Global Hfq-mediated RNA interactome of nitrogen starved Escherichia coli uncovers a conserved post-transcriptional regulatory axis required for optimal growth recovery

Abstract The RNA binding protein Hfq has a central role in the post-transcription control of gene expression in many bacteria. Numerous studies have mapped the transcriptome-wide Hfq-mediated RNA–RNA interactions in growing bacteria or bacteria that have entered short-term growth-arrest. To what extent post-transcriptional regulation underpins gene expression in growth-arrested bacteria remains unknown. Here, we used nitrogen (N) starvation as a model to study the Hfq-mediated RNA interactome as Escherichia coli enter, experience, and exit long-term growth arrest. We observe that the Hfq-mediated RNA interactome undergoes extensive changes during N starvation, with the conserved SdsR sRNA making the most interactions with different mRNA targets exclusively in long-term N-starved E. coli. Taking a proteomics approach, we reveal that in growth-arrested cells SdsR influences gene expression far beyond its direct mRNA targets. We demonstrate that the absence of SdsR significantly compromises the ability of the mutant bacteria to recover growth competitively from the long-term N-starved state and uncover a conserved post-transcriptional regulatory axis which underpins this process.


Introduction
In all three domains of life, post-transcriptional regulation of mRNA by small regulatory RNA (sRNA) represents an important mechanism to fine tune the flow of genetic information.In gram-negative bacteria such as Esc heric hia coli , sRNAs posttranscriptionally regulate more than half of all mRNAs, and many of these interactions are facilitated by the activity of an RNA binding protein (RBP), especially Hfq and ProQ (1)(2)(3)(4).
Hfq is a hexameric Sm / Lsm-type RBP that is conserved in many bacteria.Both the protein itself and its associated sRNAs have been studied in most detail ( 1 ,5 ).Most Hfqdependent sRNAs act at the ribosome binding site (RBS) by short imperfect base pairing, most commonly to repress but in some cases also to activate mRNA translation (2)(3)(4)(5)(6)(7)(8).Further, sRNA can bind internal to a target mRNA, or within its 3 UTR.This type of interaction can either protect a ribonuclease cleavage site and thereby stabilise the mRNA to maximise chances of it being translated ( 9 ,10 ) or contribute to the recruitment of the mRNA to the RNA degradosome complex for degradation ( 11 ).
Individual sRNAs typically regulate multiple target mRNAs and, conversely, a target mRNA can be regulated by multiple different sRNAs.Further, some sRNAs interact with other sRNAs; this process, called 'sponging', represents a mechanism of regulating sRNAs by sequestering the sRNA in an inactive complex or causing their degradation ( 12 ).Additionally, sRNAs can be sponged by mRNAs ( 12 ), preventing the regulatory action of the sRNA, whilst having little regulatory effect on the mRNA.Together, this results in a complex network usually referred to as the Hfq-mediated RNA-RNA interactome, which if captured by a suitable method will predict a large part of the post-transcriptional regulation of gene expression taking place under the growth condition of interest.Recently, several methods have been developed for the capture of RNA-RNA interactomes and successfully applied to Esc heric hia coli , Salmonella enterica , Vibrio cholerae , Staphylococcus aureus , Clostridioides difficile and Pseudomonas aeruginosa (13)(14)(15)(16)(17)(18)(19)(20)(21).Importantly, these studies have underscored the view that post-transcriptional regulation confers enormous regulatory flexibility, fine-tuning and diversity to bacterial gene expression.One of these methods, RIL-seq (RNA interaction by ligation and sequencing) is a recently developed methodology, which allows genome wide mapping of RNA-RNA interactions through ligating interacting RNA molecules on RNA binding proteins (RBPs) such as Hfq ( 22 ).
Our current understanding of Hfq-mediated RNA-RNA interactomes is restricted to actively growing bacteria, bacteria experiencing iron limitation or bacteria that have entered short-term growth-arrest ( 4 , 17 , 20 ).It is widely accepted that active growth is a luxury that bacteria can only occasionally afford.Whether in a host during infection or any other natural environment, bacteria frequently exist in a growtharrested state.To what extent Hfq-mediated RNA interactomes contribute to gene expression in growth-arrested bacteria or recovery remains unknown.Nitrogen (N) is an essential constituent for the biosynthesis of proteins and nucleic acids and N-starved bacteria are prevalent in many mammalianintestinal systems and several fresh water-, terrestrial-and marine-ecosystems ( 23 ,24 ).Whereas the transcriptional basis of the adaptive response to N starvation has been broadly studied ( 25 ,26 ), its post-transcriptional regulatory basis is rel-atively poorly understood, and only one sRNA is linked the adaptive response to N starvation (see later).To address the gaps in our understanding of the roles of the Hfq-mediated RNA-RNA interactome in growth-arrested bacteria and in the adaptive response to N starvation, we applied RIL-seq to identify and compare the Hfq-mediated RNA-RNA interactome in E. coli at four different growth states: bacteria growing exponentially in N replete conditions (N+), short-( ∼20 min; N-) and long-( ∼24 h; N-24) term N-starved bacteria and bacteria ∼2 h following growth-recovery from N starvation (N-24 + 2).We reveal that the Hfq mediated RNA-RNA interactome is extensive and temporally dynamic in growtharrested bacteria.By comparing how the Hfq mediated RNA-RNA interactome impacts gene expression at a proteome wide scale in long-term growth-arrested bacteria, we uncover a conserved post-transcriptional regulatory axis in growth-arrested bacteria that underpins growth recovery from long-term N starvation.

