The hallmarks of a tradeoff in transcriptomes that balances stress and growth functions

ABSTRACT Fast growth phenotypes are achieved through optimal transcriptomic allocation, in which cells must balance tradeoffs in resource allocation between diverse functions. One such balance between stress readiness and unbridled growth in E. coli has been termed the fear versus greed (f/g) tradeoff. Two specific RNA polymerase (RNAP) mutations observed in adaptation to fast growth have been previously shown to affect the f/g tradeoff, suggesting that genetic adaptations may be primed to control f/g resource allocation. Here, we conduct a greatly expanded study of the genetic control of the f/g tradeoff across diverse conditions. We introduced 12 RNA polymerase (RNAP) mutations commonly acquired during adaptive laboratory evolution (ALE) and obtained expression profiles of each. We found that these single RNAP mutation strains resulted in large shifts in the f/g tradeoff primarily in the RpoS regulon and ribosomal genes, likely through modifying RNAP-DNA interactions. Two of these mutations additionally caused condition-specific transcriptional adaptations. While this tradeoff was previously characterized by the RpoS regulon and ribosomal expression, we find that the GAD regulon plays an important role in stress readiness and ppGpp in translation activity, expanding the scope of the tradeoff. A phylogenetic analysis found the greed-related genes of the tradeoff present in numerous bacterial species. The results suggest that the f/g tradeoff represents a general principle of transcriptome allocation in bacteria where small genetic changes can result in large phenotypic adaptations to growth conditions. IMPORTANCE To increase growth, E. coli must raise ribosomal content at the expense of non-growth functions. Previous studies have linked RNAP mutations to this transcriptional shift and increased growth but were focused on only two mutations found in the protein’s central region. RNAP mutations, however, commonly occur over a large structural range. To explore RNAP mutations’ impact, we have introduced 12 RNAP mutations found in laboratory evolution experiments and obtained expression profiles of each. The mutations nearly universally increased growth rates by adjusting said tradeoff away from non-growth functions. In addition to this shift, a few caused condition-specific adaptations. We explored the prevalence of this tradeoff across phylogeny and found it to be a widespread and conserved trend among bacteria.


Introduction
Maintaining optimal fitness in microorganisms requires navigating tradeoffs in resource allocation 1 . Expression profiling has shown that increasing ribosomal content ("greedy" genes) increases growth rates and additionally that decreasing the expression of stress response genes ("fearful" genes) is highly correlated with increasing growth rates in bacteria 1 . Thus, the fear vs. greed (f/g) tradeoff emerges as a key feature of transcriptome reallocation, in which cells that favor faster growth face the cost of diminished responsiveness to stresses 2 .
The f/g tradeoff was recently demonstrated as a property of the E. coli transcriptome through the use of a novel transcriptomic analysis method that defines sets of genes that are independently modulated, forming data-driven regulons termed iModulons 3 . The f/g tradeoff is characterized by the strong negative correlation between the activity levels of the RpoS (fear) and Translation (greed) iModulons. The f/g tradeoff involves an upregulation of ribosomal proteins that often are the limiting factor for increasing growth rate 4 and a concurrent downregulation of stress-related genes. The tradeoff involves competition between the housekeeping and stress sigma factors (RpoD and RpoS), binding of ppGpp and DksA, and other regulatory mechanisms [5][6][7] . Many of these mechanisms directly involve RNA polymerase (RNAP).
Furthermore, the f/g tradeoff has been observed as an evolutionary adjustment in various studies [8][9][10] and has been well documented in adaptive laboratory evolution (ALE) experiments 3,8,11,12 . In ALEs, strains are grown and propagated in the same condition for long periods of time, which creates a selection pressure that encourages endpoint strains to be enriched for mutations which increase growth rate. Therefore, ALE strains are expected to be as "greedy" as possible under the conditions used.
RNAP mutations have been shown to balance the f/g tradeoff and RNAP is one of the most common mutation targets during ALEs 13 . In a detailed study of two RNAP mutations found in the catalytic center, it was hypothesized these RNAP mutations adjust the tradeoff by destabilizing the rpoB-rpoC interface, thus affecting the binding of ppGpp to RNAP 2 . While many ALE mutations cluster in the catalytic center of RNAP, there are numerous other RNAP mutations found in ALE endpoint strains. These mutations can be found near regulator binding sites, regions known to be related to antibiotic resistance, important structural elements such as the flap domain and trigger loop, and in regions with no clear annotations 14-22 . Previous to this study, the transcriptional impact of these RNAP mutations far from the catalytic core was unclear. Convergent RNAP mutations have been found in specific stress adaptation experiments [23][24][25][26] often leading to the assumption that RNAP mutations reflect media adaptations, missing their underlying role in the f/g tradeoff. Here, we sought to expand our knowledge of the f/g tradeoff through a multi-scale study. We introduced twelve RNAP mutations and used computer simulations to infer how these mutations destabilize the rpoB-rpoC interface. We then obtained transcriptomes in various experimental conditions and used iModulon analysis to demonstrate that, despite structurally distinct locations, these mutations universally downregulate stress-related genes and upregulate growth-related genes in addition to some condition-specific adaptations. Finally, we compared the transcriptomes of various species to find that that f/g tradeoff is widely found across phylogeny. Thus, our multi-scale study elucidated key features of a central transcriptomic tradeoff between fear and greed and proposed that it is a general principle in microbiology.

