Evolution of pH-sensitive transcription termination in Escherichia coli during adaptation to repeated long-term starvation

To gain insight into how microbes adapt to recurrent nutrient resource fluctuations following prolonged periods of famine, we evolved 16 E. coli populations in parallel to a RLTS regime of 100-d feast/famine cycles for a total of 900 d (64). Every 100 d, we transferred 1 mL of the aged culture into 9 mL of fresh LB-Miller broth, replenishing each culture’s resources. At this time, pretransfer aliquots of the populations were frozen to maintain a historical record of the evolution experiment. Growth in LB broth is supported through the catabolism of amino acids and leads to alkalinization of the medium within 24 h which remains between pH 8 and pH 9 until additional resources are provided (21, 22, 66). As such, E. coli cultures are subjected to resource-depleted, alkaline conditions for the vast majority of the RLTS regime, essentially 99 d (SI Appendix, Fig. S1A). Every 100 d, when cultures are replenished with resources, the medium pH decreases dramatically from approximately pH 9 to pH 7, before slowly returning to an alkaline pH over the next 24 h. Thus, populations adapted to these conditions must not only survive extended starvation under alkaline conditions but also cope with the sudden downward shift in pH and repletion of resources concurrently.

Metagenomic sequencing throughout the 900-d experiment revealed that approximately half (7/16) of the populations evolved fixed mutations in rho, which encodes the Rho-dependent transcription termination factor (Fig. 1A and SI Appendix, Fig. S2A). The majority of rho mutations (5/7) occurred in the first half of the RLTS experiment and swept to fixation within the first 400 d (Fig. 1B). By 700 d, rho mutations were fixed in a total of seven populations after which time no additional rho mutations were detected in the remaining nine populations. All identified rho mutations occur within the primary RNA-binding site of Rho. Thus, we suspect these mutations are likely to perturb interactions between Rho and its RNA target (Fig. 1 C and E and SI Appendix, Fig. S3). Furthermore, all five rho mutations that are fixed within the first 400 d result in an R109H nonsynonymous substitution, suggesting it may contribute to adaptation in RLTS conditions. After reconstructing the rho R109H allele into experimental ancestor background, we found that the R109H allele did not confer any considerable growth effects in conditions that previously produced other experimentally evolved rho mutations, including heat stress (56), exposure to ethanol (61), and osmotic stress (67) (SI Appendix, Fig. S4). This lack of phenotype is consistent with Rho’s role as a pleiotropic “master-regulator” (68) and indicates that any phenotypes associated with the rho R109H allele may be contingent on specific conditions that occur during RTLS or dependent on an earlier arising mutation.

Since prior investigations implicate the evolution of arginine to histidine substitutions as a response to pH (40, 41, 69, 70), and histidine residues can serve as pH sensors (39, 71, 72), we refocused our investigation to include genes shaped by positive selection during RTLS and associated with pH homeostasis. We found that all populations with mutant rho alleles also contain mutations in the recently annotated ydcI gene, which encodes a LysR-type transcriptional regulator reported to affect the expression of around 60 genes in E. coli, some of which are associated with the pH response (65, 73) (Fig. 1 A and B and SI Appendix, Fig. S2B). Mutations in ydcI are pervasive and evolved in the vast majority (13/16) of RTLS populations (SI Appendix, Table S1). In each of these cases, the ydcI mutation arose simultaneously or before the rho mutation, suggesting that mutations in ydcI may be necessary for the fixation of mutations in rho.

In contrast to the localized mutations identified in rho, the mutations in ydcI are diverse, ranging from putative loss of function mutations—caused by IS-element insertions and small indels—to nonsynonymous SNPs whose functional consequences on YdcI activity are currently unknown (Fig. 1D). Nonsynonymous mutations in ydcI tend to cluster into three general genic regions. Using previous structural data for LysR-type proteins and the AlphaFold predicted structure for YdcI, we mapped the locations of these mutations to specific domains to gain insight into their potential functional impact (Fig. 1F). The affected residues R10 and R30 are found in the predicted helix–turn–helix domain and could lead to an abolishment of DNA-binding activity. Residues F74, D83, and S91 are all found in the α4 linker helix, and residues V181 and L198 are present in the junction between the two regulatory domains, regions that are critical to dimerization and effector binding, respectively, in well-studied LysR-type proteins (74). Together, this suggests that mutations resulting in loss or attenuation of YdcI function are under positive selection in RLTS conditions, and disruption of YdcI function may be necessary for the rho R109H allele to be beneficial.