Bacterial strains and plasmids
The sdsR E. coli strain MG1655 was constructed using λ red recombination.The sdsR and sdsR yeaG BW25311 mutant E. coli strains were created by P1 phage transduction of the sdsR deletion into either WT or yeaG BW25311 strains.All plasmids used in this study are listed in Supplementary Table S1 .p5UTR-yeaG, containing a translational GFP fusion of the regulatory region of yeaG was constructed as described previously ( 27 ) using PCR products amplified from gDNA (positions −93 to +63 relative to AUG; A is +1).Inserts were restricted with NheI and NsiI and ligated into an equally treated pXG10 plasmid backbone.Plasmid variants p5UTR-yeaG MUT and pSdsR COMP were constructed by PCR-based site-directed mutagenesis of p5UTR-yeaG and pSdsR respectively.pSdsR was constructed using pKF68-3 ( 28 ) as a template, with the Salmonella sdsR changed for E. coli sdsR using PCR products amplified from gDNA and constructed by Gibson assembly ( 29 ).pBR322-sdsR was constructed by Gibson assembly using PCR product amplified from the region surrounding and including sdsR ( ∼350 bp upstream to ∼250 bp downstream of sdsR ).

Bacterial growth conditions
Bacteria were grown in Gutnick minimal medium (33.8 m M KH 2 PO 4 , 77.5 mM K 2 HPO 4 , 5.74 m M K 2 SO 4 , 0.41 mM MgSO 4 ) supplemented with Ho-LE trace elements ( 30 ), 0.4% (w / v) glucose as the sole carbon source and NH 4 Cl as the sole N source.Overnight cultures were grown at 37 • C, 180 rpm in Gutnick minimal medium containing 10 m M NH 4 Cl.For the N starvation experiments and recovery experiments, 3 m M NH 4 Cl was used.Unless stated otherwise, for all growth assays bacteria were subcultured in growth media in a 48well plate for a starting OD 600nm of 0.05, and OD 600nm was measured every 15 min in either a SPECTROstar OMEGA or SPECTROstar Nano plate reader (BMG LABTECH).The proportion of viable cells in the bacterial population was determined by measuring colony forming units (CFU) from serial dilutions on lysogeny broth agar plates.Overexpression experiments used pBAD18y eaG , pBAD18y eaG -K426A (or pBAD18empty as the empty vector control), in Gutnick minimal medium supplemented with 0.125% (w / v) l -arabinose at an OD 600nm of approximately 0.8 , for induction of yeaG expression.

