A mutant RNA polymerase activates the general stress response, enabling E. coli adaptation to late prolonged stationary phase

Escherichia coli populations undergo repeated replacement of parental genotypes with fitter variants deep in stationary phase. We isolated one such variant, which emerged after three weeks o of maintaining an E. coli K12 population in stationary phase. This variant displayed a small colony phenotype, slow growth and was able to outcompete its ancestor over a narrow time window in stationary phase. The variant also shows tolerance to beta-lactam antibiotics, though not previously exposed to the antibiotic. We show that an RpoC (A494V) mutation confers the slow growth and small colony phenotype to this variant. The ability of this mutation to confer a growth advantage in stationary phase depends on the availability of the stationary phase sigma factor σS. The RpoC (A494V) mutation up-regulates the σS regulon. As shown over 20 years ago, early in prolonged stationary phase, σS attenuation, but not complete loss of activity, confers a fitness advantage. Our study shows that later mutations enhance σS activity, either by mutating the gene for σS directly, or via mutations such as RpoC (A494V). The balance between the activities of the housekeeping major sigma factor and σS sets up a trade-off between growth and stress tolerance, which is tuned repeatedly during prolonged stationary phase. Importance An important general mechanism of bacterial adaptation to its environment involves adjusting the balance between growing fast, and tolerating stresses. One paradigm where this plays out is in prolonged stationary phase: early studies showed that attenuation, but not complete elimination, of the general stress response enables early adaptation of the bacterium E. coli to the conditions established about 10 days into stationary phase. We show here that this balance is not static and that it is tilted back in favour of the general stress response about two weeks later. This can be established by direct mutations in the master regulator of the general stress response, or by mutations in the core RNA polymerase enzyme itself. These conditions can support the development of antibiotic tolerance though the bacterium is not exposed to the antibiotic. Further exploration of the growth-stress balance over the course of stationary phase will necessarily require a deeper understanding of the events in the extracellular milieu.

prolonged stationary phase: early studies showed that attenuation, but not complete elimination, of 23 the general stress response enables early adaptation of the bacterium E. coli to the conditions 24 established about 10 days into stationary phase. We show here that this balance is not static and 25 that it is tilted back in favour of the general stress response about two weeks later. This can be 26 established by direct mutations in the master regulator of the general stress response, or by 27 mutations in the core RNA polymerase enzyme itself. These conditions can support the 28 development of antibiotic tolerance though the bacterium is not exposed to the antibiotic. Further 29