Previous investigations have identified arginine to histidine substitutions that alter protein functionality in alkaline conditions (41, 70). Thus, we hypothesized that the activity of Rho^R109H may differ from that of WT Rho (Rho⁺) under alkaline conditions. To this end, we performed in vitro assays evaluating the capability of Rho to function as a helicase, terminate transcription, and bind RNA in neutral (pH 7) and alkaline (pH 9) conditions (Fig. 2 and SI Appendix, Fig. S5). At pH 7, we found no significant differences in the helicase activity (t test; P = 0.829) or transcription termination patterns of Rho^R109H and only modest differences in RNA affinity (t test, P = 0.033) compared to Rho⁺. By contrast, at pH 9, Rho^R109H exhibits significant reductions in its helicase activity (t test; P = 5.54 × 10⁻⁵), a complete inability to terminate transcription, and a significant decline in RNA affinity (t test, P = 0.003), while Rho⁺ remained functional. Moreover, Rho⁺ binding to RNA is largely unaffected by the change in pH conditions (t test, P = 0.068), while there is a significant reduction in Rho^R109H RNA binding as pH increases (t test, P = 0.004). These data collectively demonstrate that the activity of the Rho^R109H mutant is similar to the WT protein under neutral conditions, but is strongly reduced under alkaline conditions due, in part, to compromised RNA binding.

Of the transcripts where termination is Rho dependent, a fraction of termination events require an additional transcription-termination factor, NusG (47, 49), such as transcription termination that represses toxic prophage genes (46). These NusG-stimulated terminators have been shown to contain distinct rut features (47, 49) and are less sensitive to two disruptive primary binding site mutations, Rho^F62S and Rho^Y80C, than canonical terminators (75). To test the possibility that the R109H mutation may differentially impact the two classes of terminators, we conducted in vitro transcription experiments with two DNA templates encoding the NusG-stimulated terminators located within aspS and yaiC (49). With Rho⁺, termination at the aspS and yaiC sites is strongly stimulated by NusG at both pH 7 and pH 9 (SI Appendix, Fig. S6). Termination with the Rho^R109H mutant is also stimulated by NusG at pH 7, albeit to a lesser extent than Rho⁺. At pH 9, however, Rho^R109H can no longer trigger termination at the aspS and yaiC sites, even in the presence of NusG. These data indicate that pH regulates the activity of the Rho^R109H mutant at both NusG-stimulated and canonical Rho-dependent terminators (Fig. 2B).

The finding that Rho^R109H displays altered activity under alkaline conditions implies there must be changes in intracellular pH (pHi) to reveal the pH-dependent phenotype. To ascertain whether the introduction of the rho R109H and/or ΔydcI alleles results in an alteration of pHi homeostasis, we measured pHi across a range of pH-adjusted buffers (pH 6.92 to 9.59) by staining cells with BCECF-AM (76–79) and compared their pHi response to that of WT cells. As the intensity of the fluorescent signal emitted from BCECF-stained cells depends on pH (78), we quantified the pHi of single cells with fluorescent microscopy.

Across the environmental pH conditions, WT and rho R109H single-mutant cells exhibited similar pHi values with a fitted range from 6.7 to 9.05 and 7.04 to 8.96, respectively (Fig. 3 and SI Appendix, Figs. S8 and S9). Alternatively, cells containing a single ΔydcI mutation showed a reduced pHi range of 7.48 to 8.92, as ΔydcI cells maintain a higher pHi than WT cells in acidic and neutral conditions (Fig. 3 and SI Appendix, Fig. S10). This effect is amplified in the double mutant rho R109H/ΔydcI cells which exhibit a pHi range of 7.53 to 9.67, maintaining a significantly elevated pHi across the entire range of environmental pH conditions compared to the WT (sign test; n = 8; P = 0.008) (Fig. 3C and SI Appendix, Fig. S11). Because pHi elevation in neutral conditions is only realized in mutant backgrounds containing ΔydcI, it is likely that double mutant rho R109H/ΔydcI cells constantly maintain an intracellular pH environment that continuously manifests the pH-sensitive Rho termination activity via reduced RNA binding.