RIL-seq experimental procedure
RIL-seq experiments were performed as described in ( 17 ,22 ).Briefly, E. coli strains carrying either WT or 3XFLAG tagged hfq were grown in Gutnick minimal medium with 3 mM NH 4 Cl until OD 600 of 0.3 (N+), ∼20 min following growth arrest (N-), 24 h following growth arrest (N-24) and ∼2 h following addition of 10 mM NH 4 Cl to N-24 cultures (N-24 + 2).Approximately 80 ODs were taken from three biological replicates.The samples were cross-linked under a 256 nm UV light source and pelleted in ice-cold 1xPBS.Pellets were lysed in NP-T buffer (50 mM NaH 2 PO 4 , 300 mM NaCl, 0.05% Tween, pH 8.0) supplemented with 1:200 protease inhibitor cocktail Set III, EDTA-free (Calbiochem, #539134) and RNase inhibitor (final concentration of 0.1 U / μl) (Takara, #2313A).Lysates were incubated with anti-Flag (M2 monoclonal antibody, Sigma-Aldrich, #F1804) bound protein A / G magnetic beads (Thermo-Fisher) for 2 h at 4 • C with rotation followed by three washing steps with lysis buffer.Beads were treated with an RNase A / T1 mix for 5 min at 22 • C in an RNase-inhibitor free lysis buffer.Samples were washed three times with lysis buffer supplemented with 3.25 μl of SUPERase In RNase inhibitor (Thermo-Fisher, #AM2696, 20 U / μl).The trimmed ends of RNAs were cured by PNK treatment (New England Biolabs, #M0201) for 2 h at 22 • C with agitation, followed by 2 washing steps at 4 • C. Hfq-bound RNA duplexes were ligated with T4 RNA ligase I enzyme in the following buffer: 8 μl T4 ligase buffer, 7.2 μl DMSO, 0.8 μl ATP (100 mM), 32 μl PEG 8000, 1.2 μl RNase inhibitor, 23.6 μl of water, 140 U of T4 RNA ligase I enzyme (New England Biolabs, #M0437M).Samples were incubated O / N at 22 • C with agitation, followed by three steps of washing with lysis buffer, at 4 • C. The RNAs were eluted from beads with a proteinase K (Thermo Fisher Scientific, #2313A) digestion for 2 h at 55 • C followed by LS Trizol extraction, as per manufacturer instruction.Purified RNA was resuspended in 7 μl of nuclease-free water and quality controlled on a Bioanalyzer picoRNA chip.RNA samples were sent to Vertis Biotechnologie AG for downstream processing.Briefly, oligonucleotide adapters were ligated to the 5 and 3 ends of the small RNA samples.First-strand cDNA synthesis was performed using M-MLV reverse transcriptase and the 3 adapter as primer.The resulting cDNA was amplified with PCR using a highfidelity DNA polymerase and TruSeq-designed primers compatible with Illumina sequencing chemistry.The barcoded libraries were equimolarly pooled and sequenced with an Illumina NextSeq 500 in paired-end mode with 2 × 40 cycles.

