The rpoS Gene Is Predominantly Inactivated during Laboratory Storage and Undergoes Source-Sink Evolution in Escherichia coli Species

Abstract

The rpoS gene codes for an alternative RNA polymerase sigma factor, which acts as a general regulator of the stress response. Inactivating alleles of rpoS in collections of natural Escherichia coli isolates have been observed at very variable frequencies, from less than 1% to more than 70% of strains. rpoS is easily inactivated in nutrient-deprived environments such as stab storage, which makes it difficult to determine the true frequency of rpoS inactivation in nature. We studied the evolutionary history of rpoS and compared it to the phylogenetic history of bacteria in two collections of 82 human commensal and extraintestinal E. coli strains. These strains were representative of the phylogenetic diversity of the species and differed only by their storage conditions. In both collections, the phylogenetic histories of rpoS and of the strains were congruent, indicating that horizontal gene transfer had not occurred at the rpoS locus, and rpoS was under strong purifying selection, with a ratio of the nonsynonymous mutation rate (Ka) to the synonymous substitution rate (Ks) substantially smaller than 1. Stab storage was associated with a high frequency of inactivating alleles, whereas almost no amino acid sequence variation was observed in RpoS in the collection studied directly after isolation of the strains from the host. Furthermore, the accumulation of variations in rpoS was typical of source-sink dynamics. In conclusion, rpoS is rarely inactivated in natural E. coli isolates within their mammalian hosts, probably because such strains rapidly become evolutionary dead ends. Our data should encourage bacteriologists to freeze isolates immediately and to avoid the use of stab storage.

INTRODUCTION

The rates of recombination and mutation in a genome are critical issues because the diversity generated by these mechanisms is the substrate for selection. Some regions of bacterial chromosomes are more prone to mutation (1–4) or recombination (5) than other regions, due to physical or physiological constraints. Then, selection can favor diversification, for example, that of cell surface proteins such as beta barrel porins (6), and recombination events, which can restore particular functions, as in the case of the mismatch repair (MR) genes (7, 8).

There is growing evidence that selection also acts on central regulators to allow adaptation by rewiring of networks. In vitro experimental models of Escherichia coli evolution have shown that adaptation occurs through modifications of global regulatory networks, which are often interconnected (9–13). These networks control DNA superhelicity (12), the stringent response (12), and the general stress response, via RpoS (9, 10, 13). RpoS is an alternative sigma factor of RNA polymerase that regulates more than 10% of E. coli genes and plays a critical role in survival under conditions of exposure to several types of stress, including acid, heat, and oxidative stress (14, 15). A particular case of in vitro evolution is the inactivation of crl (which modulates the balance between different sigma factors in the RNA polymerase holoenzyme, particularly RpoS [16–18]) or rpoS (19–21), during stab storage of E. coli or Salmonella strains or during nutrient starvation, which favors the inactivation of the general stress response. Regulatory networks are also dynamic in vivo. In nature, a wide continuum of phenotypes is observed in isolates from individual patients suffering from severe E. coli extraintestinal infections due to a tradeoff between self-preservation and nutritional competence as a consequence of variable levels of RpoS (22). Interestingly, the rpoS gene is located on the lagging strand, where the rate of point mutations is higher than on the leading strand, probably due to conflicts between replication and transcription (3). rpoS is also located close to the MR gene mutS, in a region that has high genomic variability as a result of horizontal gene transfer (23, 24).

In collections of natural E. coli and Salmonella isolates originating from the environment or various hosts, either as commensal bacteria or as pathogens, the frequency of inactivating alleles of rpoS is very variable and ranges from less than 1% to more than 70% of strains (25–30). However, these data are difficult to interpret, because inactivation can depend on laboratory storage conditions, as discussed above. We sought to gain insight into the forces selecting for rpoS mutations; therefore, we studied the evolutionary history of the rpoS gene and compared it to the phylogenetic history of bacteria in two collections of 82 human commensal and extraintestinal E. coli strains. These strains were representative of the phylogenetic diversity of the species and differed only by their storage conditions.

MATERIALS AND METHODS

Strains.

