Repeated phenotypic evolution by different genetic routes: the evolution of colony switching in Pseudomonas fluorescens SBW25

Repeated evolution of functionally similar phenotypes is observed throughout the tree of life. The extent to which the underlying genetics are conserved remains an area of considerable interest. Previously, we reported the evolution of colony switching in two independent lineages of Pseudomonas fluorescens SBW25 (Beaumont et al., 2009). The phenotypic and genotypic bases of colony switching in the first lineage (Line 1) have been described elsewhere (Beaumont et al., 2009; Gallie et al., 2015). Here, we deconstruct the evolution of colony switching in the second lineage (Line 6). We show that, as for Line 1, Line 6 colony switching results from an increase in the expression of a colanic acid-like polymer (CAP). At the genetic level, nine mutations occur in Line 6. Only one of these - a non-synonymous point mutation in the housekeeping sigma factor rpoD - is required for colony switching. In contrast, the genetic basis of colony switching in Line 1 is a mutation in the metabolic gene carB (Beaumont et al., 2009). A molecular model has recently been proposed whereby the carB mutation increases capsulation by redressing the intracellular balance of positive (ribosomes) and negative (RsmAE/CsrA) regulators of a positive feedback loop in capsule expression (Remigi et al., 2018). We show that Line 6 colony switching is consistent with this model; the rpoD mutation generates an increase in ribosome expression, and ultimately an increase in CAP expression.


INTRODUCTION
The repeated appearance of similar phenotypes is a striking feature amid the immense diversity of life. Many phenotypes have evolved multiple independent times in different lineages (Conway Morris, 1999). Examples include the evolution of analogous wing-like structures for flight in pterosaurs, birds, insects and bats (Alexander, 2015), C4 photosynthetic pathways in plants (Sage et al., 2011), and single-lens camera eyes in vertebrates and molluscs (Ogura et al., 2004). An intriguing aspect of repeated phenotypic evolution is the extent to which the underlying genetics are also conserved. It is commonly thought that the degree to which two organisms are related correlates with the degree of genetic parallelism underpinning evolutionary innovations, to the extent that repeated phenotypic evolution has historically been divided into 'parallel evolution' (the evolution of one phenotype from similar genetic backgrounds), and 'convergent' evolution (the evolution of one phenotype from distantly related organisms). The accumulation of genetic data in recent years has shown this assumption to be in need of revision. For example, clonal populations of Escherichia coli adapt to thermal stress via different genetic routes (Riehle et al., 2001), while pigmentation changes in mice and lizards are both underpinned by mutations in the Mc1r gene (Nachman et al., 2003;Rosenblum et al., 2004). The increasing number of examples of disparity between degree of relatedness and genetic parallelism (reviewed in Arendt and Reznick, 2008) hints at the underappreciated and poorly understood complexity of biological systems.
An evolution experiment with populations of the model bacterium Pseudomonas fluorescens SBW25 (Beaumont et al., 2009) has provided the opportunity to characterize a case of repeated phenotypic evolution in unusual detail. Twelve independent populations were subjected to multiple rounds of selection for novel colony morphologies. In every population, each round of selection concluded with the isolation of a single colony that had a phenotype different to that of the immediate ancestor for continuation into the subsequent round of selection ( fig. 1A). The final result was 12 independent evolutionary lineages, each with a clearly defined history of colony phenotypes and underlying genetic changes. Two lineages (Lines 1 and 6) converged on a similar striking capacity to stochastically switch -at high frequency -between different colony morphologies.
Colony switching in Line 1 has been extensively investigated (Beaumont et al., 2009;Rainey et al., 2011;Libby and Rainey, 2011;Gallie et al., 2015;Remigi et al., 2018). Emergent genotype 1B 4 produces a mixture of opaque and translucent colonies, and a corresponding mixture of capsulated and non-capsulated cells ( fig. 1B, 1C; Beaumont et al., 2009). The capsule consists of a colanic acid-like polymer (CAP), the ON/OFF expression of which leads to colony switching (Gallie et al., 2015). Nine mutational steps occurred during the evolution of 1B 4 ( fig. 1B). The first eight occur in genes involved in the production of c-di-GMP, a secondary messenger that affects the expression of a well-characterized acetylated cellulosic polymer (ACP, cellulose; Spiers et al., 2003;McDonald et al., 2009). The final mutation affected the central metabolic gene carB (c2020t; amino acid change R674C). This mutation, which is alone sufficient to cause colony switching, perturbs intracellular pyrimidine pools (Gallie et al., 2015 ; fig. 1D). Pyrimidine deficiency in 1B 4 has recently been shown to generate an increase in intracellular ribosome concentration, leading to the proposal of a translational control model for capsule switching (Remigi et al., 2018; see also fig. 6A). Briefly, the model proposes that capsule switching results from competition for binding sites on the mRNA of pflu3655-pflu3657, which encodes transcriptional regulators of CAP biosynthetic genes; ribosome binding results in translation (and promotion of capsulation), while RsmAE/CsrA binding inhibits translation (favouring the non-capsulated state; Remigi et al., 2018). The ribosome increase in 1B 4 is expected to tip the balance of the switch in favour of translation, increasing the probability of capsulation.

