{"paper_id":"31dafeb3-7e3d-4619-adee-fea5bad85abc","body_text":"1 \n \nPulsatile basal gene expression as a fitness determinant in bacteria \nKirti Jain, Robert Hauschild, Olga O Bochkareva †, Roderich Roemhild, Gasper Tkačik, Calin C Guet* \nInstitute of Science and Technology Austria \n† Present Address: Centre for Microbiology and Environmental Systems Science, Division of \nComputational System Biology, University of Vienna \n \n*Corresponding author  \nEmail: calin@ist.ac.at \n \nAbstract \nActive regulation of gene expression, orchestrated by complex interactions of activators and \nrepressors at promoters, controls the fate of organisms. In contrast, basal expression at \nuninduced promoters is considered to be a dynamically inert mode of non-functional “promoter \nleakiness”, merely a byproduct of transcriptional regulation. Here, we investigate the basal \nexpression mode of the mar operon, the main regulator of intrinsic multiple antibiotic \nresistance in Escherichia coli, and link its dynamic properties to the non-canonical, yet highly \nconserved start codon of marR across Enterobacteriaceae. Real -time single -cell \nmeasurements across tens of generations, reveal that basal expression consists of rare \nstochastic gene expression pulses, which maximize variability in wildtype and, surprisingly, \ntransiently accelerate cellular elongation rates. Competition experiments show that basal \nexpression confers fitness advantages to wildtype across several transitions bet ween \nexponential and stationary growth by shortening lag times. The dynamically rich basal \nexpression of the mar operon, has likely been evolutionarily maintained for its role in growth \nhomeostasis of Enterobacteria within the gut environment, thereby allo wing other ancillary \ngene regulatory roles to evolve, e.g. control of costly -to-induce multi -drug efflux pumps. \nUnderstanding the complex selection forces governing genetic systems involved in intrinsic \nmulti-drug resistance is crucial for effective public health measures. \n \n \n \n \n  \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n2 \n \nBasal gene expression , also known as promoter leakiness, is a characteristic of \nbacterial promoters that occurs in their OFF state due to the presence of a repressor \nor the absence of an activator. Unlike induced or constitutive expression  in the ON \nstate, basal expression is generally not thought of as a functional mode of gene \nexpression, but rather a s a lack thereof. While selection can tune various aspects of \ngene induction, it is unclear how it could act, if at all, on the basal expression mode . \nGiven that promoter leakiness can be detrimental  (1–3), it could be under negative \nselection. However, we wondered whether there are any alternative basal expression \nmodes that could  have regulatory functions in their own right and thus be positively \nselected for . Recent studies uncovered the existence of a much more dynamic, \npulsatile basal expression mode for several bacterial genes. Such a basal mode can \ngenerate phenotypic diversity in a clonal population and has thus been rationalized as \na possible bet hedging mechanism (4–7). Essential to this  explanation are two \npremises. The first is “frequency matching”: bet hedging conveys a long -term fitness \nbenefit when it generates phenotypes in proportion to the frequencies of the \nenvironments for which these phenotypes are advantageous (8). The second is the \nexistence of a growth rate “cost” for a pulse: some (small) fraction of cells undergoing \na pulse pay  this cost upfront, in order to survive, or be more competitive , if a rare \nexternal stress should occur in that moment. While the bet hedging explanation is \nattractive, the implied growth rate costs and benefits,  as well as fitness effects more \nbroadly, are rarely measured (4). This motivates a fundamental question: Are the two \npremises of bet hedging met or should one seek alternative explanations for the \nevolutionary maintenance of a pulsatile basal expression mode? \nHere we turn to the marRAB operon, initially discovered as the genetic determinant of \nmultiple antibiotic resistance, and a paradigmatic example of a highly complex \nbacterial regulatory circuit (9, 10). The repressor MarR and the activator MarA form a \nnegative and a positive autoregulatory loop, respectively, and this unique topology of \ntwo interlocked loops jointly controls the mar function (11, 12). While MarR is a local \nregulator of  marRAB operon, MarA is a global regulator at the heart of one of the \nlargest E. coli regulons, encompassing over  30 genes, involved in multi -drug efflux, \npH regulation, outer membrane permeability, biofilm formation, and virulence (13–18). \nModelling and experimental studies suggest that the mar interlocked regulatory loops \ncould lead to pulsatile basal mode, resulting in phenotypic heterogeneity of expression \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n3 \n \nthat could support bet hedging (11, 19). Nevertheless, the characterization of the basal \nexpression mode for the mar operon and its functional and fitness implications beyond \nbet hedging and antibiotic stress remains unexplored. \n \nConservation of GTG start codon in marR across Enterobacteriaceae  \nAs MarR controls the repressed state of the mar operon and therefore its basal \nexpression, the unusual presence of a weak GTG start codon in marR piqued our \ninterest (20). Non-ATG start codons, i.e. , GTG and TTG, initiate ~8% genes  in \nGammaproteobacteria (21), reducing the translation efficiency of these genes so that \nsignificantly lower expression is achieved than if genes used ATG. To determine \nwhether the GTG start of marR is a historical contingency or the outcome of selection, \nwe constructed the marR phylogenetic tree and determined GTG prevalence across \nGammaproteobacteria. Among 889 representative genomes , marR homologs were \nfound in ~300 species , distributed across 20 distinct bacterial families ( Fig. 1A). \nInterestingly, we observed  that marR belongs to the marRAB operon only in  \nEnterobacteriaceae. Within all other bacterial families, marR-type transcription factors \nform operons with emrAB-type efflux pump genes (Fig. 1A, S1). \nThe phylogenetic tree corroborates an evolutionary scenario in which marRAB operon \nevolved only once and was vertically inherited (Fig. 1A). Its formation in the ancestor \nof Enterobacteria coincides with a change of the marR start codon from the canonical \nATG to the noncanonical GTG. While the marR-emrAB family has a strong ATG start \ncodon, marR in marRAB operons uses the weaker GTG  variant, with very few \nexceptions (Cronobacter, Jejubacter, Pluralibacter, Salmonella, Shimwellia , \nTenebrionocola), where the putatively even weaker TTG is used . Furthermore, a \nswitch from GTG to ATG occurred only in one Klebsiella clade. Few Enterobacteria \nharbor both, marRAB and marR-emrAB operons, (Cedecea, Kosakonia, Phytobacter, \nRaoultella), providing evidence for horizontal gene transfer of marR-emrAB into some \nEnterobacteriaceae (Fig. 1A). In addition to the start codon, the ribosome binding site \n(RBS) is also a  determinant of translational efficiency. With a single exception, the \nRBS of marR in marRAB is fully conserved among Enterobacteria (Fig. S2). Taken \ntogether, our phylogenetic analysis strongly suggests that the prevalent utilization of \nweak marR start codons across marRAB operons, in conjunction with a particular RBS \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n4 \n \nvariant, is selectively favored for a yet uncharacterized, but likely general, \nphysiological role. \n \nPulsatile basal expression of mar operon \nTo ask how the conserved GTG start codon affects marRAB function and fitness, we \nconstructed scarless marR mutants with alternative start codons (ATG and TTG) in \nEscherichia coli. In addition, in the ATG* mutant we combined the ATG start codon \nwith a stronger RBS (Fig. 1B).  \nTo investigate the basal mar expression mode  in single cells, we measured the \nfluorescence output of a Pmar-venus promoter fusion using time-lapse microscopy in \na microfluidic device  (22, 23) . We simultaneously  monitored a chromosomal \nconstitutive PR-mCherry as a control. Average background-corrected Pmar expression \ndepended significantly on the choice of start codon (Fig. 2A). TTG yielded 3-4 fold \nhigher Pmar expression than the wildtype (GTG), consistent with the expectation that \nTTG leads to weaker repressor translation. In contrast, the strong, canonical ATG start \ncodon reduced Pmar basal expression below the wildtype (GTG) levels. The ATG* \nmutant that combines the canonical ATG start codon with a strong RBS , abolished \nmost of Pmar expression and accessed a nearly complete OFF promoter state (Fig. \n2B).  \nWe next characterized the overall variability of the basal expression mode by \ncomputing the coefficient of variation (CV) of Pmar-venus fluorescence across 10 \nhours of observation for each of the ~180 independent mother cells per genotype (Fig. \n2C). The wildtype (GTG) showed maximum expression variability, followed by TTG \n(despite having higher  mean expression than that of the wildtype), and then ATG. \nThese three strains have at least 2-fold higher variability than constitutive controls. For \nATG*, the CV was only slightly elevated relative to the control (Fig. 2C, S3). \nThe observed high Pmar variability in single cells traces its origin to gene expression \npulses: transient, stochastic, high-amplitude activations of transcription plainly visible \nin all strains ( Movies M1-M5). To extract and statistically characterize the se pulses \n(Fig. S4), we first decomposed the observed Pmar-venus fluorescence distributions \ninto a Gaussian mixture . The frequent lower-amplitude component corresponded to \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n5 \n \nbaseline fluctuating Pmar expression level s, whereas the rarer component \ncorresponded to sporadic high -amplitude pulse -like excursions (Fig. 2D, S5). This \nmotivated a 85-percentile threshold (see Methods) for extracting the pulses which \ncould subsequently be aligned to their respective start times (Fig. 2E, S6) and \nquantified. \nWe report a similar frequency of pulsing in the wildtype (GTG), TTG and ATG strains, \nof one pulse per approximately 7 -8 hours. For ATG*, pulses appear to be much less \nfrequent, but our detection may be biased by their low signal-to-noise ratio (Table 1). \nThe pulse duration distribution was exponential for the wildtype (GTG), TTG, and ATG \nstrains for which it could be reliably estimated, with similar average duration of ~33-37 \nminutes per pulse (Table 1, Fig. 2G, S7). The key difference between the strain s lay \nin the overall mar expression that affects the baseline as well as pulse amplitudes \n(Table 1, Fig. S8). This expression changed several-fold depending on the MarR start \ncodon, implying that the MarR translation efficiency can tune Pmar expression by \ndetermining the promoter activity level outside and during the pulse. When expressed \nas fold-change increase over their respective baseline Gaussian components – which \nwe refer to as pulse “signal-to-noise” ratio (SNR) – differences between strains were \nsmaller but significant: pulse amplitudes ranged from ~1.3 – 1.7, with the maximal \nSNR reached in the wildtype (GTG) (Table 1, Fig. S8A). This difference at the level of \npulse characteristics is responsible for the maximal CV in mar expression reported for \nthe wildtype (GTG). Finally, after z-scoring and accounting for the individual durations \nof the pulses , we find that pulses nearly collapse onto a universal shape, indicating \nthat most of the variability  across genotypes  is accounted for by the statistics we \nextracted (Fig. S8B).  \nCan pulsing be modeled as a stationary stochastic point process? If that were the \ncase, pulse numbers should be Poisson-distributed over individual cells of the same \ngenotype. We report strong and highly significant deviations from this expectation for \nthe wildtype (GTG), TTG, and ATG strains but not for ATG* and controls (Fig. 2F, S9), \neven though inter-pulse intervals are exponentially distributed for all strains (Fig. 2H). \nSpecifically, the observed pulse count distributions are under-dispersed compared to \nPoisson, suggesting a more regular pulsing , possibly due to finite pulse duration  or \npulse-pulse correlations. Despite these quantitative deviations, individual pulses could \nbe interpreted in the bet-hedging framework as stochastic switches into an alternative \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n6 \n \n(high marA) phenotype once every ~14 – 16 generations and lasting for about one \ngeneration. \nWe next assessed the cost of mar basal mode expression. TTG cells had a \nsignificantly lower long-term elongation rates than wildtype (GTG) cells, whereas the \nelongation rates of ATG and ATG* were marginally higher than for the wildtype (GTG) \n(Fig. 3A). This corresponds to the ordering of mean mar expression levels across \nstrains (Fig. 2B) and is consistent with the expectation that higher overall mar activity \nis costly. A detailed analysis, however, revealed a surprising finding. We compared \nthe single-cell long-term elongation rates to the instantaneous elongation rates during \ndifferent phases of the pulse. We expected the elongation rates to slow down around \na pulse and subsequently return to the long -term average. In contrast, for all strains \nbut ATG* we observed significantly increased elongation rates in the time window 0-\n20 minutes after we identify the pulse start (Fig. 3B, C). The growth advantage could \nbe caused by differences in pulse amplitudes, where different sets of mar regulon \ntargets are engaged by different levels of MarA. This selective targeting is plausible, \nsince genes in the mar regulon are known to respond continuously and with different \nsensitivities to MarA levels (24). Taken together, larger baseline mar expression has \na cost, while a rare transient pulse confers a n advantage. Therefore, selection may \nhave to navigate this tradeoff in an environment-dependent way.  \nTo verify that low and transient (as opposed to high and persistent)  mar expression \nduring the pulse is necessary for an elongation rate advantage, we exposed ~110 cells \nper strain in our microfluidic device to 2mM salicylate to induce Pmar expression (Fig. \n3E). Induction caused a prolonged increase in marRAB expression, with the largest , \n~6-7 fold induction in the wildtype (GTG) and ATG, followed by TTG (4 -5 fold) and \nATG* (~3 fold)  (Fig. 3F). In terms of absolute expression , these levels were \nsubstantially (2x to 4x, depending on the strain) above the pulse amplitudes in the \nbasal mode (Table 1). Induction brought about a concomitant ~10% decrease in the \nelongation rate in all strains (Fig. 3D), consistent with previous reports and our \nexpectation that prolonged and strong marRAB expression is detrimental likely \nbecause it engages additional, more costly-to-express mar regulon targets (24).  \nThe induction experiment revealed another significant difference between the wildtype \n(GTG) strain and the strain with the canonical start codon (ATG), which emerged when \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n7 \n \nwe analyzed the inflection times of individual cell induction curves (Fig. 3G). The \nwildtype (GTG), which exhibits maximal expression variability in the basal mode, \nsurprisingly had the least variable induction curves and thus the most synchronized \nresponse to induction.  In comparison, the mutants show a two-fold reduction in \nsynchronicity. Overall this suggests that the unique interlocked regulatory circuit with \nshort-lived MarA  results in pr ecise induction to sudden stress (25), indicating that \nresponse speed and synchrony may be at a premium for the enterobacterial ecological \nniche. The observed difference between the wildtype (GTG) and the canonical start \ncodon ( ATG) strain motivated us to focus next on the fitness effects of mar in \nenvironments where mar expression mode transitions and timing could be relevant. \n \nFitness advantage for the wildtype basal expression mode across growth cycles \nThe surprising observation that basal mode pulse brings about a transient elongation \nrate advantage suggests a role of mar expression in physiology  and growth \nhomeostasis, which is in line with the subtle influence of mar in the transition from \nexponential to stationary phase  and back (26, 27). We therefore decided to conduct  \npairwise competition experiments to assess the performance of our strains across the \nentire growth cycle. We competed the wildtype (GTG) vs ATG or vs ATG* strains, and \nvs GTG itself (as a control), across four serial growth cycles over four days (together \n> 40 generations) in LB media without any external inducers . The key question was \nhow ATG and ATG*, the two strains with more efficient MarR repressor translation and \nthus lower baseline mar expression, compare against the wildtype (GTG) when the \ncells are forced to undergo repetitive transitions between exponential, stationary, and \nlag phases.  \nStarting from a 1:1 ratio, we saw the wildtype (GTG) increase to a ratio of 2:1 over the \ncourse of 53 generations in competition with the ATG  strain; the same 2:1 ratio was \nreached in  35 generations in competition with ATG*. The effective selection \ncoefficients were -0.013 and -0.020 per generation for ATG and ATG* , respectively, \nrelative to wildtype (GTG) (Fig. 4A and S10). The fitness advantage of the wildtype \n(GTG) in the absence of external inducers implies a functional role of the conserved \nweaker GTG start codon for cell physiology during serial growth cycles, likely via its \neffects on the pulsatile basal mode of mar expression. To understand how the wildtype \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n8 \n \n(GTG), which is at a disadvantage during exponential growth, outcompeted the two \nmutant strains, we measured how quickly various strains recovered from the lag phase \nand transitioned to exponential growth. We report a delay of around 8-12 minutes for \nATG and ATG* mutants vs the wildtype (GTG) in LB medium that matched conditions \nused in the competition experiment . The delay further increased by up to 25 -50 \nminutes for nutrient poor M9 media (Fig. 4B).  \nIn summary, these results suggest that the wildtype (GTG) strain compensates for its \nslower exponential growth rate by shortening its lag time. It is instructive to consider a \nsimple back -of-the-envelope calculation  using realistic estimates from Fig. 3. If the \nlong-term exponential growth rate of the wildtype (GTG) is 5% slower than that of the \nATG mutant but it emerges from lag phase with a 24 min advantage , then the ATG \nstrain would require 8 hours of uninterrupted growth to compensate for its delay and \nreach the same population size as the quicker-to-emerge but slower-to-grow wildtype \nstrain. While Enterobacteria growth cycles are certainly more complex in the wild than \nin our lab setup, this simple estimate shows that the strategy of shortening the lag time \ncould be competitive in practice. \n \nDiscussion \nThe process of turning genes ON and OFF is fundamental to life , and regulatory \nnetworks that control it have been the object of intense study in developmental, \nevolutionary, molecular, and systems biology. Nevertheless, the question of whether \nthe properties of the basal expression in the OFF state of promoters can be selected \nfor was rarely, if at all, considered. We showed that the mar operon basal expression \nmode is highly dynamic, consisting of pulsatile gene expression at the single cell level, \nand we measured the fitness consequences of such dynamics. To this end, we used \nsynonymous codon mutations for the start codon of marR, which we uncovered to be \nevolutionarily conserved and tightly coupled to the interlocked regulatory architecture \nof the marRAB operon. \nWe find that the  weaker, non-canonical, GTG start codon act s as a regulatory knob \nensuring the presence of sporadic transcription pulses that are much less pronounced \nor absent if marR and marA are translated with similar speed, as is the case in the \nvarious start codon mutants we measured. The GTG start codon endows the wildtype \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n9 \n \nbasal expression mode with several unique characteristics : highest expression \nvariability due to highest “signal -to-noise” ratio of individual pulses,  transient growth \nadvantage during a pulse, and synchronized expression during transition to the \ninduced state. Altogether, these quantitative expression characteristics lead to a  \nrobust fitness advantage for the wildtype  that becomes apparent in environments in \nwhich cellular physiology switch es between exponential , lag phase, and stationary \ngrowth. More broadly, this points to the involvement of marRAB operon in general \ngrowth homeostasis. \nWhile the  marRAB operon can be induced  by various  metabolic intermediates, no \nmain physiological inducers have been characterized  so far  (28). However, the \npresence of nearly identical  mar systems across most  Enterobacteria and the \nremarkable conservation of the non-canonical GTG start codon of marR suggest that \nthe cellular processes controlled by the mar operon confer adaptations that are most \nlikely linked to Enterobacteriaceae physiology and their ecological niche (29). \nEssential to this  ecological niche is that bacteria spend a significant  fraction of their \nexistence inside the guts of various animals. The quasi regular intervals of feeding that \nare characteristic of the gut environment  point towards a selective pressure for \noptimizing cycles of lag, exponential , and stationary phases. Our results support the \nidea that the mar pulsatile basal expression mode could be an essential building block \nof this physiological adaptation to the intestinal lifestyle, allowing the basal expression \nmode to be evolutionarily maintained.  \nA supporting observation for this hypothesis  is the quantitative match between the \nmultiple timescales  related to the mar operon in  Escherichia coli : the typical time \nbetween two  mar pulses, the timescale at which the shorter lag of wildtype would \nbalance out its slower growth compared to marR start codon mutants, and the daily \nfeeding cycles, which are all on the order of ~10 hours. In the bet hedging framework, \nthis could represent a quantitative case of frequency matching: yet instead of thinking \nof very rare stresses (such as antibiotic exposure) , the system has evolved to match \nthe frequency of environmental transitions typical for the ecological niche of \nEnterobacteriaceae. Constant selective pressure to maintain the mar basal expression \nmode for such physiological raison d’etre would enable the same system to be co -\nopted for other hypothesized functions  in its induced mode : to generate diversity \nnecessary for bet -hedging against much rarer stresses, or to induce costly stress -\n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n10 \n \nresistance genes when needed , and thus allow MarA to b ecome a global \ntranscriptional regulator. In addition to cyclic nutrient availability, t he gut ecology is  \nalso characterized by host defense mechanisms and antimicrobials secreted by other \nmicrobiota. Thus, co-opting regulation of physiological adaptation  to cyclic nutrient \navailability and antibiotic resistance mechanisms would be consistent with the global \nregulatory role of mar (30). \nThe naming of the mar operon – an acronym for multiple antibiotic resistance coined \nover 40 years ago by George and Levy  (9, 31) – was based on the conferred  broad \nantibiotic resistance phenotype in the induced state (32). By now, it is clear that the \nmar operon has not evolved “for” antibiotic resistance, which is at best an ancillary \nfunction – albeit one of fundamental importance for public health . To control it  and \ncounteract the loo ming multi-drug resistance epidemic , our work demonstrates the \nacute need to understand the role of  marRAB operon in bacterial physiology and \ngrowth homeostasis, in line with Seoane & Levy who argued early on for an alternative \nrole of mar as a conveyor of ‘multiple adaptational response’ (10).  \n \nMaterials and Methods \nComputational Genomics \nGenomes of Gammaproteobacteria were downloaded from RefSeq database (33) \nusing the PanACoTA pipeline (34), module ‘download’ with filters: genome collections \n= ‘Reference’ or ‘Representative’; assembly level = ‘complete genomes’ or \n‘chromosome’ or ‘scaffold’. Then we used the PanACoTA module ‘annotate’ to predict \nand functionally annotate CDS in the genomes. \nmarR genes were found using HMM for OG #1S2AX comprising E.coli marR (UniProt \nentry ID:P27245), from eggNOG5 database with e-value threshold 10-35 (35, 36). We \nadditionally used HMMs for OG #1S26B and OG #1RPCJ which contain slyA (UniProt \nentry ID:P0A8W2) and mprA (UniProt entry ID:P0ACR9), correspondingly to include \ninto the analysis more members of marR family in order to better resolve the gene tree \n(Fig. S1). \nTo find ribosomal binding sites (RBSs) and correct putative errors in gene start \nprediction, marR upstream regions of 100 bp length were extracted and aligned taking \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n11 \n \ninto account the known RBS of marR in E. coli (20). Four downstream genes were \nused for the annotation of the marR genomic context. \nThe alignments of protein sequences and gene upstreams was made using Muscle  v. \n3.8.31 (37). The gene trees were constructed using IQtree v.1.6.12 with 1000 \nbootstrap runs (38). Visualization and annotation of the trees was done using Itol \nserver (39). The logo of PmarRAB upstream was created usin g WebLogo 3 web -\nbased application (40). \n \nStrains and Media \nAll experiments were performed using the derivate of Escherichia coli K-12 MG1655 \nstrain, with incubations at 37°C and aeration. \nLysogeny Broth (LB) media was used in all experiments except when noted. For \nplates, 1.5% agar was added to LB. For microfluidics experiment, 0.01% Tween 20 \nwas used to prevent attachment of bacterial cells to the PDMS device. Media and \nantibiotics were from Sigma, Sylgard for making PDMS was from Dow Chemicals. List \nof strains and primers used in this study are listed in (Table ST1, ST2). \nStrain Construction  \nThe DIRex method was used to generate scar -free point mutations for changing the \nstart codon and R BS for marR (41). Briefly, the method uses a single λ Red \nrecombineering step (pSIM5-Tet temperature sensitive) and a semi -stable AcatsacA \nintermediate. The desired changes were introduced through custom-made oligos with \na homology to the target region. The AcatsacA cassette codes for three genes which \nhelp in selection and counter selection steps a) cat leading to chloramphenicol \nresistance (use 12.5 mg/L chloramphenicol to select for AcatsacA formation), b) \namilCP, present as dual inverted copies, causing AcatsacA+ colonies to be of blue \ncolor helping in selection, and c) sacB, sucrose sensitivity gene (use 5% s ucrose to \nselect for self-excision), helps in counter selection generating scar-free mutants.  \nWe use a constitutive chromosomal PR-mCherry reporter as control (22). Pmar-venus \nreporter is on a low copy plasmid (23).  \n \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n12 \n \nMicrofluidics Set Up and Imaging \nWe used the same microfluidic chip as previously used in our lab (22). The length of \nthe growth channel is 24µm and the width of these growth channels ranged from \n1.2µm to 1.4µm and the height is approximately 1.1µm. To make mother machine \ndevices from  Epoxy replica, we used PDMS in a ratio of 10:1 (Sylgard and curing \nagent) and mixed and degassed in Thinky Machine (THINKY ARE-250) for 2 minutes \neach. Further degassing was done after pouring on the epoxy replica using a \ndesiccator. Curing of PDMS was do ne overnight in an incubator at 80°C. Next, the \nPDMS device was peeled out carefully from the epoxy replica and holes were punched \nusing an electro polished 18ga needle. The device was cleaned with scotch tape and \nthe cover slip (24mm x 50mm, thickness 0.1 7mm+/-0.005) was cleaned with \nisopropanol. Device was then bonded using plasma bonding technique (Harrick PDC-\n002 plasma cleaner, medium power for 1 min, for both PDMS and cover slip) followed \nby gently placing on the cover slip. After bonding, it was kept  on a hot plate (~80°C) \nfor one hour.  \nBefore starting the experiment, the device was wetted with 0.01% Tween 20 for a \ncouple of minutes followed by blowing out. This step also ensures that the bonding is \nleak-proof. Next, a pellet from exponential grown cells (overnight culture of the desired \ngenotype in LB plus Tween 0.01% is grown and sub -cultured 1:1000 and grown for \naround four hours and centrifuged at 4000 Xg for three minutes) were loaded using a \npipette. After confirming the loading of cells by checking under the scope, media flow \nwas connected with polyethylene tubing (BTPE -50Instech). Image acquisition settings \nwere kept identical throughout all experiments (Exposure time for mCherry: 200ms \nand venus: 300ms) with an image interval of 90 secs. Imag es were acquired with an \nOlympus IX83 inverted fluorescence microscope, a 100X NA 1.45 objective, with a \ncustom made autofocus, and a Hamamatsu Orca Flash4.0v2 camera (42). \n \nImage Analysis \nPre Segmentation: \nImages of the channel areas within the microfluidic chip were cropped and background \nand shading corrected (42). \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n13 \n \nSegmentation \nBacteria segmentation was carried out using Cellpose (43). A custom model was \ntrained on a dataset of over 2000 hand-labeled cells selected from a diverse range of \nexpression levels and morphologies. \nTracking \nA customized Matlab script was employed for tracking, generating lineage trees, and \nconducting further analyses. In the initial stage, the microfluidic channels were \nautomatically detec ted, and cells located outside the channels were eliminated. A \nheuristic method was applied to address missing cell detection and correct under -\nsegmentation errors. Subsequently, each channel underwent individual tracking: The \nlink cost function that establishes connections between cell detections at consecutive \ntime points to form tracks, took into account the specific characteristics of the mother \nmachine. This was achieved by assigning a higher link cost to reverse movement, \ninstances where cells swapped positions within the channel, and a reduced overlap in \ncell segmentation compared to the segmentation at the previous time point. Cell \ndivisions were identified when two cells overlapped with the same segmentation from \nthe preceding frame. Empty channels were omitted for clarity. \nGrowth rate  \nWe defined and quantified the growth rate and the promoter activity as done by Kim \net al.  (6, 44). The growth rate 𝑔 is calculated as the logarithm of the ratio of the area \nof the cell immediately prior the cell division 𝐴𝑑 and the area at the initial time point \nfollowing the last cell division 𝐴0 : 𝑔 = log (\n𝐴𝑑\n𝐴0\n) /∆𝑡. \nAs we did not observe any long term change in the average cell size, and we set \n(\n𝐴𝑑\n𝐴0\n) = 2. \nAccording to this definition, it can be inferred that the growth rate remains constant \nbetween individual cell divisions. \nInstantaneous elongation rate and cell division events \nThe instantaneous elongation rate 𝑅 is calculated as 𝑅 =\n𝑑𝐿\n𝑑𝑡. We derived the length 𝐿 \nfrom a linear fit of the cell segmentation area, rather than an ellipse fit to the segmented \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n14 \n \ncell, as this method was more robust against segmentation errors. To smooth out the \ngrowth rate, we applied a running average over a 15 -minute window. Cell division \nevents were identified by employing criteria that recognize when the segmented area \nof a cell approximately halves during division. Cell division events were excluded from \nthe growth rate analysis.  \nPulse detection and Analysis \nThe distribution of combined raw fluorescence intensities is accurately modeled as a \nsum of two distinct Gaussian distributions. This modeling provides a natural cutoff \nvalue, effectively differentiating baseline expression fluctuations from stochastic pulse-\nlike expressions. Using this cutoff value, we perform a z -transform on each cell's raw \nfluorescence intensities, utilizing the mean and standard deviation of the baseline \nexpression. Pulses are identified by applying a threshold to the z-score (Fig. S4). This \nmethod is employed for both fluorescence derived from the Pmar-venus promoter \nfusion and the chromosomal constitutive PR-mCherry expression. Additionally, we \nconducted a stochastic simulation replicating the characteristics of the constitutive \nexpression. Analysis of the pulse length histogram in all three cases supports a cutoff \nfor determining the duration of genuine gene express ion pulses: Pulses shorter than \n15 minutes are deemed random baseline fluctuations, whereas longer pulses, \nindicative of actual gene expression, are compiled for further examination. A linear \nregression on the inverse slope of the histogram of true pulses provides an average \npulse length comparable to the directly calculated mean pulse duration. We also fitted \nPoisson distributions to the number of pulses per cell. \nFor visualization, pulses are aligned temporally, setting the start time of each pulse to \nt=0 and sorting them by duration. This approach allows for the calculation of the \naverage intensity over time from Pmar-venus by normalizing each pulse's intensity to \nits peak value. However, the transient pulse intensity is affected by two factors: the \ntransient nature of each pulse and the distribution of pulse duration. To isolate the \nimpact of varying pulse duration, we normalize the duration of each pulse to one. \nConsequently, the ensemble average of all duration -normalized pulses reveals the \nstereotypical shape of a pulse. \nInflection point \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n15 \n \nIn order to quantify the synchronicity of induction, the inflection point of the \nfluorescence 𝐼 was determined by finding the peak of  \n𝑑𝐼̅\n𝑑𝑡 with 𝐼̅ being the ten min \nmoving average of 𝐼. The induction magnitude 𝑀, (= fold change of 𝐼) is given by the \nratio of the average 𝐼 at a fixed time interval before and after the inflection point: \n 𝑀 = 𝐼(𝑡 + ∆)̅̅̅̅̅̅̅̅̅̅̅ 𝐼(𝑡 − ∆)̅̅̅̅̅̅̅̅̅̅̅⁄ . \nStatistical tests \nWe performed rank-sum tests for pairwise comparison of wildtype (GTG) with mutants \nfor Pmar-venus