Two collections of human natural E. coli isolates were studied, differing essentially by their storage history. The IAI collection, contains 15 commensal and 67 extraintestinal pathogenic strains (6 from blood samples, 53 from urine samples, and 8 from miscellaneous samples) isolated in the 1980s in various parts of France, including Paris, and is representative of the phylogenetic diversity of the E. coli species (31). These strains were sent to our laboratory in stab cultures and then stored in glycerol at −80°C. The strains were stored again in stab cultures for a brief period due to the relocation of the lab in the 1990s, followed again by storage at −80°C. Therefore, it can be considered that the strains of the IAI collection were subjected to five different subcultures plus two stab storage episodes during their time in the laboratory. The “natural isolates with low subcultures” (NILS) collection was obtained for this study and mirrors the IAI collection. The origins and numbers of strains are similar between the two collections, and the phylogroups are highly comparable. The strains were isolated during January 2013 from two clinical laboratories based in different teaching hospitals in the Paris suburbs (Colombes and Bobigny). The strains were directly sampled from antibiograms on Mueller-Hinton plates. These strains were subcultured only twice and were not subjected to stab storage prior to the extraction of genomic DNA and storage in glycerol at −80°C.

MLST, phylotyping, and O typing of the strains.

Multilocus sequence typing (MLST) was performed on the IAI collection with the concatenated complete sequences of six housekeeping genes (trpA, trpB, pabB, putP, icd, and polB) (6,023 bp) as described in reference 32. Phylogenetic tree reconstruction was performed with the maximum likelihood procedure (33). IAI and NILS strains were classified according to the seven main phylogroups of the species, and strains belonging to Escherichia clade I were identified as described in references 34 and 35. The clonal group A (CGA) was identified among the D strains as described in reference 36, and the B2 group strains were assigned to the main B2 subgroups (I to X) with MLST data for the IAI collection (32) and allele-specific PCR for the NILS collection (37). O typing was performed with an allele-specific PCR assay (38, 39).

Sequencing of the rpoS gene.

The complete coding sequence of the rpoS gene was amplified by PCR with flanking primers (rpoS forward ext, 5′-ACAAGGGGAAATCCGTAAACC-3′, and rpoS reverse ext, 5′-AGCCTCGCTTGAGACTGGCC-3′) and sequenced by classical Sanger sequencing technology.

Tree comparisons, Ka/Ks ratio, and haplotype diversity.

The congruence between gene trees was estimated with the TreeOfTrees method, where each phylogenetic tree is first transformed into a tree-distance matrix to recover the tree structure independent of branch length, as described in reference 40. With this approach, a high bootstrap support value separating two sets of gene trees allows incongruent sets of gene trees to be identified. A low bootstrap value suggests that either the two sets of gene trees are congruent or there is insufficient phylogenetic information to reject the hypothesis of congruence. The ratio of the nonsynonymous mutation rate (Ka) to the synonymous substitution rate (Ks) was estimated with the KaKs_Calculator software (41) and the Tamura and Nei model of substitution (42).

Haplotypes of the rpoS gene were defined as combinations of nucleotide variations (alleles) either affecting (nonsynonymous substitutions and insertions/deletions) or not affecting (synonymous substitutions) the amino acid sequence of RpoS. Haplotype diversity was estimated based on the diversity indexes D_S and α as described in reference 43 with N, the number of strains, and S, the number of different haplotypes.

RpoS-dependent phenotypes.

The data for the IAI collection, except for the catalase activity, were available (29), whereas the data for the NILS collection were generated during this work. RpoS activity was estimated by staining glycogen with iodine as described in reference 29, because glycogen synthesis is positively regulated by RpoS, via the glgS gene (44). Plates were incubated with bacteria for 24 h at 37°C and then left at 4°C for 20 h, after which they were flooded with unstabilized Lugol's solution (I₂ concentration = 10 g · liter⁻¹). The intensity of the brown color depends on the amount of RpoS present. The results are presented as the percent color intensity relative to two controls: an E. coli K-12 MG1655 ΔrpoS strain that remained light yellow (0%) and a positive control corresponding to an E. coli natural isolate expressing high levels of RpoS, which was stained dark brown (100%) (29). Experiments were performed three times, and the median value considered. RpoS activity was also assessed by testing H₂O₂ resistance. RpoS regulates genes important for oxidative stress resistance, including katE, which encodes the catalase enzyme hydroperoxidase II. Strains were grown overnight and then centrifuged to collect the cell pellet, which was washed in phosphate-buffered saline (PBS). A suspension of the tested strain at an optical density at 600 nm (OD₆₀₀) of 0.1 was then plated to obtain an initial bacterial count. H₂O₂ was then added to a 5 mM concentration, and the culture was incubated at 37°C with shaking for 60 min. H₂O₂ resistance is expressed as a percentage and corresponds to the number of CFU at 60 min divided by that at time zero (29). Lastly, a simple assay for measuring catalase activity using a visual approach was performed as described in reference 45, except for the heat treatment at 55°C. This assay is based on the principle that the oxygen bubbles generated from the decomposition of hydrogen peroxide by catalase are trapped by the surfactant Triton X and visualized as foam, from which the height is measured in the test tube.

