Gene network analysis identifies a central post-transcriptional regulator of cellular stress survival

Cells adapt to shifts in their environment by remodeling transcription. Measuring changes in transcription at the genome scale is now routine, but defining the functional significance of individual genes within large gene expression datasets remains a major challenge. We applied a network-based algorithm to interrogate publicly available gene expression data to predict genes that serve major functional roles in Caulobacter crescentus stress survival. This approach identified GsrN, a conserved small RNA that is directly activated by the general stress sigma factor, σT, and functions as a potent post-transcriptional regulator of survival across distinct conditions including osmotic and oxidative stress. Under hydrogen peroxide stress, GsrN protects cells by base pairing with the leader of katG mRNA and activating expression of KatG catalase/peroxidase protein. We conclude that GsrN convenes a post-transcriptional layer of gene expression that serves a central functional role in Caulobacter stress physiology.


INTRODUCTION 28
Organisms must control gene expression to maintain homeostasis. A common mode of gene 29 regulation in bacteria involves activation of alternative sigma factors (σ), which redirect RNA 30 polymerase to transcribe genes required for adaptation to particular environmental conditions. 31 Alphaproteobacteria utilize an extracytoplasmic function (ECF) σ factor to initiate a gene expression 32 program known as the general stress response (GSR) ( Figure 1A). The GSR activates transcription of level, such as those controlled by small RNAs (sRNAs). Roles for sRNAs in bacterial stress response 47 systems are well described (Wagner and Romby, 2015), but remain unexplored in the 48 alphaproteobacterial GSR. 49 50 sRNAs typically function as repressors, though the regulatory roles and mechanisms of action of these 51 molecules are diverse: sRNAs can control gene expression by protein sequestration, modulation of 52 mRNA stability, transcription termination, or promotion of translation (Wagner and Romby, 2015). 53 The system properties of environmental response networks are often influenced by sRNAs, which can 54 affect the dynamics of gene expression via feedback ( We applied a rank--based network analysis approach to predict the most functionally significant genes 62 in the Caulobacter GSR regulon. This analysis led to the prediction that a sRNA, which we name GsrN, 63 is a major genetic determinant of growth and survival under stress. We validated this prediction, 64 demonstrating that gsrN is under direct control of σ T and functions as a potent post--transcriptional 65 regulator of survival across distinct conditions including hydrogen peroxide stress and hyperosmotic 66 shock. We developed a novel forward biochemical approach to identify direct molecular targets of 67 GsrN and discovered that peroxide stress survival is mediated through an interaction between GsrN 68 and the 5' leader sequence of katG, which activates KatG catalase/peroxidase expression. This post--69 transcriptional connection between σ T and katG, a major determinant of peroxide stress and 70 stationary phase survival (Italiani et al., 2011;Steinman et al., 1997), explains the peroxide sensitivity 71 phenotype of Caulobacter strains lacking a GSR system.

73
Finally, we demonstrate that RNA processing and sRNA--mRNA target interactions shape the pool of 74 functional GsrN in the cell, and that changes in GsrN expression enhance expression of some proteins 75 while inhibiting others. The broad regulatory capabilities of GsrN are reflected in the fact that a gsrN 76 deletion strain has survival defects across chemically--and physically--distinct stress conditions, and 77 support a model in which the GSR initiates layered transcriptional and post--transcriptional regulatory 78 responses to ensure environmental stress survival. 79

Iterative rank analysis of gene expression data identifies a small RNA regulator of stress survival 81
We applied a network--based analytical approach to interrogate published transcriptomic datasets 82 (Fang et al., 2013) and predict new functional genetic components of the Caulobacter GSR system. 83 We organized expression data for over 4000 genes ( Figure 1B and Figure 1--source data 1) to create a 84 weighted network. In our basic network construction, each gene in the genome was represented as a 85 node and each node was linked to every other node by a correlation coefficient that quantified the 86 strength of co--expression across all datasets ( Figure 1C). Within this undirected graph, we aimed to 87 uncover a GSR clique and thus more explicitly define the core functional components of the GSR 88 regulon.