Considering the frequent co-occurrence of the rho R109H and ΔydcI alleles in evolved populations and the necessity of both alleles for a significant alteration in pHi response, we hypothesized that the benefit of carrying the rho R109H allele could only be present in a ΔydcI background. To address this, we “replayed” one cycle of the RLTS regime by growing cultures from different starting genotypes (WT, rho R109H, ΔydcI) for 100 d and sequenced the rho and ydcI loci from the aged cultures (Fig. 4A and SI Appendix, Fig. S12). We found that when starting from a rho R109H genotype, 8/12 populations evolved putative ΔydcI mutations which swept to fixation within 100 d. Alternatively, when starting from a ΔydcI genotype, only 3/12 populations evolved a rho R109H mutation, and only one of these rho R109H mutations had fixed. Importantly, when starting from a WT genotype, we did not observe the emergence of the rho R109H allele in a population without also detecting a fixed mutation in ydcI. Thus, ydcI loss-of-function mutations are recurrent in RLTS conditions and the appearance of the rho R109H mutation is unlikely to occur without a prior mutation in the ydcI gene.

Identifying the precise conditions under which each allele confers an advantage poses a challenge due to the continual evolution of the strains and the rapid accumulation of mutations as cultures age. While this is evident after “replaying” the evolution experiment, it may also account for the lack of significant differences in population cell density between strains over 100 d (SI Appendix, Fig. S1B). Consequently, we opted to conduct relevant phenotyping experiments over shorter periods to minimize any confounding effects potentially associated with the strains’ evolution. To mimic conditions experienced by populations during the RLTS regime, we performed pairwise competitions between each mutant and the WT ancestor in unbuffered, buffered pH 9, and buffered pH 7 LB medium for 1 and 14 d (Fig. 4B). When competing the WT ancestor against the rho R109H strain, we observe an early benefit to carrying the mutant rho allele after the first day of coculture, particularly in the pH 9 medium. However, after 14 d of coculture the WT ancestor strongly outcompetes the rho R109H strain in the unbuffered, buffered pH 9, and buffered pH 7 media. In contrast to rho R109H, ΔydcI appears to exhibit considerably greater selection rates across the 14 d as the ΔydcI strain outcompetes the WT ancestor at both time points and in all of the tested conditions. Notably, when the rho R109H allele is present in a ΔydcI background, we no longer observe highly negative selection rates after 14 d. Instead, the double rho R109H/ΔydcI mutant strain outcompetes the WT strain under the unbuffered and buffered pH 9 conditions and maintains a high selection rate throughout the experiment, while in unbuffered pH 7 conditions it is neutral in coculture with the ancestor. These results indicate that the rho R109H allele is advantageous when resources are plentiful, especially in alkaline conditions, and disrupting the ydcI gene can mediate any tradeoffs associated with the rho R109H allele in resource-limited long-term stationary phase conditions.

In addition to assessing growth and survival in coculture, assessment of each strain in monoculture revealed no significant differences in the mutants’ survival over 14 d. The end of the stationary phase and subsequent onset of the death phase marks a considerable decline in population cell viability, which we consistently observe on day 3 of growth across all strains (80) (Fig. 4C and SI Appendix, Fig. S1B). Next, the transition into long-term stationary phase is characterized by a resurgence in population density following the death phase, attributed to the substantial availability of dead cells providing resources for the surviving population (80). This resurgence in population cell density is highly repeatable, occurring on the 6th day of growth for the WT ancestor, rho R109H, and ΔydcI strains (Fig. 4C). Notably, rho R109H/ΔydcI populations exhibit a resurgence a day earlier on day 5 of growth. This early population resurgence following the death phase highlights a unique advantage of rho R109H/ΔydcI as these cells can better utilize new resources as they become available, such as necromass and other metabolites released by dying cells.