Figure 1 -RNAP mutations alter the fear vs. greed tradeoff. (A)
The structure of RNAP (PDB 6OUL 27 ) is visualized using PyRosetta 28 , showing the location of mutations used in this study and highlighting some specific RNAP regions of interest [14][15][16][17][18][19][20][21][22] . The grouped mutations on the upper left are some of the most common mutations found in ALEdb 13 and are further discussed in Figure 2. (B) Laboratory evolution leads to sequence variants which adjust the composition of the transcriptome leading to faster growth and repressed stress readiness. The f/g tradeoff on the transcriptome is shown (RpoS represents fear, Translation represents greed) along with the mutations' qualitative impact on growth/stress readiness in the vertical axis visualized by the green curve. (C) The twelve mutations studied are listed, and the samples with the "misc RNAP mut" label are miscellaneous conditions in PRECISE 2.0 29 with RNAP mutations 13,29 .
RNA-sequencing data was collected under aerobic growth on glucose M9 minimal media (see Methods). Some of the RNAP mutant strains were tested under specific stress conditions that were similar to the ALE experiment in which they were originally found (see Supplemental Table 2). All but one of the 12 mutants increased the growth rate and exhibited a shift toward greed in the f/g tradeoff in the transcriptome, as detailed below (Figure 2). The exception, rpoB I966S, arose during an evolution to high temperature growth 30 and may therefore have had a stronger impact on temperature stability than regulation of expression.
RNAP mutations destabilize the rpoB-rpoC interface and affect ligand binding RNAP mutations have been shown to affect RNAP structurally in a variety of ways. Some of the most commonly found and widespread RNAP mutations are rpoC N720H, rpoB G1189C, rpoB P1100Q, and rpoB E672K (Figure 2). The physical mechanism for how these four mutations cause the tradeoff is not fully established, but some key properties are known. Structurally, they are all located near the rpoB-rpoC interface (rpoB E672K = 5.46 Å, rpoB P1100Q = 5.24 Å, rpoB G1189C = 8.97 Å, rpoC N720H = 10.09 Å). PyRosetta 28 was used to calculate the impact of these mutations on the holoenzyme and found that all destabilized the rpoB-rpoC interface (rpoB E672K = -28.40 REU, rpoB P1100Q = -23.98 REU, rpoB G1189C = -5.26 REU, rpoC 1055V = -13.91 REU). This structural destabilization likely affects the ppGpp binding site and thus modifies its regulatory role 17 .
RNAP mutations modify RNAP's interactions with sigma factors. Genes regulated by RpoS 31 , the general stress response sigma factor, showed on average a -0.33 change in log 2 transcripts per million (tpm) expression when compared to the wild-type. The relatively small change (-0.059 change in log 2 tpm) in genes regulated by RpoD, the housekeeping sigma factor, shows that these mutations differentially affect sigma factor functions (see Supplemental  Table 3). The analysis of global changes in the transcriptome is difficult due to the high number of differentially expressed genes in many comparisons. Furthermore, comparing many conditions is challenging if pairwise differential expression of genes (DEG) plots are used 32 (see Supplemental Figure 1). To overcome these challenges, we used the iModulon workflow 3,29 to identify independently modulated gene sets (iModulons) and interpret their differential activity between all conditions used. This workflow uses independent component analysis (ICA) of a compendium (X) of RNA-sequencing data, which includes our samples of interest along with a variety of other experiments which help to separate source signals associated with transcriptional regulators 3,29 . The algorithm generates two output matrices: M (whose columns highlight the genes in each iModulon) and A (whose rows show the