32
Under standard laboratory conditions, well-aerated cultures of Escherichia coli exhibit three distinct 33 phases of growth within 24 hours of inoculation: the initial slow growing lag phase, followed by a 34 period of exponential growth and finally the stationary phase during which no appreciable change 35 in cell number is observed. Most of the cells in culture lose viability within 2-5 days after reaching 36 stationary phase [(1), (2)]. The small surviving subpopulation can remain viable for at least 60 37 months in a phase known as "prolonged stationary phase" or "long-term stationary phase" (1,3,4). influence hierarchy of carbon source utilization. Several mutations in the core RNA polymerase 54 7 SC, starting from a 1000-fold minority, overtook LC after two days (Supplementary Figure S3). 123 Therefore, SC shows the ability to outcompete both ZK819 and LC in prolonged stationary phase in 124 conditions representing a narrow time-window. 125 rpoC A494V is a unique mutation in SC that causes small colony size, irrespective of background 126 To identify and characterize the genetic basis of the small colony and the GASP phenotype of SC, we 127 performed whole-genome sequencing of SC, LC and ZK819. We found several mutations common to 128 SC and LC, and identified two loci at which the two diverged. The first was at rpoS, encoding σ S . 129 Whereas ZK819 carried the attenuated rpoS819 allele, SC carried what we refer to as rpoS92. 130 rpoS819 is defined by a 46 bp duplication at the 3' end of the gene. This duplication results in a 131 longer protein with reduced activity, which may in part be due to its reduced expression at the 132 protein level. rpoS92 carries a reduplication of the 46 bp region. This introduces a STOP codon 133 prematurely for rpoS819, resulting in a σS protein variant of length intermediate between the wild 134 type and rpoS819. rpoS92 partially restores the activity of σ S that had been lost in rpoS819 [(9)]. LC, 135 in contrast to both SC and ZK819, carries the wild type rpoS ORF. 136 The second locus at which these three strains diverged was rpoC, coding for the β' subunit of the 137 core RNA polymerase. Whereas ZK819 and LC carry the wild type rpoC locus, SC has a G:A 138 transition, resulting in the replacement of the alanine at residue 494 by a valine. Mutations in the 139 core RNA polymerase have been found to confer adaptive advantages to E. coli under different 140 conditions, and have been found in at least two whole genome sequencing studies of variants that 141 emerge in prolonged stationary phase [(5), (10)]. We also identified a unique synonymous mutation 142 (H366H) in LC in the gene icd coding for isocitrate dehydrogenase, which we have not characterized 143 in this study. 144 To test the phenotypic effect of rpoC A494V mutation, we introduced it in different relevant genetic 145 backgrounds using phage mediated transduction from a thiC39::tn10 marker (CAG18500), as 146 described in a previous study [(18)]. The genetic backgrounds in which rpoC A494V was introduced 147 were selected based on their σ S status and their relevance to the strains used or isolated in this 148 study; these were LC (evolved strain, rpoS WT ), ZK819 (ancestor, rpoS819), ZK126 (rpoS WT , the parent 149 of ZK819) and ΔrpoS (ZK819::ΔrpoS). In addition, the wild type rpoC allele was transduced into SC 150 reverting the rpoC A494V mutation to the wildtype. The presence of the rpoC A494V mutation was 151 confirmed by sanger sequencing with relevant primers (Supplementary Table S1). Single-end whole 152 genome sequencing were performed on all transductant strains to check for the presence of any 153 additional indels, copy number variations and point mutations that may have been inadvertently 154 introduced during transduction and found none apart from those introduced. 155 We observed that rpoC A494V bearing transductants in all backgrounds, including ΔrpoS, produced 156 small colonies on LB agar plates and exhibited slow growth, characteristic of SC in liquid culture 157 ( Figure 5 A, B). Conversely, replacing the mutant rpoC A494V allele in SC background with the wild 158 type rpoC allele ameliorated slow growth, thus reverting the small colony phenotype to ancestral 159 colony size. These results indicate that the rpoC A494V mutation is sufficient to cause the small colony 160 phenotype to its bearer, independent of the status of σ S . 161 rpoC A494V mutant might cause inefficient transcription 162 The ability of rpoC A494V to slow growth is independent of the status of σ S . This being a mutation in 163 the core RNA polymerase, we suspected that it would have a broad impact on global gene 164 expression levels. 165 To test whether the mutation causes any alteration to transcription, we tested induction kinetics of 166 the chromosomal ara operon in rpoC mutant and wild type strains in ZK819 background. Note that 9 the ara operon comes under the control of the housekeeping sigma factor σ D and not σ S . We 168 measured the induction of the araB gene over time following the addition of arabinose to an 169 exponentially growing culture by performing RT-PCR using primers amplifying a 134 bp region at 170 the 5' end of the gene (Supplementary Table S1). To limit any population-level heterogeneity in ara 171 induction [ (19)], we used nearly 100-fold excess of arabinose for induction. 172 In rpoC WT , araB levels increase steeply (ΔΔFCinduced-uninduced = -6.05) within 50 minutes of 173 adding arabinose. In rpoC A494V , increase in mRNA is muted (ΔΔFCinduced-uninduced = ranging from 174 -0.74 to -4.54), more variable across biological replicates, and delayed, occurring after 110 minutes 175 post induction ( Figure 6A). This delayed and repressed response to arabinose induction in rpoC A494V 176 suggests that the mutation reduces the activity of RNA-pol. This is unlikely to be an effect of a 177 possible reduced arabinose transporter expression in the mutant: RNA-seq data (see below), 178 showed no significant difference in the basal expression level of araE between the wild type and 179 mutant rpoC strains. Thus slow transcription might underpin the ability of rpoC A494V to cause slow 180 growth. 181