Nucleotide sequence accession numbers.

The MLST and rpoS sequences are available in the GenBank sequence database (accession numbers KM885308 to KM885593).

RESULTS

Establishment of a collection of natural isolates mirroring the IAI collection but with no stab storage.

We sought to examine the frequency of rpoS inactivation in a collection of natural isolates; therefore, we established a collection of 82 human commensal and extraintestinal pathogenic E. coli strains subjected to a low number of subcultures and no stab storage (see Table S1 in the supplemental material). We called this collection the “natural isolates with low subcultures” (NILS) collection. These strains originated from samples similar to those used to derive the IAI collection, and the distributions of the main phylogroups (Table 1) and O types (see Tables S1 and S2) were comparable between the two collections. Of note, due to the recent spread of the ST131 clone, which belongs to B2 phylogroup/subgroup I (46), five O25b strains and one O16 ST131 strain (47) were found in the NILS collection whereas no subgroup I strain was present in the IAI collection (see Table S2). None of the collections possessed strains of the rare subgroup V.

TABLE 1.

Origins and phylogroup distributions within the IAI and NILS collections

Condition of isolation (no. of strains)	Origin(s) (no. of strains)	Phylogroup	No. of strains of the indicated phylogroup in:
			IAI collection	NILS collection
Commensal (15)	Feces (15)	A	11	11
		B1	2	1
		C	1	1
		D	0	1
		E	1	1
Pathogen (67)	Blood (6)	A	2	2
		B2	3	3
		D	0	1
		F	1	0
	Urine (53)	A	11	7
		B1	1	2
		B2	28	28
		C	4	4
		D	3	8
		E	1	1
		F	3	2
		Cl Ia	2	1
	Miscellaneous (8)	A	2	1
		B1	0	1
		B2	5	5
		D	0	1
		F	1	0

^aCl I, Escherichia clade I.

The phylogenetic history of rpoS is congruent with the phylogenetic history of the strains in both collections.

We first assessed the phylogenetic history of the IAI strains by performing MLST with six housekeeping genes that exhibit a strong phylogenetic signal but a low level of recombination (48). The IAI collection is particularly diverse and contains strains belonging to the main phylogenetic groups, i.e., A, B1, B2, C, D, E, and F (39), as well as Escherichia clade I, which is the most closely related Escherichia clade (35) (Fig. 1A). All subgroups were represented among the B2 phylogroup (32), except subgroups I and V (Fig. 1A). The obtained phylogeny was consistent with the phylogeny of the E. coli species (49). We then assessed the phylogenetic history of the rpoS gene (Fig. 1B). The phylogenetic signal was weak due to the short length of the gene. Nonetheless, the Escherichia clade, B2, D, F, and A/B1/C/E groups could be clearly delineated, except for the D phylogroup IAI24 strain, which clustered with the F strains. The clustering of A/B1/C/E strains can be easily explained by the close relationship between these phylogroups. The B2 subgroups could be distinguished within the B2 group strains (Fig. 1B). Thus, visual inspection of the two trees in Fig. 1 reveals that they have similar topology, i.e., they are congruent. We used a recently developed technique, TreeOfTrees, to estimate more quantitatively the level of congruence between the trees (Fig. 2). The comparison between sets of trees is based on several tree metrics, which produced a unique tree labeled by the genes of each tree. The robustness values of the edges of this tree can be used to detect genes with different evolutionary histories. We compared gene trees of each of the six MLST genes, the combined MLST tree, and the rpoS tree. All the bootstrap values were low enough (below 80%) to show that all the gene trees were congruent. Therefore, this statistical bootstrap resampling method confirms the observations made from the visual inspection of the trees in Fig. 1. We sequenced the rpoS gene in the NILS collection (see Fig. S1 in the supplemental material) to verify that this congruence between species and rpoS phylogeny was not limited to the IAI collection, and we divided the strains into the main phylogroups and D and B2 subgroups by PCR-based approaches (32, 35–37). Once again, we observed very good congruence between the rpoS gene and phylogroups and subgroups. Of note, strains from subgroup I, including ST131 O25b and O16 types, belong to a specific cluster in the rpoS tree (see Fig. S1).