90
To identify uncharacterized genes that are strongly associated with the GSR, we utilized an iterative 91 ranking approach related to the well--known PageRank algorithm (Brin and Page, 1998). We defined 92 the "input" set as the experimentally--defined regulators of σ T ( Figure 1D), optimized parameters 93 through a systematic self--predictability approach (Figure 1--figure supplement 1A and Materials and 94 methods--Iterative rank parameter tuning), and applied iterative ranking to compute a ranked list of 95 genes with strong associations to the input set ( Figure 1--source data 2). We narrowed our ranked list 96 by performing a promoter motif search to predict direct targets of σ T . A gene encoding an sRNA with 97 a consensus σ T binding site in its promoter, ccna_R0081 (Landt et al., 2008), was a top hit in our rank 98 list. We hereafter refer to this gene as gsrN (general stress response non--coding RNA).

100
To test whether gsrN transcription requires the GSR sigma factor, σ T , we generated a transcriptional 101 reporter by fusing the gsrN promoter to lacZ (P gsrN lacZ). Transcription from P gsrN required sigT ( Figure  102 A notable feature of GsrN is the presence of two isoforms by Northern blot. Probes complementary 148 to the 5' portion of GsrN reveal full--length (≈100 nucleotide) and short (51 to 54 nucleotides) 149 isoforms while probes complementary to the 3' portion reveal mostly full--length GsrN ( Figure 3A and 150 To test the function of the 5' portion of GsrN, we integrated a gsrN allele that contains only the first 172 58 nucleotides (Δ59--106), and lacks the transcriptional terminator (gsrNΔ3') into the vanA locus 173 ( Figure 4A). This short gsrN allele complemented the ∆gsrN peroxide survival defect ( Figure 4B). The 174 gsrNΔ3' allele produced a 5' isoform that was comparable in size and concentration to the wild--type 175 5' gsrN isoform. Since the transcriptional terminator of gsrN was removed, we also observed a run--on 176 ~2 00nt transcript from gsrNΔ3' ( Figure 4C).

178
To test the necessity of the 5' portion of GsrN in peroxide stress survival, we deleted nucleotides 10 179 to 50 of gsrN at its native locus ( Figure 4D). The gsrNΔ5' strain had a peroxide viability defect that 180 was equivalent to ΔgsrN. Ectopic expression of either full--length gsrN or gsrNΔ3' in the gsrNΔ5' strain 181 complemented its peroxide survival defect ( Figure 4E).

183
Several RNAs, including katG mRNA, co--purify with GsrN 184 We developed a forward biochemical approach to identify molecular partners of GsrN. The 185 Pseudomonas phage7 (PP7) genome contains hairpin (PP7hp) aptamers that bind to PP7 coat protein 186 (PP7cp) with nanomolar affinity (Lim and Peabody, 2002). We inserted the PP7hp aptamer into PP7hp fusion transcripts were enriched in our purification ( Figure 5B). Electrophoretic separation of 202 the eluate followed by silver staining revealed no significant protein differences between 203 GsrN(37)::PP7hp and the negative control (data not shown). We identified and quantified co--eluting 204 RNAs by RNA--seq.