iModulon's activity in every sample). Detailed information on each iModulon is available at iModulonDB.org 33 and this study focuses primarily on the "E. coli PRECISE 2.0" dataset 29 ; an E. coli database of RNA-sequencing data obtained under 422 growth conditions. Principal component analysis (PCA) of the iModulon activity matrix (A) shows that much of its variance and thus expression variation in general is explained by the RpoS and Translation iModulons' activities (see Supplemental Table 4). The RpoS iModulon is the largest and the Translation iModulon is the fifth largest contributing factor to the highest variance explaining principal component (PC). GadX and ppGpp iModulons are also highly contributing factors to large variance explaining PC's, adding additional dimensionality to f/g that is further explored in Figure 3. The f/g tradeoff is thus a major contributor to variation in the composition of the transcriptome.
The new RNA-sequencing data from the twelve new RNAP mutant strains was analyzed using ICA 3 . The iModulon activity levels in the new samples were compared to those in PRECISE 2.0. This database was used to compute the iModulons structure of the E. coli transcriptome 3 and the gene composition of the key fear and greed iModulons is found in Supplemental Table 5. All of the four common ALEdb mutations strongly downregulate the activity of the RpoS iModulon and upregulate the activity of the Translation iModulon.
All of the twelve mutations introduced, except for rpoB I966S, have a large impact on the activity level of the RpoS iModulon similar to the two previously studied RNAP mutations 2 . The mutations in the catalytic center have the largest impact on RpoS iModulon activity levels, but mutations distant from this location can also strongly impact the activity of this iModulon. This suggests there is more complexity to the physical mechanism of this transcriptomic effect than the destabilization of the rpoB-rpoC interface. Both the frequency of occurrence and the effect of these mutations imply they are commonly fixed during growth rate selection (i.e., maximization of 'greed'). Thus, a number of structurally similar mutations have a strong and similar effect on the composition of the transcriptome.
Genome-scale models of proteome allocation quantitatively estimates the growth benefit of maximizing greed functions While iModulons are an informative approach to reveal the hallmarks of changes in the expression state, they are not directly representative of the composition of the proteome. Creating iModulons from expression data requires the input RNA-sequencing data to be both centered to a control and normalized. This means the activity levels of iModulons for samples are entirely relative to each other and their magnitude range is constrained by the variance of the PRECISE dataset. We thus deployed a genome-scale model to reproduce the f/g tradeoff which allowed us to infer absolute measures of the proteome of cells undergoing said tradeoff. A genome-scale metabolism and expression (ME) model 34 was run to maximize growth while constraining RpoS iModulon-associated reactions to a specified lower bound.
The resulting RpoS and Translation iModulons' proteomic computed mass fractions were highly anticorrelated (-0.9994) (see Supplemental Figure 2). A unit activity increase in the Translation iModulon has a 650% stronger effect on said iModulons' genes' proteome mass fraction than it does in the RpoS iModulon (see Methods). This implies that the small activity increases of the Translation iModulon seen in the f/g tradeoff and in the RNAP mutations may be having a larger effect than appears on the cell's phenotype. This computational model also indicates that forced expression of the stress readiness genes reduce the expression of the growth promoting genes as experimentally observed.