4
In this work, we characterize the phenotypic and genetic bases of colony switching in the second emergent genotype, 6B 4 . Comparisons with 1B 4 demonstrate that 6B 4 colony switching is a very similar phenotype realized by a different genetic route. We also show that the two genetic routes are reconciled at the molecular mechanistic level.  (Beaumont et al., 2009). Populations were subjected to bouts of selection in static or shaken liquid KB. After each bout cells were plated on KB agar and a single colony with novel morphology was used to start the subsequent round in the opposite environment. (B) Each of the nine strains in the Line 1 evolutionary series (SBW25à1B 4 ) has a colony phenotype distinct from that of its immediate ancestor. New mutations are noted as "gene (mutation)" at the point of occurrence.

6B 4 shows colony and capsule instability
The evolutionary history of 6B 4 includes ten colony phenotypes, with translucent-opaque colony instability emerging after nine rounds of evolution ( fig. 2A) 6 insertion site determined for each (supplementary table S1). Microscopic screening of cells showed capsule production to be eliminated in 43 genotypes, and severely reduced in a further nine genotypes. Three genotypes showed an increase in capsule production.
Of the genotypes with eliminated or reduced capsule production, 41 (75%) contained insertions in genes required for the production of CAP, a polymer previously described as the structural basis of the 1B 4 capsule (Gallie et al., 2015). These include insertions in genes predicted to encode CAP precursor biosynthetic machinery (e.g., algC), CAP biosynthetic machinery (20 genes: wcaJ-wzc) and probable CAP regulators (pflu3655, pflu3656, pflu3657, gacA/gacS). A direct deletion of the CAP biosynthetic locus from 6B 4 resulted in loss of both cell capsulation and colony instability ( fig. 3A, supplementary text S2). Together these results demonstrate that the structural basis of the 6B 4 capsule is encoded by the wcaJ-wzb locus.
A transcriptional fusion of lacZ to wcaJ was constructed in 6B 4 and the phenotype analysed on LB+Xgal plates ( fig. 3B, supplementary text S2). The resulting 6B 4 -wcaJ-lacZ colonies were a mixture of white (with a high proportion of Capcells), and blue (with a high proportion of Cap + cells). The same construction in 6A 4 -the immediate ancestor -resulted in uniform colonies, showing that CAP expression is at least partially controlled at the level of transcription (later corroborated by RNA-seq data; see supplementary tables S3-S5).  relative to 6A 4 , the expression of D-Fucose (Fuc), D-glucuronic acid (GlcA), D-Galactouronic acid (GalA) and two unknowns are increased. Each of these is also increased in 1B 4 relative to 1A 4 , indicating that the 1B 4 and 6B 4 capsule polymers are the same.
Thus far, the transposon mutagenesis, strain constructions and structural analysis of the capsule polymers (and later, RNA-seq data) point to the same phenotype for 1B 4 and 6B 4 : switching between opaque and translucent colonies caused, at the single cell level, by ON/OFF expression of CAP. The only difference observed between the two genotypes lies in the frequency of capsulation and size of capsules (both increased in 6B 4 relative to 1B 4 ).