206
We employed two approaches to identify RNAs enriched in GsrN(37)::PP7hp fractions relative to the 207 negative control fractions. A conventional RNA--seq pipeline (Tjaden, 2015) quantified mapped reads 208 within annotated gene boundaries as a first pass ( Figure 5C and Figure 5--source data 1). To capture 209 reads in non--coding and unannotated regions, and to analyze reads unevenly distributed across 210 genes, we also developed a sliding window analysis approach. Specifically, we organized the 211 Caulobacter genome into 25 base--pair windows and quantified mapped reads in each window using 212 the EDGE--pro/DESeq pipeline (Anders and Huber, 2010;Magoc et al., 2013). Together these two 213 quantification strategies identified several mRNA, sRNAs, and untranslated regions enriched in the 214 GsrN(37)::PP7hp pull--down fraction ( Figure 5D and Figure 5--source data 2). We applied IntaRNA 215 (Mann et al., 2017) to identify potential binding sites between GsrN and the enriched co--purifying 216 Most bacterial sRNAs regulate gene expression at the transcript and/or protein levels through 240 Watson--Crick base pairing with the 5'end of their mRNA targets (Wagner and Romby, 2015). We 241 sought to test whether GsrN affected the expression of katG. GsrN did not effect katG transcription in 242 exponential or stationary phases, or in the presence of peroxide as measured by a katG--lacZ 243 transcriptional fusion (Figure 6--figure supplements 1A--C). However, katG is transcriptionally 244 regulated by the activator OxyR, which binds upstream of the predicted katG --35 site (Italiani et al., 245 2011). To decouple the effects of OxyR and GsrN on katG expression we generated a strict katG 246 translational reporter that contains the mRNA leader of katG fused to lacZ (katG--lacZ) constitutively 247 expressed from a σ RpoD --dependent promoter. In both exponential and stationary phases, katG--lacZ 248 activity is reduced in ∆gsrN and enhanced in gsrN ++ strains compared to wild type ( supplement 1E). We conclude that GsrN enhances KatG protein expression, but not katG 251 transcription.

253
We then used this translational reporter to investigate a predicted binding interaction between the 254 unpaired 5' loop of GsrN and a G--rich region at the 5' end of the katG transcript. Specifically, the first 255 7 nucleotides of katG mRNA (Zhou et al., 2015) is complementary to 7 nucleotides in the single--256 stranded 5' loop of GsrN, including 4 of the 6 cytosines ( Figure 6A). We disrupted this predicted base 257 pairing, mutating 5 of the 7 nucleotides in the putative katG target site and GsrN interaction loop. 258 These mutations preserved GC--content, but reversed and swapped (RS) the interacting nucleotides 259 ( Figure 6A). We predicted that pairs of wild--type and RS mutant transcripts would not interact, while 260 base pairing interactions would be restored between RS mutant pairs. 261 262 Mutating the predicted target site in the katG 5' leader ablated GsrN--dependent regulation of the   To assess the relative effects of GsrN on katG transcript and protein levels in vivo, we directly 298 measured both by dot blot and Western blot, respectively. In untreated and peroxide treated 299 cultures, katG transcript levels trended lower in ∆gsrN and higher in gsrN ++ compared to wild type. 300 These differences are not statistically significant (Student's t--test, p=0.39) in untreated cultures; 301 however, KatG protein tagged with the M2 epitope was reduced 2--fold in ∆gsrN lysates relative to Steady--state katG transcript levels differ significantly between ∆gsrN and gsrN ++ in peroxide treated 307 cultures (Student's t--test, p<0.01) ( Figure 7A). KatG protein tagged with the M2 epitope was reduced 308 3--fold in peroxide treated cells in ∆gsrN lysates relative to wild--type; KatG--M2 levels in gsrN ++ were 309 increased in both untreated and peroxide treated cells ( Figure 7B). These data support a model   katG was the only gene that that was significantly enriched in the pull--down and differentially 332 expressed in the proteomic studies ( Figure 8A). These results provide additional evidence that katG is 333 a major target of GsrN, and that GsrN functions to enhance KatG expression at the post--334 transcriptional level.