The fear vs. greed tradeoff additionally involves GAD and ppGpp iModulons
The f/g tradeoff was first visualized using the activity levels of the Translation and RpoS iModulons 3 . Since this study was published, the number of transcriptomes for E. coli has quadrupled 35 . The analysis of the larger data sets reveals additional dimensionalities to the f/g tradeoff. Several additional iModulon activity levels are correlated with growth rates, including the GadX and ppGpp iModulons. The former is related to acid stress and the latter to protein translation rates and the stringent response. GadX is highly correlated with RpoS (0.71) and negatively correlated with growth rates (-0.30) while ppGpp is strongly correlated with Translation (0.74) and has a weak positive correlation with growth rates (0.16). Correlation plots for each of these iModulon activity pairings are given in Figure 3A-F.

Figure 3 -Reflections of the fear vs. greed tradeoff transcriptome in the relative activity levels of the translation and stress iModulons. (A-F) These plots show the relationship in activity levels between the greed (Translation and ppGpp) and fear (RpoS and GadX) iModulons. (G) The activity levels of various growth-and stress-related iModulons for the RNAP mutants, along with some other iModulons highly affected by said mutations. The gray dots are the activity levels of the other iModulons for all of the mutants. Red labeled iModulons are plotted in panels A-F.
While the core group of common RNAP mutations downregulate stress-related iModulons and upregulate growth-related iModulons (Figure 3G), other RNAP mutations have more specific effects that are adaptations to the environments from which they were found. Supplemental Figure 4 shows two of these such mutations (rpoB R200P and rpoA G315V) from our set of twelve mutations.
The RpoB R200P mutation reflects a specific selection condition. It is found commonly in replicate methionine tolerance evolutions 25 and it has two effects on the transcriptome: (1) during growth on methionine it activates the Translation iModulon and downregulates the RpoS iModulon to increase the growth rate compared to wild-type; and (2) during growth on M9 glucose it activates anaerobic response genes found in the Fnr-1, Fnr-2, and Anaero-related iModulons. Methionine contains sulfur and is thus a common target of reactive oxygen species (ROS) in E. coli 36 . The RpoA G315V mutation, which is further discussed in the Supplemental Information, adjusts the transcriptome similarly to a crp mutation 37 and is commonly found in a pgi replacement evolution where it appears to be an adaptation to a failed replacement 24 .
Thus, there are RNAP mutations outside the core of the enzyme that reflect condition-specific effects on the transcriptome (see Supplemental Information for more cases). This observation leads to a wider examination of the effects of synergistic mutations with RNAP mutations that are selected under specific conditions. The genetic basis for the fear vs. greed tradeoff during ALE is condition-dependent The fear and greed iModulons are correlated for both unevolved samples (-0.57 correlation) and evolved samples (-0.39 correlation, see Figure 4A), although evolved samples strongly favor greed. Different stressors lead to specific transcriptional adjustments along the f/g tradeoff to best favor growth as is annotated in Figure 4A.
In most laboratory evolutions with high stress conditions, evolution downregulates the RpoS iModulon over time. The cells initially use the RpoS iModulon to respond to nearly any stress, but eventually tune the stress response to the specific environment. In a reaction oxygen species experiment (labeled ROS TALE) 38 , initially the RpoS iModulon was highly active but as the cells evolved on paraquat most of the iModulon was downregulated while the expression of oxidative response genes in the iModulon were left largely unmodified (see Supplemental Figure 3). This transcriptional regulatory network adjustment, which was driven by convergent mutations, enabled the cells to grow faster in a ROS stress environment.
This evolutionary pull towards growth, and thus greed, in ALEs necessitates condition-specific convergent synergistic mutations ( Figure 4B). These mutations are not strictly limited to RNAP. For example, oxyR is a common mutation target for evolution in oxidative stress 8 and a topA mutation was a convergent target in a heat tolerance evolution 39 . This trend is widespread among ALEs, as 89% of experiments in ALEdb contain at least one gene that is mutated in 50% or more of their endpoint strains. If excluding RNAP genes, this drops to 80%. While RNAP mutations have the ability to favor growth across a wide variety of conditions, different genes are often better mutational targets for specific conditions.

The fear vs. greed tradeoff is found in WT and across growth conditions
This ceaseless pull towards greed and away from stress readiness, however, is largely limited to laboratory conditions. The lack of overlap between the natural variants 40 and the ALEdb mutations seen in Figure 4C implies that there are highly divergent evolutionary pressures on wild-type strains and their ALE counterparts.
It should be noted, however, that f/g changes are not limited to mutations acquired during evolution, as Figure 4D shows how the transcriptome composition falls on the tradeoff line in nutrient limited growth. Furthermore, when limiting nutrients drive the culture into the stationary phase, a time series of points shows how the transcriptome composition moves down the f/g tradeoff line. This movement shows how lower growth rates on entry into the stationary phase come with an increase in stress readiness. The f/g tradeoff is thus reflected in the various growth states of the WT strain as well as a transition in physiological states.
The fear vs. greed tradeoff is found across the phylogenetic tree Finally, we searched the phylogenetic tree for other organisms exhibiting the f/g tradeoff. First we analyzed data from a multi-strain E. coli ALE study 41 . This analysis shows that the tradeoff was found in all the E. coli strains of the study (Figure 5A). Second, we examined iModulonDB 33 for the presence of the f/g in other species (Figure 5D-K). The tradeoff was found in seven bacteria. Although the gene composition of the fear iModulons varies between species (likely a consequence of differing stresses in their natural environments), all of the primary greed iModulons consist of a highly similar set of ribosomal subunits and translational associated functions (Figure 5B-C). The presence of the f/g tradeoff across such a wide range of species implies that may be a global property of bacterial transcriptomes. 41  33 . The p-value is calculated using a t-distribution Wald Test. The names of the iModulons are pulled from their respective data sets. Mycobacterium tuberculosis' "Positive regulation of growth" iModulon mostly consists of stress-related antitoxin genes.