Nine mutations detected in 6B 4
Next, the genetic basis of 6B 4 capsule switching was investigated. Whole genome sequencing of 6B 4 identified seven mutations. This was surprising, as nine mutations were expected -one per round of REE selection (see fig. 1A). Sanger sequencing of the identified loci at each bottleneck in the evolutionary series revealed two gaps at the beginning: SBW25à6B 0 (selection round 1) and 6B 0 à6A 1 (selection round 2; table 1, fig. 2A).
Extensive previous knowledge suggested that these two genotypes almost certainly carried mutations in one of three loci (wsp, aws, mwsR) (McDonald et al., 2009). Sanger sequencing of wspF revealed a point deletion in 6B 0 (∆t475) that was absent in 6A 1 ( fig. 2A). The loss of the wspF ∆t475 mutation was not repeated among twenty independent repeats of a single round of REE from 6B 0 . The absence of the wspF deletion in 6A 1 most likely represents a rare mutational reversal (table 1, fig. 2A).
The nine Line 6 mutations occur in a modular, paired fashion. The first six mutations occur in previously identified c-di-GMP producing loci (awsX/awsR, wspF/wssB); mutations in these loci are known to cause the gain and loss of cellulose production and wrinkly spreader colony morphology (McDonald et al., 2009;Beaumont et al., 2009;Gallie et al., 2015;Lind et al., 2015;Lind et al., 2016). The sixth mutation -an in-frame, six base pair deletion in the cellulose biosynthetic gene wssB -completely abolishes cellulose production ( fig.   2A), and appears to render downstream genotypes unable to produce the wrinkly spreader colony morphology by further mutation. Accordingly, the next pair of mutations occur in an unrelated locus: nlpD (pflu1301), which encodes a lipoprotein predicted to have a function in cell wall formation and cell separation in a range of bacteria (Stohl et al., 2015;Lind et al., 2016;Tsang et al., 2017;Yang et al., 2017). The first of these, mutation seven, generates a nonsense mutation in nlpD resulting in the production of cell chains and round colonies in 6B 3 ( fig.   2A). This mutation has previously been reported to generate a cell chain phenotype in SBW25 (Lind et al., 2016), and similar mutations have been reported in Escherichia coli (Uehara et al., 2010), Vibrio cholerae (Möll et al., 2014) and Yersinia pestis (Tidhar et al., 2009). In short, NlpD is an activator of cell division protein AmiC; inactivation of NlpD leads to incomplete cell division. Mutation eight converts the nlpD nonsense mutation into a tryptophan residue, reversing the cellular and colony phenotypes in 6A 4 ( fig. 2A). The final mutation, with which colony switching emerges, is in rpoD (t1682c, resulting in amino acid change V561A).

8
This gene encodes the housekeeping sigma factor (σ 70 ) that controls transcription of many genes involved in cell growth and division (Schulz et al., 2015).
There are two notable points of similarity and contrast between the evolutionary histories of 6B 4 ( fig.   2A) and 1B 4 ( fig. 1B). Firstly, both lineages begin in a similar fashion with mutations affecting cellulose production and wrinkly spreader colony morphology. In Line 6, mutational routes to the wrinkly spreader phenotype are presumably rendered inaccessible by the sixth mutation (in wssB), providing an opening for a pair of mutations in nlpD. Contrastingly, cellulose production is not abolished in Line 1, with 1B 4 staining positive for cellulose and forming wrinkly spreader-like mats. Accordingly, Line 1 mutations are in cellulose-affecting loci up until the final, switch-causing mutation. Secondly, the final mutation in each lineage -that with which colony switching emerges -is a non-synonymous point mutation in different and, at first glance, functionally unrelated housekeeping genes.  ; fig. 4). The rpoD mutation was then engineered into the distant ancestor, SBW25, in the absence of any other mutations. The resulting genotype, SBW25-rpoD*, also showed distinct colony types and a high level of capsulation. A capsule counting assay revealed that while the rpoD mutation alone was sufficient to cause switching, SBW25-rpoD* showed a lower degree of capsulation than 6B 4 (t-test p=1.6x10 -3 ; fig. 4C). Therefore, while the rpoD mutation does cause CAP switching, one or more of the other six mutations contribute quantitatively to the capsule switching phenotype. This is in contrast to the c2020t carB mutation in Line 1, which accounts alone for the level of CAP switching seen in 1B 4 (Gallie et al., 2015).