and PR-mCherry expression. For elongation rate during pulse and \nelongation rate upon induction, we performed t-tests. \n \nCompetition Experiment \nWe used two different marker methods: fluorophores (mCherry and Venus) (Fig. 4A) \nand resistance (chloramphenicol) (Fig. S10). Each genotype was grown overnight in \nfour replicates for each marker separately. Competitions were set up as head-to-head \ncompetitions of two genotypes, with dye-swap controls, where each marker was used \nfor half of the replicates. The optical density (OD) was measured and pairwise cultures \n(wildtype (GTG): wildtype (GTG), wildtype (GTG):ATG, wildtype (GTG):ATG*) were \nmixed in a 1:1 ratio, diluted 1000x as to permit for ten generations of growth and \nincubated at 37°C for 24 hou rs. The 1000x -fold dilution was repeated for three \nconsecutive days so that the genotypes were in direct competition for a total of 40 \ngenerations. In parallel, the cultures were plated on LB agar for CFU quantification of \nthe genotypes. Plates were imaged  using a custom -build fluorescence macroscope \nand fluorescent colonies were counted using ImageJ (42). When using the resistance \nmarkers, cells were plated on both LB agar and LB agar with chloramphenicol. \nSelection coefficients were calculated as the slope of the linear model fit to the log \nratio of genotypes (using natural log) over generat ions of competitive growth.  We \ncorrected for fitness costs of the markers by subtracting the baseline selection \ncoefficients of control competitions where the same genetic background was \ncompeted against itself with different markers. We then used a gener alized linear \nmodel with genotype as fixed factor and experiment as random factor and a post-hoc \ntest with user -defined contrasts to statistically test for significant selection between \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n16 \n \ngenotype pairs and estimate the selection coefficient. These analyses were done \nusing the statistical software R and the packages multcomp and nlme. \n \nPopulation measurements of growth rate and lag phase \nGrowth rate and growth lag were measured by strongly diluting overnight cultures into \n0.3 mL of fresh media in Honeycomb pl ates and measuring OD during exponential \nregrowth at 37°C with vigorous shaking every 4 min using the Bioscreen C plate reader \n(OY Growth Curves, Helsinki, Finland; Ref. FP -1100-C) system for at least 6 hours. \nExponential growth rates were estimated as doublings per hour using the slope of the \nlinear model fit to the plot of log -transformed OD over time in hours during the \nexponential growth phase. Accuracy of exponential growth was assess using R 2 \nvalues of the fit, which was >0.94 for all growth curves (average 0.98). Lag phase was \nestimated as the time to restart exponential growth of OD, for which we used the time \nat which OD reached above the detection threshold of OD = 0.004. Time delay of \nmutant strains over the wild type strain was calculated by subtracting the average lag \ntime of the wild type from the lag times of the mutants. To allow for estimation of lag \nphase growth curves were started with equal ODs. ODs of overnight cultures were \nnormalized using OD immediately before inoculation into prepared Honeycomb plates. \nWithout this correction, start growth time is dependent on starting cell density (45). We \nused ANOVA to test for significant differences in growth parameters between strains \nand three different growth condi tions, where “LB, fresh” refers to regrowth in LB \nfollowing dilution of a 16h overnight culture in LB, “LB aged” refers to regrowth in LB \nfollowing dilution of a 48h culture in LB and “M9 glycerol aged” refers to regrowth in \nM9 minimal medium with 0.2% glycerol as the sole carbon source following dilution of \na 48h culture in that medium. \n  \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n17 \n \nReferences \n1.  A. L. Koch, The protein burden of lac operon products. J. Mol. Evol. 19, 455–\n462 (1983). \n2.  D. M. Stoebel, A. M. Dean, D. E. Dykhuizen, The cost of expression of \nEscherichia coli lac operon proteins is in the process, not in the products. \nGenetics 178, 1653–1660 (2008). \n3.  J. M. L. Ho, C. A. Miller, S. E. Parks, J. R. Mattia, M. R. Bennett, A suppressor \ntRNA-mediated feedforward loop eliminates leaky gene expression in bacteria. \nNucleic Acids Res. 49, e25–e25 (2021). \n4.  O. Patange, C. Schwall, M. Jones, C. Villava, D. A. Griffith, A. Phillips, J. C. W. \nLocke, Escherichia coli can survive stress by noisy growth modulation. Nat. \nCommun. 9, 5333 (2018). \n5.  E. Nadezhdin, N. Murphy, N. Dalchau, A. Phillips, J. C. W. Locke, Stochastic \npulsing of gene expression enables the generation of spatial patterns in Bacillus \nsubtilis biofilms. Nat. Commun. 11, 950 (2020). \n6.  J. M. Kim, M. Garcia-Alcala, E. Balleza, P. Cluzel, Stochastic transcriptional \npulses orchestrate flagellar biosynthesis in Escherichia coli. Sci. Adv. 6, \neaax0947 (2020). \n7.  N. M. V. Sampaio, C. M. Blassick, V. Andreani, J.-B. Lugagne, M. J. Dunlop, \nDynamic gene expression and growth underlie cell-to-cell heterogeneity in \nEscherichia coli stress response. Proc. Natl. Acad. Sci. 119, e2115032119 \n(2022). \n8.  E. Kussell, S. Leibler, Phenotypic diversity, population growth, and information \nin fluctuating environments. Science 309, 2075–2078 (2005). \n9.  A. M. George, S. B. Levy, Amplifiable resistance to tetracycline, \nchloramphenicol, and other antibiotics in Escherichia coli: involvement of a non-\nplasmid-determined efflux of tetracycline. J. Bacteriol. 155, 531–540 (1983). \n10.  A. S. Seoane, S. B. Levy, Characterization of MarR, the repressor of the \nmultiple antibiotic resistance (mar) operon in Escherichia coli. J. Bacteriol. 177, \n3414–3419 (1995). \n11.  J. Garcia-Bernardo, M. J. Dunlop, Tunable stochastic pulsing in the Escherichia \ncoli multiple antibiotic resistance network from interlinked positive and negative \nfeedback loops. PLoS Comput. Biol. 9, e1003229 (2013). \n12.  M. K. Prajapat, K. Jain, S. Saini, Control of MarRAB operon in Escherichia coli \nvia autoactivation and autorepression. Biophys. J. 109, 1497–1508 (2015). \n13.  T. M. Barbosa, S. B. Levy, Differential expression of over 60 chromosomal \ngenes in Escherichia coli by constitutive expression of MarA. J. Bacteriol. 182, \n3467–3474 (2000). \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n18 \n \n14.  P. J. Pomposiello, M. H. J. Bennik, B. Demple, Genome-Wide Transcriptional \nprofiling of the Escherichia coli responses to superoxide stress and sodium \nsalicylate. J. Bacteriol. 183, 3890–3902 (2001). \n15.  R. G. Martin, J. L. Rosner, Genomics of the marA/soxS/rob regulon of \nEscherichia coli: identification of directly activated promoters by application of \nmolecular genetics and informatics to microarray data. Mol. Microbiol. 44, 1611–\n1624 (2002). \n16.  