FIG 1.

Phylogenetic trees of the 82 E. coli strains from the IAI collection reconstructed with PHYML (33) from the complete sequences of six housekeeping genes (6,023 bp) (A) and the complete rpoS sequences (993 bp) (B). The trees were rooted on Escherichia clade I strains IAI32 and IAI45. All bootstrap values above 70% are shown. The phylogroups, the B2 subgroups (SG), and clonal group A (CGA) are indicated.

FIG 2.

Tree representation of the distance matrix between gene tree structures. Gene tree structure comparisons were between trees based on rpoS, six individual housekeeping genes (trpA, trpB, pabB, putP, icd, and polB), and concatenation of the six housekeeping genes (MLST). Numbers are bootstrap values.

Thus, there was high congruence between the MLST and rpoS trees in both IAI and NILS collections, with no detectable horizontal gene transfer affecting rpoS phylogeny. This indicates that the strains and the rpoS gene share a similar evolutionary history.

The frequency of RpoS inactivation depends on the collection studied.

The nucleotide sequencing identified 54 rpoS haplotypes (combinations of alleles affecting or not the amino acid sequence of RpoS) in the IAI collection (29) and 26 in the NILS collection. We translated the nucleotide sequences of rpoS and analyzed the amino acid sequences of the various haplotypes. In the IAI collection, 41 strains (50%) had rpoS alleles encoding proteins with amino acid sequence variation, of which 26 (63.4%) corresponded to a truncated protein. Among these 41 strains, only two strains (IAI15 and IAI17) had the same amino acid (and nucleotide) sequence. However, only one strain in the NILS collection (1.2%) possessed an rpoS allele encoding a variant protein: the NILS50 strain, which had a frameshift allele (c.92_126del, p.V31SfsX12) (Fig. 3). Overall, there were 40 haplotypes leading to amino acid variations in the IAI collection but only one haplotype in the NILS collection. The remaining haplotypes (14 and 25 for the IAI and NILS collections, respectively) did not affect the amino acid sequence of RpoS.

FIG 3.

Diagrammatic representation of the RpoS alleles encoding proteins with amino acid variation identified in 42 strains among the 164 studied from the E. coli IAI and NILS collections. Black regions represent RpoS structural domains σ₁ to σ₄ (64). The mutations are shown at scale all along the protein (above, truncating alleles; below, nontruncating alleles). Each mutated strain is represented with a symbol defined in the boxes.

The truncating mutations were not randomly distributed along the gene and were clustered toward the 5′ end of the gene. In total, 59% of these mutations were located in the upstream fifth of the gene (versus 20% and 0% of the nontruncating and missense mutations, respectively). Thus, only a small part of the protein was translated in most cases, suggesting that the total inactivation of the protein is favored.

We used iodine staining and measured the H₂O₂ resistance and catalase activity of the strains to verify that RpoS was functionally inactive in the mutated strains. We first noted that all phenotypes used to assess the functionality of RpoS were correlated with Spearman's rho correlation coefficients ranging from 0.3 to 0.45 (P < 0.01) (data not shown). Iodine staining was significantly higher in wild-type strains than in strains possessing nontruncating or truncating alleles of rpoS (65% versus 17% and 11%, respectively), and the same was true of H₂O₂ resistance (81% versus 10% and 3%, respectively) and catalase activity (9 versus 6 and 5 mm of foam height, respectively) (Wilcoxon test, P < 0.05) (Fig. 4). Lastly, the median value of iodine staining in the NILS collection (72%) was significantly higher than that in the IAI collection (25%) (P < 0.01). Similarly, strains in the NILS collection were also more resistant to H₂O₂ stress than strains in the IAI collection (median values, 85% and 22%, respectively) (P < 0.01) and had a greater catalase activity (median values, 10.5 and 6, respectively) (P < 0.01).

FIG 4.

RpoS-dependent phenotypes (iodine staining [A], H₂O₂ resistance [B], and foam height formed under H₂O₂ stress [C]) of the 164 E. coli strains studied, presented as box plots according to the rpoS sequence. WT, NTA, and TA correspond to wild-type rpoS and nontruncating alleles and truncating alleles of rpoS, respectively. The asterisks indicate significant differences between groups of strains (Wilcoxon test, P < 0.05).

Collectively, these data indicate that RpoS is differentially inactivated in the two collections and is frequently inactivated in the IAI collection and almost never in the NILS collection.