Repeated evolution of switcher genotypes reveals additional rpoD mutations
To identify additional mutations able to cause capsule switching in 6A 4 , new switcher genotypes were evolved from 6A 4 . Each of 56 independent microcosms was inoculated with 6A 4 and put through a single round of the REE (Beaumont et al., 2009). Nine new switcher genotypes were isolated from nine independent microcosms (genotypes Re1-Re9; see supplementary text S1). Sequencing of rpoD revealed a single, non-synonymous point mutation in each; eight of the new switchers (Re1-Re8) contained mutation a1723c leading to amino acid change T575P, while one (Re9) carried a1745c causing amino acid change Q582P. All three rpoD mutations (t1682c, a1723c, a1745c) are located in the H-T-H motif that interacts with the -35 consensus sequence of σ 70 dependent promoters ( fig. 5A; Hu and Gross, 1988;Siegele et al., 1989). Interestingly one mutation, T575P, leads to a significantly higher capsulation rate (t-test p<0.001; fig. 5B-C).  The equivalents of the above comparisons have been previously published for Line 1 (GEO database submission number GSE48900; (Gallie et al., 2015). While the numbers of differentially expressed genes are much higher in the Line 1 comparisons -most likely attributable to there being only a single biological replicate for each Line 1 morphotype -the overall pattern remains; the highest number of differentially expressed genes is between 1A 4 and 1B 4 -Cap + , and the lowest between 1B 4 -Capand 1B 4 -Cap + . A "comparison of comparisons" was performed, whereby each of comparisons A, B and C for Line 6 was equated to the Line 1 counterpart. Lists of shared and unique genes for comparisons A, B and C were generated (supplementary table S6). For comparison C, B 4 -Capversus B 4 -Cap + , 26 genes are common between Line 6 and Line 1; nine of these are more highly expressed in Capforms compared to the Cap + , and include four flagella genes and five genes of unknown function. The remaining 17 genes are more highly expressed in the Cap + forms than the Cap -, and include seven CAP genes, a transcriptional regulator, an inorganic ion transport gene and eight genes of unknown function.
Together, the results corroborate the finding that capsules and flagella are mutually exclusive. A similar finding has recently been reported in Cronobacter sakazakii, in which induction of colanic acid biosynthesis is accompanied by a reduction in flagella gene expression (Chen et al., 2018).

Ribosomal genes and proteins are overexpressed in 6B 4
The recently proposed ribosome-RsmAE model of 1B 4 capsule switching postulates that capsulation is controlled by the combined intracellular pool of ribosomes and RsmAE (Remigi et al., 2018); fig. 6A).  ; supplementary text S3). These results are consistent with the rpoD mutation increasing the probability of capsulation by generating an increase in ribosome expression.   . 6B); in particular, deletion of rsmA1 resulted in significantly higher levels of capsulation (one-sided t-test p=0.001112), bringing the percentage of capsulated cells to almost 100%. Secondly, the model predicts RsmAE activity to be reduced by deletion of mvaT, which encodes a transcriptional repressor of rsmZ -itself a negative regulator of RsmAE -in Pseudomonas aeruginosa (Brencic et al., 2009). Deletion of mvaT (pflu4939) from 6B 4 or 1B 4 results in a significant increase in capsulation (p<0.01; fig. 6C, supplementary text S2).
In contrast to the above, any bias of the switch machinery in favour of RsmAE is expected to inhibit translation of pflu3655-3657 mRNA, and thus reduce capsulation. Support for this side of the model comes from the transposon mutagenesis screen (supplementary table S1). Firstly, three capsule-reducing insertions were obtained in the GacA/GacS two-component sensory system, which is a negative regulator of RsmAE in γproteobacteria (Lapouge et al., 2008) fig. 6A). Inactivation of these genes is expected to increase RsmAE expression and decrease capsulation. Indeed, 6B 4 -TnCre-gacA and 6B 4 -TnCre-gacS (Cre-deleted forms of the transposon mutants, see methods and supplementary text S2) showed a complete absence of capsulation ( fig.   6D). Secondly, four transposon insertions were identified in genes involved in the production of mature tRNAs: two in gidA/mnmG (pflu6129), one in truA (pflu4189) and one in thiI (pflu0349) (reviewed in Yacoubi et al., 2012). Each of these insertions resulted in a reduction in capsulation ( fig. 6D). While not lethal, disruption of each of the three tRNA modification genes is expected to reduce translational speed (Yacoubi et al., 2012; reviewed in Shepherd and Ibba, 2015), suggesting a role for efficient translation in 6B 4 capsulation.
The ability to increase and decrease 6B 4 capsules by manipulating components of the 1B 4 ribosome-RsmAE circuitry (as predicted by the model) demonstrates that the same intracellular architecture underpins switching in both genotypes.