336
Given our transcriptomic and proteomic datasets, we reasoned that GsrN may contribute to other 337 phenotypes associated with deletion of the GSR sigma factor, sigT. Indeed, the ∆gsrN mutant has a 338 survival defect after exposure to hyperosmotic stress, similar to ΔsigT ( Figure 8D). As we observed for 339 peroxide stress, overexpression of gsrN protects cells under this physicochemically--distinct condition. 340 Hyperosmotic stress survival does not require katG ( Figure 8D), providing evidence that a separate 341 GsrN regulatory target mediates this response. Unlike hydrogen peroxide ( Figure 7A), hyperosmotic 342 stress induces GsrN expression ( Figure 8E). This is consistent with previous transcriptomic studies in 343 Caulobacter in which hyperosmotic stress, but not peroxide stress, activated GSR transcription 344 The GSR system is broadly conserved in Alphaproteobacteria. Given the importance of GsrN as a post--351 transcriptional regulator of the Caulobacter GSR, we reasoned that functionally--related sRNAs might 352 be a conserved feature of the GSR in this clade. To identify potential orthologs of gsrN, we surveyed 353 the genomes of Alphaproteobacteria that encoded regulatory components of the GSR system and for 354 which transcriptomic data were publically available.

DISCUSSION 372
We sought to understand how GSR transcription determines cell survival across a spectrum of 373 chemical and physical conditions. To this end, we developed a directed gene network analysis 374 approach to predict genes with significant functional roles in the Caulobacter GSR. Our approach led 375 to the discovery of gsrN, a small RNA of previously unknown function that is a major post--376 transcriptional regulator of stress physiology. 377 378

Role of GsrN in mitigating hydrogen peroxide stress 379
Hydrogen peroxide can arise naturally from the environment and is also produced as an aerobic  This stated, the difference in hydrogen peroxide susceptibility between ∆sigT and ∆gsrN ( Figure 1E) 396 may be explained in part by the fact that dps is still expressed in ∆gsrN.

398
Post--transcriptional gene regulation by GsrN is a central feature of the general stress response 399 Alternative sigma factor regulation is a major mechanism underlying transcriptional reprogramming 400 PP7hp insertion mutants with low 5' isoform levels did not complement the peroxide viability defect 448 of ∆gsrN. Processing to a short 5' isoform may be necessary for GsrN to bind katG mRNA and regulate 449 KatG expression. Alternatively, cleavage may not be required for function, and lack of 450 complementation by certain hairpin insertion mutants may be due to PP7hp interfering with target 451 recognition or simply reducing total levels of GsrN. Regardless, our data clearly show that GsrN is 452 cleaved in a regular fashion to yield a 5' isoform that is very stable in the cell ( 1C). GSR activation during this period potentially protects cells from endogenous stressors that arise 463 from upregulation of anabolic systems required for growth and replication. In the future, it is of 464 interest to explore the hypothesis that the GSR system provides both basal protection against 465 endogenous stressors generated as a function of normal metabolism, and induced protection against 466 particular stressors (e.g. hyperosmotic stress) encountered in the external environment. 467 C. crescentus was cultivated on peptone--yeast extract (PYE)--agar (0.2% peptone, 0.1% yeast extract, 478 1.5% agar, 1 mM MgSO 4 , 0.5 mM CaCl 2 ) (Ely, 1991) at 30°C. Antibiotics were used at the following 479 concentrations on this solid medium: kanamycin 25 µg/ml; tetracycline 1 µg/ml; and chloramphenicol 480 2 µg/ml.

Strain construction 496
All C. crescentus experiments were conducted using strain CB15 (Poindexter, 1964) and derivatives 497 thereof. Plasmids were conjugated into CB15 (Ely, 1991) using the E. coli helper strain FC3 (Finan et 498 al., 1986). Conjugations were performed by mixing the donor E. coli strain, FC3, and the CB15 499 recipient strain in a 1:1:5 ratio. Mixed cells were pelleted for 2 minutes at 15,000xg, resuspended in 500 100 µL, and spotted on a nonselective PYE--agar plate for 12--24 hours. Exconjugants containing the 501 desired plasmid were spread on PYE agar containing the plasmid--specified antibiotic for selection. 502 The antibiotic nalidixic acid (20 µg/ml) was used to counter--select against both E. coli strains (helper 503 and plasmid donor).