Given the widespread presence of Rho across the bacterial domain and its essential role in many prokaryotes (81), we investigated the prevalence and distribution of a histidine residue at analogous positions within Rho (Fig. 5). Our initial search of the RefSeq database did not return any bacterial sequences with histidine at this location. However, a deeper search using PHI-BLAST identified several sequences from species isolated from diverse environments that harbor a histidine residue at positions analogous to the E. coli Rho R109. Many of these isolates originated from alkaline environments or those which fluctuate between alkaline and neutral conditions. The soil bacterium Inquilinus limosus was isolated from a region that reports alkalization of the soil (82, 83), while the Poribacteria genome was identified as part of a microbial mat in hypersaline, alkaline water (84, 85). The pathogen Bartonella bacilliformis, the causative agent of Carrion’s disease, a neglected tropical disease, also encodes a histidine residue in Rho that may confer a pH-sensing phenotype as the pathogen alternates between neutral human blood and the alkaline midgut of sand flies (25, 86). The remaining species are isolated from aquatic environments with reported pH gradients. Planctomycetota was identified in sediment from the Auka marine hydrothermal vent (pH 6.5) in the Pacific Ocean (pH 8) (87), while Elusimicrobia was isolated from groundwater with a pH ranging from pH 6.5 to 8.5 (88, 89). Together, the absence of a histidine residue in the analogous position 109 of Rho in various lab-adapted bacterial reference strains, as well as the identification of Rho H109 residues in isolates that experience alkaline environments, is consistent with our results showcasing that this allele provides beneficial effects in alkaline environments.

All bacterial strains originate from PFM2, a prototrophic derivative of E. coli MG1655 K-12 that was provided by Dr. Patricia Foster (Indiana University) (103). The rho R109H substitution was cloned into the PFM2 ΔaraBAD background using the Church protocol (104). The ΔydcI and rho R109H/ΔydcI mutants were generated by moving the ydcI723(del)::kan cassette from the Keio collection strain JW5226 (105) to the PFM2 WT and rho R109H strains, respectively, using P1vir transduction (106). Mutations were confirmed via PCR and Sanger sequencing. The rho R109H mutation was confirmed using a forward primer sequence of 5′-GGTGATGGCGTACTGGAGAT-3′ and a reverse primer sequence of 5′-GTGCCACAATCAGACCACG-3′. For the ydcI locus, mutations were confirmed using the forward primer 5′-CGGCACGATTAACTGAGGCCGT-3′ and reverse primer 5′-CGCCGTGATGCTGTTCGCCAA-3′.

Mutations in rho and ydcI were initially detected during the analysis of longitudinal metagenomic sequencing of 16 E. coli populations that were experimentally evolved in 100-d feast/famine cycle conditions (64). These populations originate from four different genetic backgrounds: PFM2; PFM2, ΔaraBAD; PFM2, ΔmutL; or PFM2, ΔaraBAD, ΔmutL. The deletion of the ΔaraBAD locus serves as a neutral marker that enables the detection of cross-contamination between lines during experimental evolution by plating on TA agar (10 g/L tryptone, 1 g/L yeast extract, 5 g/L NaCl, 16 g/L agar, 10 g/L L-arabinose, and 0.005% tetrazolium chloride) (107), while the deletion of ΔmutL disrupts methyl-directed mismatch repair and allows for the examination of the evolutionary process in a mutator phenotype. The relevant genotypes of populations adapted to 100-d feast/famine cycles can be found in SI Appendix, Table S1.

The quaternary structure of YdcI was predicted with SWISS-MODEL from an alpha-fold predicted structure for YdcI (P77171) on UniprotKB. As YdcI is annotated as a LysR-family transcriptional regulator, which are typically homotetramers (108, 109), we focused our search on the homology options in the SWISS-MODEL repository that were homotetramers and based on OxyR—another member of the LysR family (templates: 4 x 6g.1.A and 6g1d.1.A). Residues that acquired nonsynonymous mutations throughout experimental evolution were highlighted in PyMol. For the structure of Rho, the most recent and most detailed cryoEM reconstruction of an E. coli Rho–RNA complex was used (#8E6W), while PyMol was used to highlight features of interest.

Plasmid pET28bR109H for overexpression of the R109H mutant was obtained by site-directed mutagenesis of plasmid pET28bRho (kindly provided by Pr. James Berger, Johns Hopkins University) using the NEBuilder HiFi kit (New England Biolabs). The R109H mutant was overexpressed in BL21(DE3)pLysS cells carrying the pET28bR109H plasmid and purified as described for WT Rho (110).