DISCUSSION
In this work 6B 4 has been extensively characterised. Its phenotype and genotype have been compared with those previously reported for 1B 4 -a strain evolved in parallel to, but independently of, 6B 4 (Beaumont et al., 2009;Gallie et al., 2015). 6B 4 and 1B 4 populations show elevated levels of CAP-based capsule expression and colony switching ( fig. 1B-E, 2A). The phenotype is realised by two distinct genetic routes, culminating in a mutation in either rpoD (Line 6) or carB (Line 1). Both mutations promote increased translation of mRNA encoding positive regulators of the CAP biosynthetic machinery. These regulators also activate their own transcription, forming a positive feedback loop that results in bistable capsule expression (Remigi et al., 2018 ; fig. 6A).
Line 1 and Line 6 were derived from a single clonal ancestor (P. fluorescens SBW25). This means that the genotypes of interest, 6B 4 and 1B 4 , share an evolutionary history of many millions of years followed by a comparatively minuscule period of several weeks of independent evolution in experimental microcosms. Given the extensive shared history, it is not surprising that the same phenotype emerged in both lineages. It is surprising, however, that such different molecular pathways generate the same phenotype.
Repeated phenotypic evolution has been documented many times in both the laboratory (e.g., Riehle et al., 2001;Cooper et al., 2003;Fong et al., 2005;Ostrowski et al., 2005;Bantinaki et al., 2007;Meyer et al., 2012;Lindsey et al., 2013) and natural populations (Nachman et al., 2003;Rosenblum et al., 2004;Stern and Frankel, 2013;Riveron et al., 2014). In many of these examples, repeated phenotype evolution is determined by changes in the same gene or molecular pathway (e.g., Meyer et al., 2012;Lindsey et al., 2013;Riveron et al., 2014). The fact that colony switching in Lines 1 and 6 arise by different molecular routes -despite extreme shared ancestry -is surprising. At first glance, rpoD and carB seem functionally unrelated and, as such, it is natural to assign them to separate functional compartments. However, as this work shows, the two genes are connected at the level of their effects on ribosomes: both mutations increase expression of ribosomal genes (supplementary table S7; Gallie et al., 2015;Remigi et al., 2018) and thus tip the balance of the switch in favour of CAP mRNA translation ( fig. 6A).
The precise molecular mechanisms by which rpoD and carB mutations alter ribosomal gene expression remain to be elucidated. However, it is conceivable that the rpoD mutation directly increases transcription from one or more ribosomal genes. A point mutation in Salmonella typhimurium rpoD has recently been shown to increase transcription from rpsT (Knöppel et al., 2016). In the case of the carB mutation, which perturbs intracellular pyrimidine pools (Gallie et al., 2015), the reported influence of nucleotide triphosphate (NTP) concentrations on rrn promoters may play a mechanistic role (Gaal et al., 1997;Schneider et al., 2002;Murray et al., 2003;. If cellular components show a high degree of connectivity, it follows that many other factors could also affect the switch circuitry. Possible candidates include those affecting capsule expression and identified via the transposon mutagenesis screens (e.g., hslO, sahA, ndk; supplementary table 1; Gallie et al., 2015).
In stark contrast to the disparate molecular evolution of 6B 4 and 1B 4 , repeated bouts of evolution from the same immediate ancestor of the 6B 4 switching genotype, namely, 6A 4 , resulted in re-evolution of the switching genotype by mutations solely in rpoD ( fig. 5A). Similar repeated bouts of evolution from 1A 4 (the immediate ancestor of the Line 1 switching genotype) resulted in switching types with mutations in genes encoding the determinants of pyrimidine biosynthesis (five in carB, one in pyrH; Gallie et al., 2015). In other words, the comparatively tiny portion of evolutionary history for which Line 6 and Line 1 diverged -several weeks compared with millions of years of common history -has a significant impact on molecular evolution.
The distinct classes of switcher mutations in Lines 1 and 6 are likely to result from positive epistatic interactions: while both types of switch-causing mutations presumably arise in both backgrounds, rpoD mutations provide a significant growth advantage in 6A 4 (and, by extension, carB/pyrH mutations in 1A 4 ). The growth advantage afforded by rpoD mutations in 6A 4 can be seen in fig. 5D, where each of three rpoD mutants grows more quickly and to a higher final density than ancestral 6A 4 in shaken KB. The advantage of the t1682c rpoD mutation (that from 6B 4 ) disappears in the absence of the previous six mutations (SBW25 grows faster than SBW25-rpoD* ( fig. 5D)). Precisely which of the first six mutations in the Line 6 evolutionary series contribute to the observed epistatic effect remains to be tested. However, the two nlpD mutations immediately preceding the rpoD mutation are prime candidates for two reasons. Firstly, nlpD is the only locus that is mutated in Line 6 but not Line 1 ( fig. 1B, 2A). Secondly, nlpD is immediately upstream of rpoS (pflu1302), which encodes the stationary phase sigma factor RpoS (σ 38 ). RpoS and RpoD (together with other sigma factors) compete for binding of core RNA polymerase (Ishihama, 2000;Mauri and Klumpp, 2014), and so their relative intracellular concentration affects the expression level of their respective regulons (Gross et al., 1998;Mauri and Klumpp, 2014). It is possible that the nlpD mutations, in addition to altering colony morphology via a reduction of NlpD/AmiC activity, also alter the expression of rpoS. Indeed, a promoter for rpoS has previously been reported within E. coli and P. aeruginosa nlpD (Takayanagi et al., 1994;Lange and Hengge-Aronis, 1994;Kojic and Venturi, 2001). A change in RpoS concentration could conceivably set the stage for compensatory mutations in RpoD.
Understanding the molecular bases of adaptive phenotypes continues to present significant challenges even when aided by high-throughput genomic technologies. As shown here and elsewhere (Larsen et al., 2008;Gallie et al., 2015;Bershtein et al., 2015;Grenga et al., 2017;Carvalho et al., 2018), mutations -particularly those in central metabolism -can have complex effects that extend well beyond the immediate neighbourhood of gene function. That point mutations in two seemingly unrelated genes (rpoD and carB) can generate stochastic capsule switching draws attention to the interconnectedness of cell physiology and highlights the extensive mutational opportunities available to evolution.