505
Gene deletion and nucleotide substitution strains were generated using the integrating plasmid 506 pNPTS138 (Ried and Collmer, 1987). pNPTS138 transformation and integration occurs at a 507 chromosomal site homologous to the insertion sequence in pNPTS138. Exconjugants with pNPTS138 508 plasmids were selected on PYE agar plates with 5 µg/ml kanamycin; 20 µg/ml nalidixic acid selected 509 against the E. coli donor strain. Single colony exconjugants were inoculated into liquid PYE or M2X for 510 6--16 hours in a rolling 30°C incubator for non--selective growth. Nonselective liquid growth allows for 511 a second recombination event to occur, which either restores the native locus or replaces the native 512 locus with the insertion sequence that was engineered into pNPTS138. Counter--selection for the 513 second recombination of pNPTS138 was carried out on PYE agar with 3% (w/v) sucrose. This selects 514 for loss of the sacB gene during the second crossover event. Colonies were subjected to PCR 515 genotyping and/or sequencing to identify to confirm the allele replacement. Protein expression plasmid pMal was used to express a maltose binding protein (MBP) fused to the N--560 terminus of a Pseudomonas Phage 7 coat protein fused to a His--tag at its C--terminus (to generate 561 MBP--PP7CP--His). The PP7CPHis protein sequence was amplified out of pET283xFlagPP7CPHis and 562 inserted into pMal at SalI and EcoRI restriction sites. pET283xFlagPP7CPHis was a gift from Alex 563 Ruthenburg and originates from Kathleen Collins (Addgene plasmid # 28174). 564

Experimental Method Details 565
Hydrogen peroxide/osmotic stress assays 566 Liquid cultures were passaged several times before stress treatment to insure that population growth 567 rate and density was as consistent as possible prior to addition of hydrogen peroxide (oxidative 568 stress) or sucrose (hyperosmotic stress). Briefly, starter cultures were inoculated in liquid M2X 569 medium from colonies grown on PYE--agar plates. Cultures were grown overnight at 30°C in a rolling 570 incubator. Overnight cultures were then diluted back to an optical density reading of 0.05 at 660 nm 571 (OD 660 =0.05) and grown in a rolling incubator at 30°C for 7--10 hours. After this period, cultures were 572 re--diluted with M2X to OD 660 =0.025 and grown for 16 hours at 30°C in a rolling incubator. After this 573 period, OD 660 was consistently 0.85--0.90. These cultures were then diluted to OD 660 =0.05 and grown 574 for 1 hour and split into two tubes. One tube received stress treatment and the other tube was 575 untreated. Treated cultures were subjected to either hydrogen peroxide or sucrose. 576 577 For stress treatment, we used a freshly prepared 10 mM H 2 O 2 solution diluted from a 30% (w/w) 578 stock bottle (stock never more than 3 months old) or a stock of 80% (w/v) sucrose. The amount of 10 579 mM H 2 O 2 added for stress perturbation depended on the volume of the culture and the desired final initially diluting into 96--well plates. 5 μL spots from each dilution were plated on PYE--agar. Once 585 spots dried, plates were incubated at 30°C for 2 days. Clearly visible colonies begin to form after 36 586 hours in the incubator.

588
The difference in colony forming units (CFU) between treated and untreated cultures was calculated 589 using the following formula: (1) Where represents the countable (resolvable) dilution in which colonies are found in the treated 591 sample dilution series and y represents the untreated sample dilution. 592 593

β--galactosidase gene expression reporter assays 594
To assess reporter gene expression, liquid cultures were passaged several times as described in the 595 hydrogen peroxide/osmotic stress assays section above. However, cultures were placed in a 30°C 596 shaker instead of a 30°C rolling incubator. Exponential phase cultures were harvested when the last 597  Chloroform--treated cells were vortexed for 5--10 seconds to facilitate permeabilization. Z buffer and 609 ONPG were added directly to chloroform--permeabilized cells. Reactions were incubated in the dark at 610 room temperature and quenched with 1 mL of 1 M Na 2 CO 3 .