Duplex unwinding experiments were performed as described (110) with minor modifications. Briefly, 5 nM RNA:DNA duplexes bearing a ³²P label and a rut site [substrate C (111)] was incubated for 5 min at 37 °C with 20 nM Rho hexamers in helicase buffer (150 mM potassium acetate and 20 mM of MES, HEPES, or EPPS at the indicated pH). Then, 1 mM MgCl₂, 1 mM ATP, and 400 nM Trap oligonucleotide (complementary to the DNA strand of the duplexes) were added to the reaction mixture before incubation at 30 °C. Reaction aliquots were withdrawn at various times, quenched with two volumes of quench buffer (20 mM EDTA, 0.5% SDS, 150 mM sodium acetate, and 4% Ficoll 400), and analyzed by 9% polyacrylamide gel electrophoresis (PAGE) and phosphorimaging (Typhoon FLA-9500 instrument and ImageQuant TL v8.1 software, Cytiva).

Transcription termination experiments were performed as described (112) with minor modifications. Briefly, DNA template encoding the λtR1 terminator (0.1 pmol), E. coli RNA polymerase (0.4 pmol; New England Biolabs), Rho (1.4 pmol), and SUPERase-In (0.5 U/µL; Ambion) were mixed in 18 µL of transcription buffer (150 mM potassium acetate, 5 mM MgCl₂, 1 mM DTT, 0.05 mg/mL BSA, and either 20 mM HEPES, pH 7.0, or EPPS, pH 9.0) and incubated for 10 min at 37 °C. Then, 2 µL of initiation mix (2 mM ATP, GTP, and CTP, 0.2 mM UTP, 2.5 µCi/µL of ³²P-aUTP, and 250 µg/mL rifampicin) were added to the mixtures before further incubation for 20 min at 37 °C. Reactions were quenched with 70 mM EDTA, 0.1 mg/mL tRNA, and 0.3 M sodium acetate, phenol extracted, concentrated by ethanol precipitation, and then analyzed by denaturing 8% PAGE and phosphorimaging.

Rho–RNA dissociation constants (K_d) were determined using a filter-binding assay, as described (113). Briefly, ~10 fmoles of ³²P-labeled RNA substrate (RNA strand from substrate C) were mixed with various amounts of Rho in 100 µL of binding buffer (0.1 mM EDTA, 1 mM DTT, 150 mM potassium acetate, 20 µg/mL BSA, and either 20 mM HEPES, pH 7.0, or 20 mM EPPS, pH 9.0). After incubation for 10 min at 37 °C, the samples were filtered through stacked [top] nitrocellulose (Amersham Protran) and [bottom] cationic nylon (Pall Biodyne B) membranes using a Bio-dot SF apparatus (Bio-Rad). The fractions of free and Rho-bound RNA (retained on the nylon and nitrocellulose membranes, respectively) as a function of Rho concentration (expressed in hexamers) were then determined by phosphorimaging of the membranes.

Intracellular pH was determined using the pH-sensitive fluorescent dye BCECF-AM. In brief, the fluorescence of this dye depends on the environmental pH, with high pH environments showing increased levels of fluorescence. It is a ratiometric dye, meaning that we utilize the ratio of fluorescence emission (525 nm) when excited at two different wavelengths (470 nm and 440 nm). Because the relative intensities in these two channels depend on both the imaging conditions (light intensity, emission filters, exposure time, etc.) and the pH, the conversion from the ratio of intensities to pH units needs to be calibrated utilizing the same approach as the experimental collection. Here, we used the ionophore nigericin as a way to collapse the pH gradient between the inside of control cells and their external buffered environment. This allowed us to measure intensities ratios and build a conversion function from intensities to pH while the BCECF dye was still in the presence of cellular components.

Cultures of each experimental strain, including an additional WT culture used for calibration of ratiometric fluorescence calibration (control), were grown overnight in LB broth. Due to the significant autofluorescence exhibited by LB broth, it could not be utilized for this microscopy-based approach. To address this issue, we thoroughly washed and resuspended cells from the overnight cultures in Ringer’s buffer, which is specifically formulated to minimize osmotic stress and commonly used to wash cells prior to staining and visualization on a microscope (114, 115). Specifically, a 1 mL aliquot from each culture was washed two times via centrifugation and resuspended in 500 µL of unbuffered Ringer’s buffer (1/4X) (Oxoid). We stained cells for 30 min at 37 °C with 2 µM BCECF-AM (Invitrogen). In addition to BCECF-AM staining, we added 5 µL of nigericin (Sigma Aldrich) to the tube containing the WT ladder standard to allow intracellular pH to equilibrate with the external environment. Following staining, we performed two additional washes via centrifugation using 500 µL of unbuffered 1/4X Ringers’ solution (Oxoid). 100 µL aliquots of each experimental strain and the ladder sample were transferred into five separate tubes. We centrifuged each tube for 1 min at 13,000×g and resuspended the pellets in one of seven 1/4X buffered Ringers’ solutions (pH 6.92, 7.25, 7.72, 8.33, 8.69, 9.06, and 9.59). We prepared buffered Ringers’ solutions using 50 mM of 1,3-Bis[tris(hydroxymethyl)amino]propane (Acros Organics) and adjusted the pH to the appropriate level using 1 M HCl. Once resuspended into pH-adjusted solutions, we imaged cells after approximately 30 min.

