevolve in opposite directions in capsulated compared to non-capsulated genotypes (Figure 1). Indeed, in capsulated genotypes, biofilm formation and capsule production show a strong negative correlation, but in non-capsulated populations, increased biofilm positively correlates with increased production of surface polysaccharides. Similarly, in populations derived from capsulated ancestors, higher levels of capsule production positively correlate with increased hypermucoidy, whereas, as expected, no correlation is observed between increase surface-associated polysaccharides and HMP in non-capsulated populations (Figure 1).
Taken together, our results show that the presence of capsule in adapting populations strongly shapes the direction of phenotypic evolution, the coevolution of traits and suggests that capsulated and non-capsulated populations may be adapting to structured environments by different mechanisms.
The capsule shapes genotypic changes in evolving populations
To determine the genetic basis of phenotypic adaptation, we sequenced one randomly chosen clone per population. On average, we observed four mutations per clone, with significant differences in the number across strains and environments (multifactorial ANOVA, Strain, Environment P< 0.001) (Figure S1A). The presence of the capsule did not affect the total number of mutations per clone nor the number of different genes mutated.
We assigned a COG process category (Clusters of Orthologous Group) to each gene family in the ancestral genomes [41] and tested whether some cellular processes were preferentially targeted by mutations relative to the whole genome (Figure 2A). Patterns are similar in clones descending from capsulated and non-capsulated ancestors, but we observed opposite trends in defence mechanisms (V). Specifically, mutations in this group are over-represented in clones from capsulated background, but are under-represented in non-capsulated clones. Direct comparison in the number of mutations within each COG group between capsulated and non-capsulated clones showed that the latter accumulate significantly more mutations in genes associated to cell motility (N, Fisher’s test P<0.001) than the capsulated clones. This is mostly driven by mutations found in the mrkABCDF operon, which codes the type 3 fimbriae [42]. These mutations accumulate significantly more in non-capsulated populations (Chi-squared, P < 0.001). Capsulated clones are enriched in mutations involved in transcription (K, Fisher’s test P = 0.05) and cell wall, membrane and envelope biogenesis (M, Fisher’s test P = 0.005).
To precisely pinpoint the impact of the ancestral genetic background on mutations at the gene level, we established the pangenome of the three ancestral strains. There was very little overlap between the mutated genes in clones descending from capsulated and non-capsulated ancestors (Figure 2B). Dissimilarity tests revealed that there are few mutations common to capsulated and non-capsulated clones descending from the same genetic background, comparable to the differences across strains (Table S3, Figure S1B). We however identified two operons that were consistently mutated in all strains: the abovementioned mrkABCDF operon and the capsule operon (Figure S1C). Additionally, we also found numerous mutations in known capsule regulators. To systematically analyse capsule-related mutations (capsule operon and regulators), we compiled a list of 143 genes identified by mutagenesis to affect capsule production (either up or downregulation, Dataset S1, see Methods) [43, 44] and checked whether mutations occurred in genes (or neighbouring intergenic regions) homologous to known capsule-related genes [43, 44]. Mutations in capsule-related genes account for 19% of the total, yet capsule-related genes represent only ~1.9% of the genome (as calculated by the total nucleotidic length of all of such capsule-related genes in the reference genome of Kpn NTUH) [43, 44]. This indicates that they are major targets of selection, in all evolutionary treatments as they are extremely over-represented, (P < 0.001 for deviation from expectation of 18 mutations under a null assumption of random distribution of the 673 mutation events identified in the evolved clones, two-tailed binomial test). Interestingly, clones descending from capsulated ancestors mutated preferentially within the capsule operon, as already suggested by the COG analyses, whereas clones descending from non-capsulated ancestors had mutations in genes regulating capsule production (Fisher’s test, P<0.001) (Figure 2C).
To test the effect of mutations in capsule regulatory genes, we restored capsule expression by wcaJ complementation in trans and quantified capsule production of several clones in the environments in which they evolved. We tested several deletions in the hypervirulent plasmid of NTUH all of which resulted in the absence of a gene annotated as rcsA_3, a potential homolog of rmpA (40% identity). These mutations did not affect capsule production. We hypothesized that it could affect HMP, but this could not be tested as capsule complementation did not restore HMP (data not shown). This could be explained by the tight multifactorial regulation of HMP [45]. However, the mutations in capsule regulators found in the genome, all significantly reduced the amount of capsule produced (Figure 2D).