Competitive advantage conferred by rpoC A494V mutation depends on the status of σS 182
Having established that the rpoC A494V mutation can cause the small colony phenotype independently 183 of the status of σ S , we interrogated the ability of this mutation to confer GASP to its bearer. To do 184 this, we performed mixed culture competition experiments of rpoC A494V strains against rpoC WT 185 strains in all relevant genetic backgrounds in both fresh and spent media. 186 We found that rpoC A494V confers GASP on its bearer only in the ZK819 background, carrying the 187 attenuated rpoS819 allele, and not others ( Figure 5C, Supplementary Figures S4-8). In 10 day-old 188 spent media, ZK819+rpoC A494V overtook ZK819+rpoC WT from a 1000 folds lower starting frequency 189 after two days. However, SC with rpoS92 and rpoC A494V did not outcompete SC with rpoS92 and 190 rpoC WT . This suggests, under the assumption that these mutations emerged under a regime of 191 natural selection, that the rpoC A494V mutant might have arisen in the rpoS819 background, with 192 rpoS92 emerging later. We also found that in the ZK819ΔrpoS background, both rpoC A494V and 193 rpoC WT exhibited precipitous drops in cell count with no viable colonies remaining after three days 194 of growth, highlighting the importance of rpoS in survival and growth in nutrient depleted/altered 195 Figure S4). 196

conditions (Supplementary
These suggest that rpoC A494V confers GASP only when σ S is severely attenuated, as seen in rpoS819. 197 However, its ability to confer GASP does require some level of functional σ S , bringing about the 198 suggestion that it acts through σ S . 199 rpoC A494V induces expression of the σ S regulon depending on the rpoS genetic background ΔrpoS). We identified differentially expressed genes that showed a log2-fold change of over 1.5 at P-204 value of less than 0.05. 205 The number of differentially expressed genes across these five backgrounds ranges from 407 to 206 654. However, the number of genes differentially expressed irrespective of the genetic background 207 was only 18. This shows that the effect of rpoC A494V on the transcriptome is dependent on the 208 background expression state established at least in part by σ S activity. 209 Because the ability of rpoC A494V to confer GASP -as well as alter gene expression states -seemed to 210 depend on the status of σ S , we tested the effect of the mutation on the expression levels of σ S targets. 211 Towards this, we compared the fold changes -between rpoC A494V and rpoC WT -of targets of σ S with 212 those of a set of targets of the housekeeping sigma factor σ D . In a second analysis, we tested for the 213 overlap between σ S and σ D target genes that are differentially expressed in rpoC A494V over rpoC WT . 214 When we looked at all genes, σ S targets showed higher positive fold changes in rpoC A494V when 215 compared with rpoC WT , only in the ZK819 background carrying the attenuated and elongated 216 rpoS819 allele ( Figure 6B). Consistent with this, sets of differentially expressed gene sets between 217 rpoC A494V and rpoC WT were significantly enriched for σ S targets in ZK819 (rpoS819) background, but 218 not the other backgrounds (Supplementary Table S2-S6). 219 A previous study had identified rpoC A494V as a mutation that enables RNAP to bypass the 220 requirement of the small molecule ppGpp in transcribing a ppGpp dependent, σ 54 promoter [ (18)]. 221 Following from this work, we analysed the expression patterns of genes known to respond to ppGpp 222 [(20)] in rpoC A494V compared to rpoC WT . We observed that positively regulated targets of ppGpp are 223 upregulated and negatively regulated targets down regulated in rpoC A494V compared to rpoC WT in the 224 ZK819 and SC backgrounds, which harbor elongated rpoS alleles with less than wild type σ S activity 225 ( Figure 6C). This applies for genes that are known to respond to ppGpp, but not known to be 226 regulated by σ S as well. While this might indicate that this effect may not be mediated by σ S , we 227 observed an opposite pattern in rpoC A494V compared to rpoC WT in the ΔrpoS background ( Figure 6C). 228 We did not observe any significant effect on ppGpp targets in the rpoS WT background. 229 rpoC A494V does not seem to induce the σ S regulon by increasing the expression levels of σ S , on the 230 basis of our observation that there is little differential expression of the rpoS gene in the RNA-seq 231 data, nor is there any increase in σ S protein levels as measured by Western blotting. In fact, Western 232 blots indicate a decrease in the levels of σ S in rpoC A494V (Figure 7). This suggests that rpoC A494V may 233 activate the σ S regulon by affecting σ-factor competition for the core RNA polymerase. There is 234 evidence [ (21)] that ppGpp reduces the ability of σ D to compete with at least two alternative sigma 235 factors including σ S . Whether the induction of the σ S regulon by rpoC A494V operates through this 12 mechanism, and whether this might even require the extended C-terminal tail that rpoS819 and 237 rpoS92 alleles add to the wild type rpoS, remains to be tested. 238 In summary, though rpoC A494V increases the expression of the σ S regulon in backgrounds with low σ S 239 activity, this effect requires the presence and expression of σ S . The manifestation of GASP in the 240 regime in which rpoC A494V displays a selective advantage might require higher σ S activity, in contrast 241 always a trade-off between growth and stress response, which is determined by this competition 260 between the two sigma factors. The expression and activity of σ S is tightly regulated at multiple 261 levels, and adjusting this switch helps the bacterial cell modulate global gene expression states 262 according to environmental cues. [ (24,25)]. Additionally, the gene for σ S is highly mutable, thus 263 enabling genetic alterations of the growth-stress tolerance trade-off (26). 264 The attenuated rpoS819 allele alters this growth-stress response tradeoff in favour of growth. In our 265 experiments we found that the rpoS819 allele, when evolved in LB media, suffers another re-  (37,38)]. Under our experimental framework, we observed that E. coli 294 diversifies into stable, regulatory mutants which phenotypically produces small colonies with 295 enhanced stress tolerance after three weeks of incubation in rich spent media. 296 In summary, we show that a mutation in the core RNA polymerase provides a competitive 297 advantage to its bearer in deep stationary phase, by altering the balance between growth and stress 298 tolerance. 299