For each condition, 7 µL of cells was sandwiched between a 22 mm square #1.5 coverslip and a standard glass slide, providing sufficient fluid volume for cells to remain in solution but in approximately the same focal plane of the microscope. Cells were imaged on a Nikon Ti inverted microscope (Nikon Instruments, Melville, NY) equipped with an Andor Zyla sCMOS camera (Oxford Instruments, Abingdon, Oxfordshire), Spectra/Aura light engine (Lumencor, Beaverton, OR), and an Apo TIRF 60X Oil DIC N2 NA 1.49 objective (Nikon Instruments, Melville, NY). The BCECF fluorescence intensity was imaged in two channels (1: Ex 470 nm, Em 525/50 nm, and 2: Ex 440 nm, Em 525/50 nm). We used a semiautomated MATLAB script (MathWorks, Natwick, MA) to segment cells and quantify the fluorescence intensity. In brief, the process of segmentation identifies which pixels in the image correspond to individual objects or regions. Within each of these regions, the total intensity from each fluorescence channel was computed. Regions around these cells were used as proxies for the local background intensity and subtracted from the total intensity in each region. From these background subtracted intensities, the intensity ratio was calculated. To assist in the automation of this process, we utilized standard approaches in image processing which are described below. Images underwent a spatial bandpass filter (spatial frequencies between 220 nm and 1,100 nm) to remove salt and pepper noise and background intensity. Objects were then segmented as individual regions if above a certain intensity threshold. Because cells were plated at a sufficiently low density to ensure single objects, a relative threshold for each image was set as the intensity such that approximately the top 1,000 pixel values were above the threshold (quantile of 0.9998). To ensure that the entire region of each cell was included, and to deal with the fact that these two imaging channels have a small spatial shift from one channel to the other, these regions were then dilated by disks of radius 2 µm. Background regions were defined by dilating these regions by a further 2 µm, and removing the objects from these masks, yielding regions that are near, but not overlapping with, individual objects. The intensity for each object in each channel was determined as the background-subtracted sum of all the pixels in each region. Lists of relative intensities of these objects were then imported into RStudio. Samples treated with nigericin were used to determine the coefficients A and B in a fit of pH_obs = A*log(intensityRatio)+B.

To statistically determine differences in intracellular pH across the WT and engineered mutant strains, we fit the relationship between intracellular pH and environmental pH to a logarithmic function [Intracellular pH ~ a*log(Environmental pH)+k] using a nonlinear least squares method in the R package “nlm.” The fit was calculated across the measured interval of pH 7 to 9.5, and CI were calculated via uncertainty propagation with the function predictNLS which calculates CI for the fitted values of nonlinear models by using first-/second-order Taylor expansion and Monte Carlo simulation.

To determine the prevalence of the rho R109H and ΔydcI alleles in the RLTS condition and to parse out any patterns in their co-occurrence, we replayed one cycle of the RLTS regime by incubating WT, rho R109H, and ΔydcI cultures for 100 d after which time we examined the rho and ydcI loci for acquired mutations. We inoculated 12 cultures of WT, rho R109H, and ΔydcI strains in 10 mL of LB broth each originating from a separate colony from an LB agar plate freshly prepared from frozen glycerol stocks. We cultivated each culture at 37 °C and shaken upright at 180 rpm for a total of 100 d. We left cultures undisturbed during this time except for a brief period every 10 d when the culture tubes were gently removed from the incubator and added sterile deionized water to achieve a volume of 10 mL to counteract evaporation. After 100 d, we removed the cultures and centrifuged them 2,500×g for 10 min. We froze the resulting pellet of bacterial biomass in liquid nitrogen and it was stored at −80 °C until DNA extraction was performed. We extracted DNA from the pellets using a DNeasy UltraClean Microbial Kit (Qiagen) and used the resulting DNA samples as template for PCR. We performed PCR using the same primers previously mentioned that were used for confirmation of the initial mutation. The PCR products were purified using the Monarch PCR & DNA Cleanup Kit (NEB) and sent for Sanger sequencing (GENEWIZ).