The inactivation of rpoS is a recent phenomenon under selection.

One way to identify the selective pressure acting on a protein-coding gene is to study the Ka/Ks ratio, where Ka is the number of nonsynonymous substitutions per nonsynonymous site and Ks is the number of synonymous substitutions per synonymous site. A ratio greater than 1 implies positive selection, a ratio less than 1 implies purifying selection to conserve protein sequence, and a ratio of 1 indicates neutral selection. We estimated the Ka and Ks in both collections, excluding the truncating alleles and the deletions of 7 and 32 amino acids (IAI58 and IAI70 strains, respectively). Ks values were similar in both IAI (54 strains) and NILS (81 strains) collections (mean values, 0.026 and 0.030, respectively); however, Ka was different and was higher in the IAI collection than in the NILS collection (mean values, 0.001 and 0, respectively; Mann-Whitney test, P < 0.0001) (Fig. 5). Consequently, the Ka/Ks ratios were significantly different between the two collections (0.043 in the IAI collection and 0 in the NILS collection; Mann-Whitney test, P < 0.0001).

FIG 5.

Ka (top panels) and Ks (bottom panels) values of the IAI (n = 54, left panels) and NILS (n = 81, right panels) collection strains, excluding the truncating alleles and the deletions of 7 and 32 amino acids (IAI58 and IAI70 strains, respectively). The ordinate corresponds to the percentage of sequence similarity determined by pairwise comparison.

We then examined whether pseudogene degeneration had occurred in the rpoS genes inactivated by truncating variations and in the two alleles with large amino acid deletions (i.e., 29 strains) by looking at the Ka/Ks of these alleles (these variations in the sequence were excluded). Due to their nonfunctional nature, pseudogenes are not expected to be under evolutionary constraint and are instead expected to be under neutral selection. Thus, pseudogenes should have Ks roughly equal to the Ka. However, these strains had a Ka of 0 and a Ks of 0.029, similar to the Ks of the strains without truncating variations, suggesting that these genes had not undergone degeneration to pseudogenes.

Thus, the rpoS gene appears to be under strong purifying selection. However, this purifying selection is counterbalanced in the IAI collection by positive selection because the Ka is higher in this collection than in the NILS collection. No pseudogene degeneration was observed in rpoS alleles inactivated by truncating variations (Ka = 0, and similar Ks values for truncating and nontruncating alleles), suggesting a recent evolutionary origin for these alleles.

The rpoS gene undergoes “source-sink” evolution.

A source-sink model was recently used to describe the adaptation of bacterial pathogens (50), which undergo continuous switching between permanent (source or reservoir) and transient (sink or virulent) habitats. In this model, the sink-adaptive alleles, which are deleterious in the source environment, continuously emerge and become extinct. An example of a gene in the E. coli species undergoing such evolution is the gene coding for fimH adhesin (51). We hypothesized that rpoS is characterized by source-sink dynamics, whereby the rpoS wild-type allele is favored in the source environment and the rpoS-inactivating alleles are favored in the sink environments. We first built an unrooted maximum likelihood tree from the rpoS haplotypes of the 135 strains of the IAI and NILS collections studied above (excluding strains with truncating alleles and large amino acid deletions). We observed characteristic source-sink dynamics with the existence of a pool of stable haplotypes (with accumulated silent variation) corresponding to the source or primary zone and a series of recently derived haplotypes with inactivating alleles (without silent variation) corresponding to the sink or external zone (50) (Fig. 6). We then used various diversity indexes (43) to show that zonal diversity was different between the source (D_S = 11.5, D_S/S = 0.4, α = 10.7) and the sink (D_S = 13.0, D_S/S = 1.0, α = infinity because all the haplotypes are identical) habitats.

FIG 6.

Unrooted maximum likelihood DNA haplotype tree for the rpoS gene from the 135 E. coli strains without truncating alleles or those containing large deletions. For the sake of clarity, haplotypes belonging to the same clade are grouped in boxes, and the box size is proportional to the number of haplotypes within the clade which is indicated, except when the clade includes only one haplotype. The primary zone (source) is separated from the external zone (sink) by the dotted line. Each nonbold branch represents synonymous variation(s), whereas each bold branch includes one nonsynonymous variation. Note that all the strains in the source have the same amino acid sequence whereas the strains in the sink all have different amino acid sequences.

The molecular evolutionary footprint of rpoS is thus typical of a gene exhibiting source-sink dynamics.