P. Casaz, L. K. Garrity-Ryan, D. McKenney, C. Jackson, S. B. Levy, S. K. \nTanaka, M. N. Alekshun, MarA, SoxS and Rob function as virulence factors in \nan Escherichia coli murine model of ascending pyelonephritis. Microbiology 152, \n3643–3650 (2006). \n17.  P. Sharma, et al., The multiple antibiotic resistance operon of enteric bacteria \ncontrols DNA repair and outer membrane integrity. Nat. Commun. 8, 1444 \n(2017). \n18.  L. M. Chubiz, The Mar, Sox, and Rob Systems. EcoSal Plus eesp-0010-2022 \n(2023). https://doi.org/10.1128/ecosalplus.esp-0010-2022. \n19.  I. El Meouche, Y. Siu, M. J. Dunlop, Stochastic expression of a multiple \nantibiotic resistance activator confers transient resistance in single cells. Sci. \nRep. 6, 19538 (2016). \n20.  R. G. Martin, J. L. Rosner, Transcriptional and translational regulation of the \nmarRAB multiple antibiotic resistance operon in Escherichia coli. Mol. Microbiol. \n53, 183–191 (2004). \n21.  F. Belinky, I. B. Rogozin, E. V. Koonin, Selection on start codons in prokaryotes \nand potential compensatory nucleotide substitutions. Sci. Rep. 7, 12422 (2017). \n22.  T. Bergmiller, et al., Biased partitioning of the multidrug efflux pump AcrAB-TolC \nunderlies long-lived phenotypic heterogeneity. Science 356, 311–315 (2017). \n23.  C. C. Guet, L. Bruneaux, P. Oikonomou, M. Aldana, P. Cluzel, Monitoring \nlineages of growing and dividing bacteria reveals an inducible memory of mar \noperon expression. Front. Microbiol. 14, 1049255 (2023). \n24.  R. G. Martin, E. S. Bartlett, J. L. Rosner, M. E. Wall, Activation of the \nEscherichia coli marA/soxS/rob regulon in response to transcriptional activator \nconcentration. J. Mol. Biol. 380, 278–284 (2008). \n25.  K. L. Griffith, I. M. Shah, R. E. Wolf, Proteolytic degradation of Escherichia coli \ntranscription activators SoxS and MarA as the mechanism for reversing the \ninduction of the superoxide (SoxRS) and multiple antibiotic resistance (Mar) \nregulons. Mol. Microbiol. 51, 1801–1816 (2004). \n26.  G. P. Price, A. C. St. John, Purification and analysis of expression of the \nstationary phase-inducible Slp lipoprotein in Escherichia coli : role of the Mar \nsystem. FEMS Microbiol. Lett. 193, 51–56 (2000). \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n19 \n \n27.  K. J. Allen, D. Lepp, R. C. McKellar, M. W. Griffiths, Targeted microarray \nanalysis of stationary phase Escherichia coli O157:H7 subjected to disparate \nnutrient conditions: Transcriptome analysis of starved E. coli O157:H7. J. Appl. \nMicrobiol. 109, 2118–2127 (2010). \n28.  L. M. Chubiz, C. V. Rao, Aromatic acid metabolites of Escherichia coli K-12 can \ninduce the marRAB operon. J. Bacteriol. 192, 4786–4789 (2010). \n29.  M. A. Savageau, Escherichia coli habitats, cell Types, and molecular \nmechanisms of gene control. Am. Nat. 122, 732–744 (1983). \n30.  J. F. Brooks, et al., The microbiota coordinates diurnal rhythms in innate \nimmunity with the circadian clock. Cell 184, 4154-4167.e12 (2021). \n31.  A. M. George, S. B. Levy, Gene in the major cotransduction gap of the \nEscherichia coli K-12 linkage map required for the expression of chromosomal \nresistance to tetracycline and other antibiotics. J. Bacteriol. 155, 541–548 \n(1983). \n32.  J. L. Rosner, Nonheritable resistance to chloramphenicol and other antibiotics \ninduced by salicylates and other chemotactic repellents in Escherichia coli K-12. \nProc. Natl. Acad. Sci. 82, 8771–8774 (1985). \n33.  N. A. O’Leary, et al., Reference sequence (RefSeq) database at NCBI: current \nstatus, taxonomic expansion, and functional annotation. Nucleic Acids Res. 44, \nD733-745 (2016). \n34.  A. Perrin, E. P. C. Rocha, PanACoTA: a modular tool for massive microbial \ncomparative genomics. NAR Genomics Bioinforma. 3, lqaa106 (2021). \n35.  The UniProt Consortium, et al., UniProt: the Universal Protein Knowledgebase \nin 2023. Nucleic Acids Res. 51, D523–D531 (2023). \n36.  J. Huerta-Cepas, et al., eggNOG 5.0: a hierarchical, functionally and \nphylogenetically annotated orthology resource based on 5090 organisms and \n2502 viruses. Nucleic Acids Res. 47, D309–D314 (2019). \n37.  R. C. Edgar, MUSCLE: multiple sequence alignment with high accuracy and \nhigh throughput. Nucleic Acids Res. 32, 1792–1797 (2004). \n38.  L.-T. Nguyen, H. A. Schmidt, A. von Haeseler, B. Q. Minh, IQ-TREE: a fast and \neffective stochastic algorithm for estimating maximum-likelihood phylogenies. \nMol. Biol. Evol. 32, 268–274 (2015). \n39.  I. Letunic, P. Bork, Interactive Tree Of Life (iTOL) v5: an online tool for \nphylogenetic tree display and annotation. Nucleic Acids Res. 49, W293–W296 \n(2021). \n40.  G. E. Crooks, G. Hon, J.-M. Chandonia, S. E. Brenner, WebLogo: A Sequence \nLogo Generator: Figure 1. Genome Res. 14, 1188–1190 (2004). \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n20 \n \n41.  J. Näsvall, Direct and Inverted Repeat stimulated excision (DIRex): Simple, \nsingle-step, and scar-free mutagenesis of bacterial genes. PLOS ONE 12, \ne0184126 (2017). \n42.  R. Chait, J. Ruess, T. Bergmiller, G. Tkačik, C. C. Guet, Shaping bacterial \npopulation behavior through computer-interfaced control of individual cells. Nat. \nCommun. 8, 1535 (2017). \n43.  C. Stringer, T. Wang, M. Michaelos, M. Pachitariu, Cellpose: a generalist \nalgorithm for cellular segmentation. Nat. Methods 18, 100–106 (2021). \n44.  L. Potvin-Trottier, N. D. Lord, G. Vinnicombe, J. Paulsson, Synchronous long-\nterm oscillations in a synthetic gene circuit. Nature 538, 514–517 (2016). \n45.  R. Hazan, Y.-A. Que, D. Maura, L. G. Rahme, A method for high throughput \ndetermination of viable bacteria cell counts in 96-well plates. BMC Microbiol. 12, \n259 (2012). \n \n \n \n \n \n \n \n \n  \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n21 \n \nAuthor contributions:  \nConceptualization: KJ and CCG \nMethodology: KJ (microfluidics, image analysis, competition and regrowth experiments), RH \n(analysis script), OOB (computational genomics), RR (competition and regrowth \nexperiments) \nInvestigation: KJ (microfluidics, image analysis, competition and regrowth experiments), \nOOB (computational genomics), RR (competition and regrowth experiments) \nVisualization: KJ, RH, OOB, RR, GT, CCG \nFunding acquisition: CCG \nProject administration: KJ and CCG \nSupervision: GT and CCG \nWriting – original draft: KJ \nWriting – review & editing: KJ, GT, CCG, RR, OOB, RH \nCompeting interests: Authors declare they have no competing interest. \nClassification: Major - Biological Sciences and Minor - Systems Biology, Microbiology \nKeywords: Gene regulation, basal expression, mar operon \nThis pdf file contains: \nMain Text \nFigures 1 to 4 \nTable 1 \n \nAcknowledgements: KJ thanks B. Wu, I. Tomanek, K. Tomasek for detailed discussions on \nthe manuscript, all other members from the Guet laboratory for helpful feedback, and R. Chait, \n& IOF IST Austria for helping with the microscope.  \nFunding: KJ acknowledges IST fellowship IC1006FELL02, RH was supported in part by CZI \ngrant DAF2020-225401 (10.37921/120055ratwvi) from the Chan Zuckerberg Initiative DAF, \nOOB acknowledges FWF Grant ESP253-B, RR acknowledges FWF Grant 10.55776/ESP219, \nCCG acknowledges FWF I5127-B. \nData availability: All data to understand and assess the conclusions of this research are \navailable in the main text and supplementary file. Raw data is at the IST repository. \n  \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n22 \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n23 \n \nFig. 1.  Distribution of marR start codons across Enterobacteriaceae. (A)  \nMaximum likelihood phylogenetic tree of marR in Gammaproteobacteria. The tree was \nconstructed using other members of marR family (full tree in Fig. S1). The marRAB \noperon is only present in Enterobacteriaceae, while in other bacterial families, a marR-\ntype transcription factor was found in conjunction with the genes forming an emrAB-\ntype efflux pump. (B) Schematic representation of marRAB operon (top), strains \nconstructed (middle), and reporter plasmid (bottom).  \n  \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n24 \n \n \n \n \n  \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n25 \n \nFig. 2.  Characterization of Pmar-venus basal expression dynamics.  (A) \nKymographs of wildtype (GTG) and mutants (TTG, ATG, and ATG*) showing Pmar-\nvenus expression for a representative mother cell imaged in a microfluidic channel \nover 10 hours. (B) Mean Pmar-venus expression in wildtype and mutants (Mean and \nSD over cells, p<10-4, rank-sum test). (C) Coefficient of variation (CV) of Pmar-venus \nexpression in wildtype and mutants compared against CV of control (constitutive \nreporter PR-mCherry expression). Pmar-venus expression CV in wildtype is \nsignificantly different from mutants (mean and SD over cells, p<10 -4, rank-sum test). \n(Pmar-venus expression CV is also significantly different from the corresponding PR-\nmCherry control for each strain, p<10-4, rank-sum test). (D) Distribution of Pmar-venus \nexpression is modeled as a sum of two Gaussian distributions, i.e., baseline and pulse \ncomponent, for wildtype. Magenta circles represent the amplitudes of individual \nextracted pulses and magenta circle with an error bar is the mean amplitude + SD over \nindividual pulses. (E) Pulses for wildtype sorted by duration and pulse start is aligned \nto 0 min on the x -axis. (F) Distribution of the number of pulses per cell in wildtype in \nDataset 1 (181 cells, 600  mins) and Dataset 2 (131 cells, 2120 mins) showing \nsignificant deviation from Poisson. Error bars represent √(Count). (G) Log histogram \nof pulse durations. Mean pulse duration (in caption) was derived from the inverse of \nthe slope of the linear fit. (H) Log histogram of time intervals between two consecutive \npulses. Inset shows the corresponding PDF. \n \n \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n26 \n \n \n  \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n27 \n \nFig. 3. Elongation rates and Pmar-venus expression quantification in baseline \n(A to C) and induced state (D -G). (A) Single-cell elongation rates in the baseline \nstate (individual cell data (circles) with mean (square) and SD). Wildtype elongation \nrate is significantly different from TTG & ATG (p<10 -4, rank -sum test) and ATG* \n(p<0.05, rank -sum test). (B) Raw data showin g instantaneous elongation rate \nnormalized to the long-term individual cell average and aligned to pulse start (vertical \nmagenta line). Pulse duration is denoted in magenta as well. (C) Mean and SE of \nelongation rates (normalized to the long-term average elongation rate of the respective \ncell) in 20 min windows (80 mins before and after the pulse start) for wildtype and \nmutants. The shaded regions illustrate the start of the pulse and the mean pulse \nduration in each strain. Stars indicate significant deviat ion from the long -term \nelongation rate (t -test). (D) Mean and SE of elongation rate change upon induction \nwith 2mM salicylate relative to their respective baselines  shows similarly-sized fold \nchange relative to pre-induction baseline and statistically significant reduction across \nstrains (p<10-4, t-test). (E) Representative raw data (Pmar-venus expression) showing \nthe transition from baseline to 2mM salicylate -induced state. (F) Individual cell data \n(circles), mean (square), and SD of fold change of Pmar-venus expression upon \ninduction with 2mM salicylate, horizontal line shows a comparison with pulse mean \nSNR in the baseline state. Wildtype expression fold change upon induction is \nsignificantly different from TTG & ATG* (p<10-4, rank-sum test) and ATG (<0.05, rank-\nsum test). (G) Individual cell data (circles), mean (square), and SD of variability in \ninduction time quantified by the time taken by expression to reach the inflection point \nfor each cell. Wildtype is significantly different from all three mutants  (p<10-4, rank-\nsum test). \n \n \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n28 \n \n \n  \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n29 \n \nFig. 4.  Competition and regrowth dynamics of wildtype (GTG) over ATG and \nATG* mutants. (A) Selection coefficient (square) was quantified from the slope of the \nlog ratios of competitor over wildtype strain per generation (circle represents mean \nand error bar represents SE over squares). ATG and ATG* lost the competition against \nthe wildtype as d etermined by negative selection coefficients that were significantly \ndifferent from the wildtype-wildtype control competition (s = -0.01, p<10-2 for ATG; s = \n-0.02, p<10-4 for ATG*, post hoc test and GLM) but not significantly different from each \nother (s = 0.007, post hoc test and GLM). mCherry (empty squares) and Venus \nfluorophores (filled squares) were used as markers to select for the respective strain \nbackground, and seven biological replicates across two independent experiments, \nincluding fluorophores  swaps, were performed for each combination. (B) Time to \nregrow from stationary phase. Data points (square), mean (circle), and SE of lag time \nto initiate exponential growth when revived from 16 hours overnight LB (fresh rich \nmedia) culture, 48 hours LB cu lture (aged rich media), and 48 hours old M9 glycerol \nculture (aged poor media). Analysis of variance of lag time delays for the three \nexperimental regrowth conditions showed a significant effect of strain (F = 9.4,  p<10-\n4, ANOVA) and a significant effect of regrowth condition (F = 7.8,  p<10-2, ANOVA). \nSample size was three to five biological replicates per condition. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint \n\n30 \n \n \nTable 1. Characterization of pulse expression, amplitude, duration, and \nfrequency in wildtype (GTG), TTG, ATG, and ATG* strains. \n \n.CC-BY 4.0 International licensemade available under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is \nThe copyright holder for this preprintthis version posted October 1, 2024. ; https://doi.org/10.1101/2024.09.30.615870doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}