612
Each reporter construct was optimized with different reaction times and different volumes of cells. Miller units were calculated as: Where A 420 is the absorbance of the quenched reaction measured at 420 nm on a Spectronic Genesys 623 20 spectrophotometer (ThermoFisher Scientific, Waltham, MA). A 660 is the optical density of the 624 culture of cells used for the assay. is time in minutes between the addition of ONPG to the time of 625 quenching with Na 2 CO 3 . is the volume in milliliters of the culture added to the reaction. Intensity of GsrN bands or katG mRNA dots was calculated by dividing the probe signal specific to 680 GsrN or katG mRNA over the probe signal specific to the 5S rRNA multiplied by 100. Normalization of 681 katG mRNA specific probes in the dot blot was carried out in a manner similar to that described for 682 Northern blot, in which the 5S rRNA probe signal was used for normalization.

Rifampicin transcription inhibition assays 685
Liquid C. crescentus CB15 cultures were passaged in the same manner outlined in the hydrogen 686 peroxide/osmotic stress assays section. However, cells for transcription inhibition assays were grown were rediluted and grown to OD 600 =0.6. Cells were then induced with 1mM IPTG for 5 hours and spun 739 down at 8000g at 4°C for 10 minutes. The cell pellet was resuspended in 6 mL of ice--cold lysis buffer 740 (125 mM NaCl, 25 mM Tris--HCl pH 7.5, 10 mM Imidazole) and mechanically lysed in a LV1 741 Microfluidizer. Lysate was immediately added to 500 μL of amylose resin slurry that was prewashed 742 with ice--cold lysis buffer. After the sample was loaded, beads were washed in 50x bead volume 743

Acid--Phenol RNA extraction 761
Samples for acid--phenol extractions were mixed with equal volumes of acid--phenol and vortexed 762 intermittently at room temperature for 10 minutes. Phenol mixture was spun down for 15 minutes at 763 maximum speed at 4°C. The aqueous phase was extracted, cleaned with an equal volume of 764 chloroform, and spun down for 15 minutes at maximum speed at 4°C. The aqueous phase was 765 extracted from the organic and equal volumes of 100% isopropanol were added. Linear acrylamide 766 was added to the isopropanol precipitation to improve pelleting (1 μL per 100 μL of isopropanol 767 sample). Samples were then incubated at --20°C overnight and spun down at 15000xg for 15 minutes 768 at 4°C. The isopropanol was aspirated, the pellet washed in 1 mL of 75% ethanol, and sample spun 769 again at 15000xg for 15 minutes at 4°C. Ethanol was removed from the RNA pellet, and pellet was left 770 to dry for 15 minutes. Pellet was resuspended in 25 μL of nuclease--free H 2 O. 771 772