To determine the fitness outcomes of the mutant strains relative to the WT ancestral strain, we performed pairwise competition assays (107, 116). Here, culture tubes containing a 50:50 starting proportion of two competing strains are grown for a given period, at which time they are sampled to determine the resulting proportions of each strain following growth. Each competing strain contains either an intact araBAD operon or ΔaraBAD genotype and can be differentiated from each other by plating on TA agar (10 g/L tryptone, 1 g/L yeast extract, 5 g/L NaCl, 16 g/L agar, 10 g/L L-arabinose, 0.005% tetrazolium chloride) where the WT araBAD strain produces light pink colonies and the ΔaraBAD strain produces bright red colonies. Overnight cultures were prepared for each strain using colonies picked from freshly streaked LB agar plates obtained from frozen stocks. For each competition, we inoculated 10 mL of LB broth with 50 µL of overnight culture from both competitors. We set up individual competitions for each sampling time point, as vortexing and resampling of the same competition tube has been shown to alter fitness outcomes (23). Before incubation, we sampled each competition to determine precise starting proportions by collecting a 100 µL aliquot, serially diluting in PBS, and plating on TA agar. All competition tubes were incubated at 37 °C until the sampling time point at which time a 100 µL sample was removed, serially diluted in PBS, and plated on TA agar. All TA agar plates were incubated at 37 °C for 24 h before colonies were enumerated. For competitions using buffered media, we added 50 mM 1,3-Bis[tris(hydroxymethyl)amino]propane (Acros Organics) and adjusted the pH to either 7.0 or 9.0 using 1 M HCl.

We assessed the survival of our strains by monitoring their population density over the course of 14 d. We inoculated three replicate overnight cultures of WT, rho R109H, ΔydcI, and rho R109H/ΔydcI strains in 10 mL of LB broth each originating from a separate colony from an LB agar plate freshly prepared from frozen glycerol stocks and incubated them for 18 h at 37 °C. The survival assay was initiated by aliquoting 100 µL of each overnight culture into 14 individual culture tubes filled with 10 mL of fresh LB broth. We incubated each culture at 37 °C, shaken upright at 180 rpm until the appropriate sampling time point. We set up individual cultures for each sampling time point, as vortexing and resampling of the same culture tube has been shown to alter fitness outcomes (23). Every 24 h, the culture tube corresponding to the sampling time point was removed from the incubator and gently vortexed. We then removed 100 µL from the culture and performed serial dilutions in PBS. We plated 100 µL aliquots of the dilutions on LB agar plates and enumerated the colonies from countable plates after 24 h of incubation at 37 °C.

Rho homologs containing a histidine at the analogous E. coli R109 residue were identified using blastp with the Pattern Hit Initiated (PHI-BLAST) algorithm (117). The following PHI pattern EH[YF][YFG][AGS][LM][LVT] targets R109H substitutions to the conserved ERFYALL motif from the E. coli Rho sequence (NP_418230.1) while allowing for any variation observed after alignment to Rho sequences with COBALT (118) across nine additional species of bacteria spanning the bacterial tree of life: Bacillus subtilis (AIY95021.1); Borreliella burgdorferi (WP_002657659.1); Deinococcus radiodurans (AAF10910.1); Helicobacter pylori (EIE30697.1); Pseudomonas aeruginosa (NP_253926.1); Salmonella enterica typhimurium (NP_462807.1); Staphylococcus aureus (YP_500838.1); Streptococcus pneumoniae (CVN06062.1); and Vibrio cholerae (WP_001054524.1) (Fig. 5).

S.B.W., B.P.B., M.B., and M.G.B. designed research; S.B.W., R.D.P.M., M.D., R.S., B.P.B., M.B., and M.G.B. performed research; R.S. contributed new reagents/analytic tools; S.B.W., R.D.P.M., M.D., B.P.B., M.B., and M.G.B. analyzed data; and S.B.W., R.D.P.M., B.P.B., M.B., and M.G.B. wrote the paper.

The authors declare no competing interest.