300
Bacterial strains and culture conditions 301 Escherichia coli K-12 str W3110 is the parent strain of Escherichia coli ZK126. The latter, when 302 evolved in prolonged stationary phase for 10 days gave rise to Escherichia coli ZK819 which has an 303 elongated and attenuated rpoS allele [(4)] called rpoS819. ZK819 was evolved in prolonged 304 stationary phase for 28 days [(5)]. The small colony form isolated is referred to as SC and its co-305 existing ancestral colony sized variant LC. 306 Additional long-term stationary phase experiments were performed to test repeatability of the 307 appearance of small colony forms. Every day, 1 ml aliquots were taken from each flask. 500ul of this 308 sample was used to prepare dilutions in 0.9% NaCl solution which were then plated on LB plates in 309 two technical replicates. The remaining 500ul solution was mixed with equal volume of 50% 310 glycerol and saved in -80 0 C as stocks. (Figure 1). The plates were incubated at 37 0 C for 16-18 hours 311 and the colonies were counted to track population changes over time and to check for 312 contamination. Images were taken with the help of Gel-Doc system. Putative small colonies, 313 whenever seen, were re-streaked to verify stability of the phenotype and a separate count of them 314 was maintained daily. 315 For measuring growth characteristics of various strains, 5 ml of LB was inoculated with single 316 colonies of respective strains and grown overnight to form the starter culture. This culture was 317 diluted 1000 fold in 30 ml LB in 100 ml conical flasks. Cultures were maintained at 37 0 C, with 318 circular shaking at 200 rpm to maintain aeration. Growth dynamics were tested in rich media (Luria 319 broth, Himedia), as well as minimal media (M9) with 0.5% glucose as carbon source. Bacterial 320 growth was measured by tracking change in optical density at 600 nm. 321

Calculation of maximum growth rate and lag time 322
To estimate the maximum growth rate, specific growth rates of each strain was calculated from 323 turbidimetric change for every 15 minutes over 20 hours. Based on the extent of initial noise a 324 sliding window of size 5-8 timepoints was taken for each strain and the median growth rate was 325 ascertained for each window. The element of the window showing maximum median was defined 326 as corresponding to the maximum growth rate for a specific strain. 327 To calculate time spent by each culture in lag phase the last point of lag was determined following 328 Bertrand (2019) [(39)]. Intersection of non-growing lag phase and exponentially growing log phase 329 was ascertained by fitting linear models to growth dynamicss. The time obtained from this 330 intersecting point was considered as the duration of the lag phase for a specific strain. 331