Discussion
We detail a general tradeoff in the bacterial transcriptome between growth rate and stress readiness. A major genetic component of this tradeoff lies in RNAP mutations, which affect the structure of RNAP and consequently the composition of the transcriptome. In RNAP mutants that arise from ALE studies, the modified transcriptome composition favors transcription of growth-related functions over stress-related functions. The tradeoff between fear and greed related functions was found across a wide range of wild-type bacterial strains. Interestingly, the fear vs. greed tradeoff has been described in many areas of science; such as economics 44 , game theory 45 , and psychology 46 . It has been elucidated here for microbiology through a multi-scale analysis.
A previous study compared two RNAP mutations 2 , rpoB E672K and rpoB E546V, and found that they destabilize the rpoB-rpoC interface 47 . Another study using in vitro assays linked an rpoC deletion from 3,611 to 3,619 bp to destabilizing the open complex of RNAP 48 . This destabilization came with decreased transcriptional pausing on the promoter, reduced RNAP's open complex half-life, and increased elongation rates 48 . The impact of these mutations have been shown to be similar to strains with modified ribosomal operons, suggesting that these mutations are likely modifying ribosomal availability and/or distribution 9 . A recent study analyzing 45,000 ALE mutations and comparing them to wild-type mutations suggests that the wild-type mutations are under negative selection pressure, while ALE mutations are under positive selection pressure. This suggests that ALE mutations represent extreme mutations extenuating a preferred trait, thus amplifying the basis for the f/g tradeoff 40 .
The current study expanded upon current knowledge 2,48 by analyzing the impact of twelve RNAP mutations to detail RNAP's role as a global master regulator of the f/g tradeoff. The detailed molecular/structural mechanisms that underlie the tradeoff are not fully understood, but appear to involve destabilization of the rpoB-rpoC interface 2 , altered kinetic and regulatory properties 48 , and changes in the sigma factor use of RNAP.
The effects that RNAP mutations have on the transcriptome composition, however, are clear. The transcriptomic re-allocation involves a consistent set of iModulons with known functions. The relationship between the proteome and transcriptome functions enable genome-scale computational biology assessment of the phenotypic consequences of the reallocation 49 . Thus, a detailed understanding of the effects of the f/g tradeoff at the systems level has emerged. As the tradeoff involves resource allocations for improved fitness, it is important to contextualize particular RNAP mutations fixed in laboratory evolution studies and seek to identify epistatic mutations that are condition specific.
Finally, the phylogenetic distribution of the f/g tradoff is broad, suggesting that this tradeoff may emerge as a universal feature of the bacterial transcriptome. Thus, one can conjecture that this tradeoff was present in the last universal bacterial ancestor, and that RNAP may have played a role as a master transcriptional regulator in early life.

Competing interests
The authors have no competing interests to declare.

Creation of RNAP Mutations
RNAP mutations were created according to the protocol outlined in Zhao et al. 50 .

RNA-sequencing
All samples were prepared and collected in biological duplicates. 3 ml of culture was added to 6 ml of Qiagen RNA-protect Bacteria Reagent after sample collection. This solution was then vortexed for 5 seconds, incubated at room temperature for 5 minutes, and then centrifuged. The supernatant was then removed and the cell pellet was stored at -80°C. The Zymo Research Quick-RNA MicroPrep Kit was used to extract RNA from the cell pellets per vendor protocol. On-columns DNase treatment was performed for 30 minutes at room temperature. Anti-rRNA DNA Oligo mix and Hybridase Thermostable RNase H 51 was used to remove ribosomal RNA. Sequencing libraries were created using a Kapa Biosystems RNA HyperPrep per vendor protocol. RNA-sequencing reads were processed using https://github.com/avsastry/modulome-workflow. Data is available at NCBI GEO GSE227624.
iModulon Computation RNA-sequencing data was used to create iModulon activity levels of the mutated strains using PyModulon 3 which is available at https://github.com/SBRG/pymodulon. Activities of iModulons were compared to samples from PRECISE 2.0 29 which is easily accessible using iModulonDB 33 .