MATERIALS AND METHODS
Bacterial strains, plasmids and media. Details of bacterial strains and plasmids used are provided in supplementary text S2. Unless otherwise stated, P. fluorescens strains were grown for 24 hours at 28˚C in shaken 30 mL glass microcosms containing 6 mL King's Medium B (KB; Ward et al., 1954). Where stated, uracil L-arginine hydrochloride and/or guanine hydrochloride (Sigma-Aldrich) were added to the medium. Cells were plated on KB or Lysogeny Broth (LB) containing 1.5% agar. Antibiotics were used at the following concentrations: tetracycline (12.5 µg mL -1 ; Tc); kanamycin (100 µg mL -1 ; Km); nitrofurantoin (100 µg mL -1 ; NF).

Microscopy.
Cell microscopy was performed using a Zeiss Axiostar Plus bright field microscope. A dissection microscope was used for colony images. Microscopy images were cropped and processed in Preview or Microsoft Word as indicated in figure legends.
Capsule counting assay. Capsule staining and the counting assay were performed as previously described (Gallie et al., 2015). Briefly, for each strain to be assayed, three (for the Cre-deleted transposon mutant assay in fig. 6D) or five (all others) single colonies were grown to stationary phase in KB cultures. Cultures were transferred to fresh KB and grown to mid-exponential phase. Cells from each culture were stained with 1:10 diluted India ink (Pébéo) and photographed under bright field x60 magnification. Capsule expression was recorded manually for 500 cells per replicate (≤100 cells assayed per photograph). Average proportions of Cap + cells were determined and statistical analyses performed.
Gene deletions and mutation construction. Gene deletions were constructed in the SBW25 background by pUIC3-mediated two step allelic exchange as described elsewhere . For further details of genetic constructs see supplementary text S2.
The construct was used to transform E. coli DH5α-λpir and transferred to 6A 4 and 6B 4 via tri-parental conjugation (with a helper strain carrying pRK2013). Successful transconjugants were purified, giving 6A 4 -wcaJ-lacZ and 6B 4 -wcaJ-lacZ. Single colonies of these constructed genotypes were grown at 26˚C for 48 hours on LB+Tc+X-gal (60 µg mL -1 ) plates prior to microscopic analysis.
Transposon mutagenesis. 6B 4 was subjected to random mutagenesis as described in Giddens et al., 2007. Approximately 10,000 transposon mutants from eleven independent conjugations were screened on LB+Km, on which 6B 4 mutants typically form opaque colonies after ~72 hours. Mutants that formed translucent or otherwise different colonies were selected and screened microscopically for obvious alterations in capsule expression.
Mutants of interest were then purified and the insertion site determined by AP-PCR. In selected strains, the bulk of the transposon was deleted leaving 189 base pair at the insertion site ("TnΔ-" genotypes) and eliminating polar effects (Giddens et al., 2007).  (Li et al., 2008)) and ELAND (Illumina, Inc.).