RNA dot blot analysis 773
Samples (≈3 μg) for dot blot analysis were mixed with equal volumes of 2x RNA loading dye as in a 774 Northern Blot, and heated for 8 minutes at 75°C. Samples were then spotted on a Zeta--Probe Blotting 775 Membrane and left to dry for 30 minutes. Spotted membrane was then subjected to two doses of 776 120 mJ/cm 2 UV radiation (Stratalinker UV crosslinker). The membrane was then prehybridized 2 777 times for 30 minutes in hybridization buffer at 65°C in a rotating hybridization oven. After pre--778 hybridization, we added radiolabeled oligonucleotide probes. Hybridization buffer with probes was 779 always prepared so that each probe's concentration was approximately 1 nM. katG mRNA was first 780 hybridized for 16 hours at 65°C in a rotating hybridization oven. Membrane was then washed with 781 wash buffer three times, 20 minutes each at 65°C in a rotating hybridization oven. The blot was 782 exposed for 48 hours to a Molecular Dynamics Phosphor screen and imaged on a Personal Molecular 783 Imager as described above. Membrane was subsequently stripped with two rounds of boiling in 0.1% 784 SDS solution and incubated for 30 minutes at 65°C in a rotating hybridization oven. Following 785 stripping, the membrane was subjected to two rounds of prehybridization and then hybridized for 16 786 hours at 65°C in a rotating hybridization oven with the probe specific to the 5' end of GsrN. 787 Membrane was then washed again with wash buffer three times for 20 minutes each at 65°C in a 788 rotating hybridization oven. This GsrN blot was exposed for 36 hours to the phosphor screen and 789 imaged. The membrane was stripped four times after GsrN probe exposure. Following stripping, 790 membrane was again subjected to two rounds of prehybridization and then hybridized for 16 hours at 791 65°C in a rotating hybridization oven with the probe specific to 5S rRNA. Membrane washed with 792 Wash Buffer three times, 20 minutes each at 65°C in a rotating hybridization oven. This 5S RNA blot 793 was exposed to the phosphor screen for 1 hour and imaged. Tween 20) overnight at room temperature on a rotating platform. Primary incubation with a 810 DYKDDDDK(i.e. M2)--Tag Monoclonal Antibody (clone FG4R) was carried out for 3 hours in 5% 811 powdered milk TBST at room temperature on a rotating platform (4 μL antibody in 12 mL). 812 Membrane was then washed 3 times in TBST for 15 minutes each at room temperature on a rotating 813 platform. Secondary incubation with Goat anti--Mouse IgG (H+L) Secondary Antibody, HRP was for 1 814 hour at room temperature on a rotating platform (3 μL antibody in 15 mL). Finally, membrane was 815 washed 3 times in TBST for 15 minutes each at room temperature on a rotating platform. 816 Chemiluminescence was performed using the SuperSignal™ West Femto Maximum Sensitivity 817 Substrate and was imaged using a ChemiDoc MP Imaging System version 6.0. Chemiluminescence 818 was measured using the ChemSens program with an exposure time of 2 minutes. 819 820 Western blot lane normalization of KatG--M2 specific bands was conducted by normalizing total signal 821 from the doublet signal in the M2 specific background to that of the non--specific band (found in 822 strains were there was no M2 tagged KatG). Samples extracted on the same day were run on the 823 same gel. Lane normalized samples were then normalized to the levels of KatG--M2 signal in the wild--824 type untreated samples for that specific gel.

RNA--seq preparation 827
Total RNA was extracted from cultures passaged similarly to the hydrogen peroxide/osmotic stress 828 assays section. However, cultures were harvested at OD 660 =0.85--0.90 from the 16--hour overnight 829 growth. Total RNA extraction followed the procedure outlined in the TRIzol extraction section. We computed Pearson's correlation coefficient based on normalized expression between all pairwise 872 combinations of genes. Correlation coefficients were organized into a numpy.matrix data structure 873 where each row and column corresponds to the same gene order. Correlation coefficients less than 0 874 were not considered for this analysis and were assigned the value 0. We refer to this matrix as the 875 Rho--matrix. The Rho--matrix is symmetric and the product of its diagonal is 1. The Rho--matrix 876 represents the weighted edges of the network, where the value of 0 demonstrates no edge is drawn 877 between nodes.

879
A one--dimensional weight matrix that corresponds to the rows and columns of the Rho--matrix was 880 constructed as a numpy.matrix data structure with all values initialized at 0. Lastly, a key array was 881 constructed in conjunction with the Rho--matrix and weight--matrix for initializing the assignment of 882 weight and obtaining the final weights of the algorithm. The weight--matrix represents the weight of 883 the nodes of the network and the key matrix represents the gene name of the node. 884 885

Iterative Ranking: Matrices and Algorithms 886
Iterative ranking algorithms are a class of analytical tools used to understand relationships between 887 nodes of a given network. The iterative ranking algorithm used to dissect the general stress response 888 in the transcription--based network follows: 889 890 Given the Rho--matrix (Ρ) and weight--matrix ( ), the weight--matrix after --iterations is Equation 8.
For Equation 8, let ∝ represent a dampening factor applied to the initialize = 0 weight of the 893 nodes, ! . The final weights of the weight--matrix as → ∞ converge to a stable solution, Equation 9 .
Algorithm and solution information was adapted from (Wang and Marcotte, 2010).