Mutation Analysis
ALEdb 13 was used for selecting the mutations for this study. Any E. coli strains on ALEdb were considered as potential sources for mutations. Mutations from the same sample but where one is from an isolate and one is from the population were considered to be just one instance of said mutation.

Structural Analysis
Structural analysis was performed using PyRosetta 28 using its default score function by mutating each residue in place and repacking the area within 10 Angstroms around it. The pdb files were downloaded from RCSB 52 . REU stands for Rosetta Energy Unit, which is PyRosetta's unit for energy. For the holoenzyme mutation impacts, the pdb files used for the holoenzyme were 1L9U, 2A6E, 2CW0, 4YG2, and 4MEY, which were selected based on a review of bacterial RNAP 53 .

Metabolic Model and Proteomic Calculations
The FoldME 34 model was used for the metabolic modeling calculations. Supplemental Figure 2 was generated by iteratively increasing the lower bounds for the genes of the RpoS iModulon and recording the proteomic mass fraction of the Translation and RpoS iModulons' genes until the model no longer ran. Proteome mass fraction to iModulon genes is the sum of the measured proteomic mass fractions of each enriched gene in an iModulon. This value is calculated for every sample and plotted against its corresponding PRECISE iModulon Activity. The proteomic calculations performed for this paper are well described in Patel et al 49 .

Supplemental Information
Other RNAP mutations of special interested from our study The rpoA G315V mutation affects the activities of Crp-1 and Crp-2 iModulons with the strongest impact on the maltose operons. This mutation was found in a pgi synthetic gene replacement ALE 24 in nearly all strains that failed to integrate the exogenous pgi replacements. Presumably the loss of pgi required large changes to sugar import systems, thus necessitating this rpoA mutation to help downregulate maltose importers 54 . The mutation's effect on the Crp-1 iModulon is similar to one reported in a study that deactivated regions of crp 37 (see Supplemental Figure 4E). This similarity increases the likelihood that the mechanism of action for this rpoA mutation is to modify the rpoA-crp binding interface.
Some mutations predicted to have a specific stress response show no clear existence of one in our study. RpoC H419P was commonly found in octanoic acid tolerance studies and thus RNA-sequencing data was gathered for it both on M9 and on octanoic acid. Initially the rpoC H419P mutants had trouble growing on octanoic acid until the concentration was lowered, so perhaps an octanoic acid-specific adjustment would have been seen if a higher acid concentration was used. RpoC H419P has been previously introduced into E. coli and shown to be an adaptation to octanoic acid 26 , so this issue is likely limited to our study.
RpoB I966S was commonly selected for in heat tolerance studies 39,55,56 and thus we chose to grow it in our study on high temperature conditions. These heat tolerance studies, however, were all carried out using E. coli B-strains and there was little effect of this mutation on the RNA-sequencing data so it appears this is a strain specific adaptation.

Supplemental Tables/Figures
Supplemental       Figure 1 -Differential expression genes (DEG) plot between wild-type and the mutated strains colored by COG categories. The median expression value from the mutated strains was used for the mutated strain values. The pairwise single mutant strain compared to the wild-type strain versions of this plot look similar. Interpreting these individual plots is highly difficult and doing so for all these plots together is nearly impossible, thus necessitating the use of iModulon analysis.
Supplemental Figure 2 -Constrained ME-model simulation predicts fear vs. greed tradeoff. Reactions associated with the Translation iModulon's genes were tightly controlled in a ME model simulation, resulting in a corresponding change in the proteomic mass fraction in both the Translation iModulon and the RpoS iModulon. Growth rates increased nominally (<1%) as RpoS decreased. Note that given the large proteomic fraction allocated to the Translation iModulon compared to that of the RpoS iModulon, its fold-changes are numerically much smaller, but represent a notable proteome reallocation.
Supplemental Figure 3 -RpoS iModulon's genes downregulating specifically over the evolution. As the cells evolve on 250 µM paraquat, many genes are downregulated from the RpoS iModulon to enable higher growth, but those related to oxidative stress are not (those