RIL-seq analysis and data presentation
Raw read pairs were quality and adapter trimmed via Cutadapt ( 42 ) version 2.5 in paired-end mode using a cutoff Phred score of 20.Read pairs without any remaining bases in at least one read of a pair were discarded (parameters: -nextseq-trim = 20 -m 1 -a A GATCGGAA GA GCA CA CGTCTGAA CTCCA GTCA C-A A GATCGGAA GA GCGTCGTGTA GGGAAA GA GTGT).Processed read pairs were further analysed using the RILseq software package ( 17 ,22 ) version 0.81 ( https://github.com/asafpr/RILseq; see https://github.com/yairra/RILseqfor latest versions) installed via Bioconda ( 31 )).In brief, read pairs were aligned to the E. coli K12 MG1655 reference genome (RefSeq assembly accession: GCF_000005845.2) using the map_single_fragments.py script with default parameters.A transcript file (TF) was generated via gener-ate_transcripts_gff.py based on an EcoCyc (version 25.5) data folder with parameter -BC_chrlist COLI-K12,NC_000913.3 for chromosome ID mapping.Afterwards, chimera mapping with the same reference genome as above was performed separately for each of the previously generated BAM files using the map_chimeric_fragments.py script.For this, default parameters were applied except for using -t TF to incorporate the transcript file in the analysis.Finally, overrepresented interacting regions were identified using the RILseq_significant_regions.py script with default parameters except for including the EcoCyc data folder (-bc_dir), a RefSeq folder with a feature table for the assembly (-refseq_dir), the chromosome ID mapping (-BC_chrlist NC_000913.3,COLI-K12)and parameters for excluding rRNA interactions (-ribozero -rrna_list rRNA,RRNA).Summary statistics for sequencing and mapping were calculated as suggested in Supplementary Table 2 of ( 19 ) and can be found in Supplementary Data 1 .Coding regions were termed CDS, annotation of 5 UTRs and 3 UTRs (termed 5UTR and 3UTR in figures and tables) was based on annotation in EcoCyc.In cases where the transcription start or termination sites of a gene were unknown (termed EST3UTR / EST5UTR for estimated UTRs), the UTRs were considered as the regions 100 nt upstream the ATG and downstream the stop codon (or shorter if these regions spanned another transcript or were more likely to be a UTR of the neighbouring transcript).Intergenic regions were termed IGR if their boundary genes were not in the same transcript or IGT if the two genes were part of the same transcript.Small RNAs that are candidate regulatory RNAs as denoted by EcoCyc were grouped as sRNAs.Transcripts antisense to genes or to IGT were termed AS and AS_IGT respectively.The same annotation was maintained in all tables and figures unless noted otherwise.ygaM.EST3UTR was re-annotated as an sRNA.For the purpose of certain data analysis and data interpretation glnA.3UTR was considered to be GlnZ, ybaP.EST3UTR was considered to be ChiX, glmZ.hemY was considered to be GlmZ and nlpD was generally considered to be rpoS.Interaction types were defined as follows: mRNA:sRNA for chimeras with one sRNA and either a CDS, 3UTR, 5UTR, EST3UTR, EST5UTR or IGT, sRNA:sRNA for chimeras with two sRNA, sRNA:tRNA for chimeras with an sRNA and a tRNA, and other interactions were defined as all other interactions not categorised by the previous definitions (including all those involving AS or IGR fragments).Principle component analysis was performed on total RIL-seq datasets following both normalisation and standardisation of raw chimera counts using the factoextra package version 1.0.7 in R 4.2.3 and plotted using the scatterplot3d package version 0.3-41.For further downstream analysis and interpretation only interactions that were detected with at least 30 chimeras, and in least two replicates from the same time-point were considered.Chimera data from individual samples can be found in Supplementary Data 2 , and summary data from all samples can be found in Supplementary Data 3 .For certain comparisons between time-points and / or replicates the number of chimeras for each interaction were normalised such that the total number of chimeras in each replicate and time-point were the same (130677).Circos plots were generated using shinyCircos ( https:// venyao.xyz/shinyCircos-V1/ ) and then edited further in Adobe Illustrator 26.5.Further information on specific Circos plots are provided in their respective figure legends.sRNA-target binding predictions were performed using IntaRNA ( 32 ).Coverage plots of chimeras were generated using the script generate_BED_file_of_endpoints.py of the RIL-seq computational pipeline; the bed files were visualized with IGV 2.12.2, and further processed with Adobe Illustrator 26.5.