DISCUSSION

The mutS-rpoS region is a striking region of the E. coli chromosome because it encompasses two major genes implicated in the adaptation of strains: mutS, a mismatch repair gene subjected to second-order selection (52), and rpoS, the global stress regulator gene (14). This region is a hot spot of both phylogenetic incongruence (5) and gene acquisition and loss (53–55), which indicates that a high rate of horizontal gene flow occurs at this locus between strains of the E. coli species. In addition, many studies have documented the inactivation of the rpoS gene not only in vivo but also in vitro, especially during the laboratory storage of strains in stab culture. However, few studies had investigated the role of mutations and recombination in the diversification of the rpoS gene, as well as the selective pressures acting on it (56). We decided to study the evolutionary history of two collections of human commensal and extraintestinal pathogenic E. coli strains differing only by their storage conditions, to eliminate ambiguities resulting from previous works where the storage conditions of the strains were often unknown. Three main conclusions can be derived from our work.

First, we found that the phylogenetic histories of the strains and of rpoS are similar (Fig. 1), indicating that rpoS itself does not undergo horizontal gene transfer. The region between mutS and rpoS is variable in length, ranging from 3 to 9 kb, and mutS is highly subject to recombination (7, 8), but rpoS appears as an anchor point bordering the recombinant region.

Second, in the NILS collection, rpoS is under strong purifying selection and has a Ka/Ks ratio substantially smaller than 1, with no nonsynonymous substitutions. This high conservation of amino acid sequence has been reported for a small set of intraintestinal pathogens (55) and indicates selective pressure for the absolute conservation of protein sequence. RpoS, a sigma subunit of RNA polymerase, is a master regulator of the general stress response of the cell and controls around 200 genes (14). Therefore, it interacts with several targets such as proteins and specific DNA sequences, and these interactions probably necessitate a conserved structure to ensure the function of RpoS and hence optimal cell fitness. Proteins that interact with a large number of molecules evolve slowly and are subject to many pleiotropic constraints that limit sequence divergence (57).

Third, we observed a clear difference between the IAI and NILS collections in terms of the frequency of inactivating alleles, either truncating or nonsynonymous. Indeed, the frequency of such alleles was high in the IAI collection, which has been subjected to stab storage, and there was almost no amino acid variation in the NILS collection, which was studied directly after the isolation of the strains from the host. We verified the functional role of the nonsynonymous and truncating variants by studying iodine staining, H₂O₂ resistance, and catalase activity phenotypes and showed a clear genotype-phenotype correlation (Fig. 4). The absence of pseudogene degeneration suggests that the truncating alleles occurred at a recent point on the evolutionary time scale. The nutrient-limited environment in stab culture is an ideal environment for the selection of rpoS inactivation. For example, there is a strong selection for rpoS mutants in succinate minimal medium (58), because RpoS inhibits succinate dehydrogenase in the tricarboxylic acid cycle (59). In natural hosts, the inactivation of rpoS occurs during the course of intestinal colonization in mice (60), and the expression of rpoS is downregulated in microevolved isolates during human extraintestinal infections (22). Similarly, the selection of E. coli isolates with inactivated rpoS alleles could be favored in the environment because soil and sediments are characterized by low nutrient availability (26). Studies similar to that reported here, but with environmental isolates, should be performed to investigate this issue. However, strains harboring inactive rpoS alleles are probably evolutionary dead ends in the long term. They can be considered transient niche specialists that are counterselected during the natural life cycle of E. coli in their hosts by particular conditions, for example, the acidic stress that they encounter in the stomach when they are ingested. These evolutionary dynamics of rpoS are best explained by the source-sink model (Fig. 6), which was originally developed for the population ecology of animals and plants and recently proposed for bacteria (50).

In conclusion, our work will help to resolve the current debate in the literature about the frequency of rpoS inactivation in E. coli. The rpoS gene is rarely mutated in natural E. coli isolates within their mammalian hosts (either commensal bacteria or pathogens). Strains that do possess a mutated rpoS gene are short-lived, probably because such strains rapidly become evolutionary dead ends. However, laboratory storage of strains in stab cultures selects for rpoS inactivation, because this provides a specific advantage in this nutrient-limited environment (61). From a practical point of view, this observation should encourage bacteriologists to make fewer subcultures and stab cultures and should instead promote the immediate freezing of isolates. This will prevent erroneous phenotyping of the strains because RpoS regulates several phenotypes, some of which are medically relevant, including biofilm development, bacterial persistence, and antibiotic resistance (62, 63).