Isolation and analysis of extracellular polysaccharide (EPS
Insertions and deletions were identified by analysing genomic regions with unusual coverage and BLAST analysis of discarded sequences. Genome sequence files available on request. Re-evolution of switchers from 6A 4 . Nine independent switcher genotypes were isolated from 6A 4 according to the REE protocol (Beaumont et al., 2009). Each switcher was purification streaked and rpoD sequenced.
Growth curves and analysis. Single colonies were grown for each strain of interest (KB agar, 26˚C, 48 h).
Eight colonies per strain were used to inoculate 200 µL KB in wells of a 96-well plate. The plate cultures were grown for 24 hours at 26˚C, 200 rpm. Each well was then mixed by pipetting and 2 µL culture transferred to 198 µL fresh KB. This second plate was then incubated at 26˚C for 72 hours in a BioTek Epoch 2 plate reader, and the OD 600 of each well measured at five minute intervals (5 seconds of 3 mm orbital shaking preceding each read). Data from each well was plotted, and the mean and standard error of eight wells per strain (minus absorbance in the media control wells) used to draw fig. 5D. V max (maximum growth rate) and lag time were calculated using a sliding window of six time points during exponential growth (between 1-24 hours, based on observation of growth curves) using Gen5 Software version 3.00.19.
RNA-seq analysis. For each of 6A 4 and 6B 4 , three single colonies were grown overnight in independent KB microcosms, diluted 1:1000 into 20 mL KB in 250 mL flasks and grown to mid-exponential phase (~OD 600 of 0.4-0.6). Total RNA was then harvested from each culture; for 6A 4 cultures, 100 µL culture were mixed with 900 µL KB, pelleted and resuspended in 1 mL of RNAlater® (Ambion®). For 6B 4 cultures, 100 µL was separated into Cap + and Capfractions by centrifugation, and each was resuspended in 1 mL RNAlater®. All mRNA extractions proceeded using a RiboPure TM Bacteria Kit (Ambion®). Normalized mRNA-seq library preparation, followed by 100 base pair paired-end Illumina HiSeq 2500 sequencing, was performed by the Australian Genome Research Facility (Brisbane, Australia; GEO submission number GSE116490). The data was analysed with Bowtie2 (Langmead and Salzberg, 2012), HTSeq (Anders et al., 2015) and the R-package DESeq2 (Love et al., 2014). First, RNA-seq datasets were mapped to the SBW25 genome (downloaded from GenBank under the accession number NC_012660) via Bowtie2 with default settings. The coverage per gene of the genome mapping was determined with HTSeq. The gene annotation for HTSeq was also downloaded from the SBW25 GenBank entry. Then, differentially expressed genes were identified by applying DESeq2. The standard workflow in https://bioconductor.org/packages/release/bioc/manuals/DESeq2/man/DESeq2.pdf was used, except that the alpha parameter was set to 0.3 to reduce the number of genes that were falsely classified as not significantly differentially expressed between the different morphotypes. Three comparisons were made: 6A 4 versus 6B 4 -Cap -, 6A 4 versus 6B 4 -Cap + , and 6B 4 -Capversus 6B 4 -Cap + (supplementary tables S3-S5). The corresponding comparisons for Line 1 are available elsewhere (Gallie et al., 2015).

Statistical analyses.
To detect differences in capsulation levels or nucleotide concentrations between two strains, two-sample t-tests (parametric or Welch) or, where normality assumptions were violated, Wilcoxon rank sum tests were applied. To detect differences in capsulation levels across three strains (while testing overexpression genotypes), one-way ANOVA or, where normality assumptions were violated, Kruskal Wallis tests were used. Exact binomial tests were used to detect differences in ribosomal gene expression between morphotypes in the RNA-seq data (see also supplementary text S3). All analyses were performed in R version 3.3.3. On graphs: *=0.05<p<0.01, **=0.01<p<0.001, ***=p<0.001.

ACKNOWLEDGMENTS AND FUNDING INFORMATION
The raw RNA-seq data is deposited at GEO (submission number GSE116490). This work was supported by The Marsden Fund (all authors) and the Max Planck Society (JG, FB, PBR).