Our results show that the presence of a capsule strongly affects the first steps of adaptation to novel environments. Capsulated clones mutate preferentially in genes affecting directly capsule synthesis whereas, in non-capsulated clones, regulators are often mutated as to diminish, but not fully abolish, capsule production.
Capsule inactivation emerges readily but rarely fix in structured environments
In the KlebEvoI experiment we showed that in well-mixed environments capsule inactivation was very common when nutrients were abundant [30]. To test whether the same trend is observed, we followed the emergence of non-capsulated clones throughout the KlebEvoII experiment. We plated all populations descending from capsulated ancestors at regular intervals and visually examined the proportion of capsulated clones (see Methods, Figure S2). In ASM and LB, not all time points for all populations could be examined due to a very persistent HMP, which precluded dilution plating. We observed the emergence of non-capsulated clones in 61 independent populations (75%), but the dynamics differed across strains and environments (Figure S2). In a majority of the populations, non-capsulated clones emerged at very low frequencies and were rapidly outcompeted. Indeed, at the end of the experiment, 67 populations out of 81 descending from a capsulated ancestor were still dominated by capsulated clones, and in 53 (65%) of them, all sampled individuals remained capsulated (Figure 3). These populations were found in all environments and in all strains, and capsule maintenance was independent of both strain and environment (Two-way ANOVA, P> 0.05). In only 7 out of 81 populations (9%), the capsulated clones were driven to extinction, contrasting with our findings from KlebEvoI obtained in well-mixed nutrient-rich environments [30].
The genetic basis of capsule inactivation in KlebEvoII followed previously observed patterns, including KlebEvoI, namely mutations in wcaJ, the first gene of the biosynthetic pathway [31, 46]. From the 81 randomly-chosen clones descending from a capsulated ancestor, many of which had mutations in the capsule operon, only 12 were non-capsulated. This was a result of IS insertions or mutations resulting in changes in the reading frame or premature stop codon in wcaJ, wzc or rfaH (Table S4), genes essential for capsule synthesis [47, 48].
The results of KlebEvoII suggest that capsule mutants emerge easily, but rarely fix under these conditions, suggesting that they are either outcompeted or their capsule restored by recombination with other clones in the population.
The mutational mechanisms are strain-specific
In the long KlebEvoII experiment, non-capsulated clones of strain Kva 342 emerged primarily by IS insertions, small mobile genetic elements (~0.7-2.5 kb) that can vary in type and copy number across genomes, whereas those of Kpn BJ1 and Kpn NTUH resulted from SNPs or single deletion or insertion of a base-pair in wcaJ (Table S4). To further extend our analyses of the genetic mechanisms by which the capsule production is interrupted and test if KlebEvoII is representative of the larger diversity of strains, we analyzed 73 non-capsulated clones descending from 16 different capsulated strains (including the three focal strains of this study) from KlebEvoI, in which populations were propagated in well-mixed environments [30], and only partially analysed in [31]. Almost half (44%) of capsule inactivations were due to an IS insertion event (Figure 4A), but this was strain-specific (Figure S3A). Some strains, including Kva 342, mutate very frequently (or exclusively) by IS insertions, whilst others seldomly. There was no correlation between the total number of IS in a strain and frequency of IS-dependent capsule inactivation (Figure S3B). Capsules were mostly interrupted by IS903 from the IS5 family, which uses a replicative mechanism of transposition. Two other capsules were interrupted by IS from the IS91 and IS3 family, in multidrug resistant NJ ST258 and Kpn ST2435, respectively. The presence and number of IS of these three families in the genome significantly correlated with the frequency of capsule inactivation by such elements (Spearman’s rho = 0.85, P < 0.001).
To analyse if similar mutational trends were found outside the capsule operon, we analysed all mutational events in the 160 end-point clones of the KlebEvoII experiment. Non-synonymous point mutations (N) were common and their frequency was constant across strains and environments (36% on average) (Figure 4B). The ratio of non-synonymous over synonymous changes (S), N/S, revealed an excess of the latter, suggesting that genomes of Kva 342 and Kpn NTUH are under purifying selection (Table S5). We observed that the second most common mutational events in Kva 342 were IS insertions (16%), but gene deletions were very frequent in Kpn NTUH (28%). Strain BJ1 mutated primarily by small base-pair deletions (42%). Multifactorial ANOVA revealed that the frequency of each mutation type depended on both strain and environment, and their interaction (P < 0.001). Taken together, our results from both KlebEvoI and KlebEvoII show that, independently of the environment structure, Klebsiella strains evolve and mutate by different mechanisms.