897
Initial weight--matrix, ( ! ), was created by assigning the weight 1.0 to the corresponding positions of 898 the seven genes known to regulate the General Stress Response (GSR) of C. crescentus: sigT, phyR, 899 phyK, sigU, nepR, lovR, and lovK. Normalization of the values of the Rho--matrix, Ρ, was performed by 900 normalizing each column such that each column has a sum equal to 1 and then repeating the same 901 normalization process by rows. 902 903

Iterative rank parameter tuning 904
Iterative rank parameters were optimized through the self--prediction of known associated 905 components of the General Stress Response (GSR). Variables tuned for exploration were the ∝ 906 parameter and the reduction of the number of edges based on correlation cut--offs. We chose to base 907 our parameters on which condition best predicted the gene phyR, when initializing the weight--matrix 908 with sigT, sigU, nepR, phyK, lovR, and lovK values of 1. Varying these two parameters showed that an 909 edge reduction of ρ > 0.9 and an alpha factor greater than 0.5 yielded the highest rank for phyR 910 (Figure 1--figure supplement 1A).

Identification of σ T --promoter motifs 918
Motif finder utilized a python script that scans 200 nucleotides upstream of annotated transcriptional 919 start sites (Zhou et al., 2015) or predicted translational start sites (TSS) (Marks et al., 2010). 920 921 We built a simple python library to take in genomic FASTA files, find specified regions of interest, and 922 extract 200 nucleotides from a given strand. We used the Caulobacter crescentus NA1000 annotation 923 (CP001340) from NCBI as the input genomic file and used the predicted TSS (when available) or 924 Output from IntaRNA comprised a csv file of target binding sites and the corresponding GsrN binding 942 sites. We extract the predicted binding sites of the targets with a python script and parsed the targets 943 into those predicted to bind the first exposed loop and the second exposed loop. Sequence logos were generated from (Crooks et al., 2004) 945 946

Phylogenetic tree construction 947
A 16S rRNA phylogenetic tree of Alphaproteobacteria was constructed by extracting 16S rRNA 948 sequences for all species listed in Figure S7A  fractions. In some cases, reads mapped outside coding sequences. Such reads mapped proximal to 987 the 5' end of annotated genes and to intergenic regions. We observed uneven read distribution 988 across some annotated genes. Cases in which reads were not evenly distributed across a gene were 989 typically not classified as significantly different from the control samples in "Expression" or "qValue" 990 by Rockhopper even when a clear bias in read density was visually evident (most often at the 5' end 991 of the gene).

993
As a second approach, we performed a systematic window annotation analysis to capture the 994 unaccounted read density differences between the two purified fractions (GsrN(37)::PP7hp and the 995 PP7hp--GsrN--3' negative control). Windows were generated by in silico fragmentation of the C. 996 crescentus NA1000 genome sequence, designating 25 base pair windows across the genome. We Multiple t--tests were conducted using all 6 LFQ value obtained across the three MaxQuant runs. We 1058 used the multiple t--test analysis from GraphPad Prism version 6.04 for MacOS, GraphPad Software, 1059 La Jolla California USA, www.graphpad.com. The false discovery rate (Q) value was set to 5.000% and 1060 each row was analyzed individually, without assuming a consistent SD. 1061   Samples on right were treated with 0.2 mM hydrogen peroxide before RNA extraction. Blots were 1345

Key Reagent And Resource
hybridized with katG mRNA, GsrN or 5S rRNA probes. katG mRNA signal normalized to 5S rRNA signal 1346 is quantified (mean ± SD, n=3, p--value estimated with Student's t--test    removed from the network.

1407
(B) The number of edges drawn for a given node shows that an edge reduction of 0.9 dramatically 1408 shifts the average number of edges per node.