Proteomics sample preparation
The proteomics analysis presented in the worked was performed as part of a larger proteomics experiment consisting for WT and 4 different mutant E. coli strains.For totalproteome analysis WT and sdsR E. coli were grown to N-24, and 20 ml of culture was washed twice with ice cold PBS and flash frozen in liquid nitrogen before further processing.Pellets were suspended in 500 μl of lysis buffer (100 mM Tris-HCl (pH7.9), 150 mM NaCl, 1.5% SDS, 2 × cOmplete Protease Inhibitor mix (Roche)) and sonicated on ice for a total of 10 min at 30% intensity, in 10 s intervals.Insoluble protein and cell debris was pelleted by centrifugation for 20 min and supernatant was collected.All samples were adjusted to an equal protein concentration with more lysis buffer.10 μg of lysate derived from our control (WT strain), sdsR strain, and other samples not discussed in this study, were subjected to an in-solution tryptic digest using a modified version of the Single-Pot Solid-Phase-enhanced Sample Preparation (SP3) protocol ( 33 ,34 ).1% SDS-containing lysates were added to Sera-Mag Beads (Thermo Scientific, #4515-2105-050250, 6515-2105-050250) in 10 μl 1 5% formic acid and 30 μl of ethanol.Binding of proteins was achieved by shaking for 15 min at room temperature.SDS was removed by 4 subsequent washes with 200 μl of 70% ethanol.Proteins were digested overnight at room temperature with 0.4 μg of sequencing grade modified trypsin (Promega, #V5111) in 40 μl Hepes / NaOH, pH 8.4 in the presence of 1.25 mM TCEP and 5 mM chloroacetamide (Sigma-Aldrich, #C0267).Beads were separated, washed with 10 μl of an aqueous solution of 2% DMSO and the combined eluates were dried down.Peptides were reconstituted in 10 μl of H2O and reacted for 1 h at room temperature with TMT16pro labelling reagent (Thermo Scientific, #A44522).To this end, 50 μg of TMT16pro label reagent were dissolved in 4 μl of acetonitrile and added to the peptides.Excess TMT reagent was quenched by the addition of 4 μl of an aqueous 5% hydroxylamine solution (Sigma, 438227).Peptides were reconstituted in 0.1% formic acid and equal volumes were mixed.Mixed peptides were purified by a reverse phase clean-up step (OASIS HLB 96-well μElution Plate, Waters #186001828BA).Peptides were subjected to an off-line fractionation under high pH conditions ( 33 ) yielding 6 fractions.

LC-MS / MS analysis
Peptides were separated using an Ultimate 3000 nano RSLC system (Dionex) equipped with a trapping cartridge (Precolumn C18 PepMap100, 5 mm, 300 μm i.d., 5 μm, 100 Å) and an analytical column (Acclaim PepMap 100.75 × 50 cm C18, 3 mm, 100 Å) connected to a nanospray-Flex ion source.The peptides were loaded onto the trap column at 30 μl per min using solvent A (0.1% formic acid) and eluted using a gradient from 2 to 80% Solvent B (0.1% formic acid in acetonitrile) over 90 min at 0.3 μl / min (all solvents were of LC-MS grade).The Orbitrap Fusion Lumos was operated in positive ion mode with a spray voltage of 2.4 kV and capillary temperature of 275 • C. Full scan MS spectra with a mass range of 375-1500 m / z were acquired in profile mode using a resolution of 120 000 with a maximum injection time of 50 ms, AGC operated in standard mode and a RF lens setting of 30%.Fragmentation was triggered for 3 s cycle time for peptide like features with charge states of 2-7 on the MS scan (data-dependent acquisition).Precursors were isolated using the quadrupole with a window of 0.7 m / z and fragmented with a normalized collision energy of 34%.Fragment mass spectra were acquired in profile mode and a resolution of 30 000 in profile mode.Maximum injection time was set to 94 ms or an AGC target of 200%.The dynamic exclusion was set to 60 s.

Proteomics data analysis
Acquired data were analyzed using FragPipe ( 35 ) and a Uniprot E. coli FASTA database (UP000000625, E. coli strain K12, ID83333, 4402 entries, last modified October 27th 2022, downloaded January 11th 2023) including common contaminants.The following modifications were considered: Carbamidomethyl (C, fixed), TMT16plex (K, fixed), Acetyl (Nterm, variable), Oxidation (M, variable) and TMT16plex (Nterm, variable).The mass error tolerance for full scan MS spectra was set to 10 ppm and for MS / MS spectra to 0.02 Da.A maximum of 2 missed cleavages were allowed.A minimum of 2 unique peptides with a peptide length of at least seven amino acids and a false discovery rate < 0.01 were required on the peptide and protein level ( 36 ).

Proteomics statistical data analysis
The raw output files of FragPipe (protein.tsv-files, ( 35 )) were processed using the R programming language (ISBN 3-900051-07-0).Contaminants were filtered out and only proteins that were quantified with at least two unique peptides were considered for the analysis.2156 proteins passed the quality control filters.Log 2 transformed raw TMT reporter ion intensities were first cleaned for batch effects using the 'removeBatchEffects' function of the limma package ( 37 ) and further normalized using the vsn package (variance stabilization normalization - ( 38 )).Proteins were tested for differential expression using the limma package.The replicate information was added as a factor in the design matrix given as an argument to the 'lmFit' function of limma.Limma analysis for sdsR as compared to WT can be found in Supplementary Data 4 .