Evolution of hypermucoidy as a byproduct of adaptation outside the host
Early on the KlebEvoII experiment, we observed that populations descending from capsulated ancestors developed the hypermucoid phenotype (HMP) in liquid (Figure S4A). Such phenomenon was exclusively observed in ASM and LB, the two media with highest carrying capacity, including in Kva 342, a strain which does not code for rmpA, a transcriptional regulator that increases capsule expression and is known to cause HMP and hypervirulence [49]. We plated all end-point populations to perform the string test, a hallmark of HMP (Figure S4B). Our results revealed that, when tested on agar, capsulated populations evolving in all environments, including soil, displayed HMP. In more than half of the populations of environmental strain Kva 342, HMP emerged de novo. However, we also observed that 12 out of the 54 populations derived from the string-test positive ancestral strains BJ1 and NTUH-K2044, no longer displayed this phenotype (Figure S4B). These results were largely in agreement with HMP quantification by slow centrifugation. Whereas populations of Kpn NTUH, only increased HMP in ASM (Figure 5B), almost all populations from strain Kva 342 and Kpn BJ1 increased HMP, irrespective of the environment (Figure 5B) (Two-way ANOVA). Overall, our results suggest that evolution in structured environments selects for hypermucoviscosity, independently of the presence of the rmp locus.
We then investigated the genetic basis of de novo emergence of HMP. Among the 13 Kva 342 clones sequenced that were capsulated and string-test positive, ten (77%) had single nucleotide polymorphisms in the active sites of the ATPase activity domain of wzc, the tyrosine kinase involved in capsule production (Table S6, Figure 5C). Introduction of the evolved wzc alleles in the ancestral Kva 342 resulted in compact, elastic and string-test positive colonies. These clones had an increased HMP relative to the ancestor (Figure 5D) but did not produce more capsule (Figure S5A). Similarly, evolved capsulated clones derived from Kpn BJ1 and Kpn NTUH with mutations in the wzc gene exhibited an increased HMP (Figure S5BC). The convergent evolution towards the HMP in structured environments is intriguing because it is a very costly (~30%) phenotype (Figure S6) during growth in well-mixed cultures [30]). This suggests that the large trade-offs between growth rate and HMP can be easily overcome in structured environments. Taken together, our results revealed that hypermucoviscosity can rapidly evolve by point mutations as a by-product of adaptation outside the host, suggesting a strong advantage in structured environments
Convergent evolution in ramA regulon results in increased sensitivity to antibiotics
Our findings that HMP can easily emerge in environments without biotic pressure could be worrisome, particularly because of the recent convergence of hypervirulent and multidrug resistant clones [50, 51]. However, we also observed that the same evolutionary conditions that selected for HMP also resulted in accumulation of mutations in the ramA/romA/ramR locus, and its regulon (47 mutations in 42 clones of Kva 342) (Table S7, Figure 6A). These loci are associated to lipid A biosynthesis, outer membrane stability [52] and resistance to antibiotics [53]. Accordingly, evolved clones with mutations in the locus, had increased sensitivity to ciprofloxacin, tetracycline and chloramphenicol (Figure S7). Introduction of the evolved alleles in the ancestral background, and reversion of one evolved allele to its ancestral sequence confirmed that these mutations were enough to increase the cell sensitivity to antibiotics (Figure 6B), and this is independent of the ancestral background (capsulated or non-capsulated).
In ten clones (6%), wzc mutations leading to HMP co-exist with those in the ramA regulon. However, binomial tests revealed that there is no dependent co-evolution between wzc mutations and ramA (P >0.05 for deviation from the 5% expected under a null assumption that these mutations do not co-evolve). Reversal of both mutations in an evolved clone or insertion of both mutations in the ancestral genotype, revealed that, despite an increased HMP, these clones were still more sensitive to antibiotics (Figure 6C).
Taken together, de novo HMP per se may not be of concern as the same abiotic conditions selecting for it, also favour mutations resulting in increased sensitivity to antibiotics.