Immunoblotting
Immunoblotting was conducted in accordance to standard laboratory protocols, with primary antibodies incubated overnight at 4 • C. A rabbit anti-YeaG antibody was produced previously ( 39 ) and used at a dilution of 1:2000.The following other antibodies were used: mouse monoclonal anti-RpoA (Biolegend, WP003) at 1:1000 dilution, HRP Goat anti-mouse IgG (BioLegend, 405306) at 1 : 10, 000 dilution and HRP Goat anti-rabbit IgG (GE Healthcare, NA934) at 1:10 000 dilution.ECL Prime Western blotting detection reagent (GE Healthcare, RPN2232) was used to develop the blots, which were analysed on the ChemiDoc MP imaging system and bands quantified using Image Lab software.

SdsR target binding site analysis
Binding site designations for mRNA targets of SdsR identified in the RIL-seq dataset were determined based on the peak of chimeric alignments to their target RNA in IGV.Designations were defined as follows: RBS-40 bp upstream to 15 bp downstream of the AUG; Internal + 3 UTR-any site > 15 bp following the AUG and before any known terminator (this designation includes targets with multiple internal binding sites); 5 UTR-in the 5 UTR but > 40 bp upstream of the AUG; multiple sites -when chimeras map to more than one designation; poorly defined -when chimeras mapped to a very broad region of the mRNA or were found in an intergenic region within the same transcript.Targets where the interaction site of SdsR mapped in an intergenic region between transcripts were excluded from binding site analysis.

Scanlag
Bacteria were grown in 3 mM Gutnick minimal medium to N-24, washed twice in 1 ml PBS, diluted between 10 −5 and 10 −6 , and 100 μl spread on Gutnick minimal media containing 0.4% glucose, 3 mM NH 4 Cl, Ho-LE trace elements and 1.5% (wt / vol) agar.Plates were incubated at 33 • C in a standard office scanner (Epson Perfection V370 photo scanner, J232D) placed in an incubator, and images were taken every 20 min over a 48 h period.Analysis of appearance time and apparent growth rate of colonies was adapted from Levin-Reisman et al. ( 40 ) using a modified code available at https: // github.com/mountainpenguin/ NQBMatlab .

Competition assay
For bacterial competition assays, bacteria were grown to the desired growth state (N-or N-24) in Gutnick minimal media supplemented with 0.4% glucose and 3 mM NH 4 Cl.In all competition assays the sdsR strain was resistant to kanamycin, and the WT was not.For competition during recovery growth, WT and sdsR bacteria were both inoculated at a starting OD 600mn of 0.025 in Gutnick minimal media supplemented with 0.4% glucose and 3 mM NH 4 Cl, for a total starting OD 600nm of 0.05.CFUs were measured by serial dilution, at the point of inoculation and at an OD 600nm of 0.4 and 0.8 -plating on both LB agar and LB agar with 50 μg / ml of kanamycin; the CFUs on the plates containing kanamycin gives the number of sdsR bacteria, and the CFU on the plates without kanamycin give the total number of bacteria -the number of WT bacteria can then be calculated by subtracting the number of sdsR bacteria from the total number of bacteria.For competition during stationary phase, both WT and sdsR bacteria were resuspended in Gutnick minimal media supplemented with 0.1% glucose and no nitrogen source; OD 600nm was measured and wild-type and sdsR bacteria were combined to an equal OD 600nm for a total final volume of 20 ml.CFUs for total and sdsR bacteria were then determined at regular intervals as above.

SdsR YeaG-GFP reporter assay
sRNA-target regulation assays were performed as previous, with some modifications ( 27 ).Plasmids pCONT, pSdsR and pSdsR COMP contained either ∼50 nt nonsense RNA fragment SdsR, or SdsR containing a compensatory point mutation to yeaG MUT (see below), respectively, from an IPTG inducible promoter.Plasmids p5UTR -yeaG contained the 5 UTR and translational start codon of yeaG (positions −93 to +63 relative to AUG; A is +1) N-terminally fused to GFP from a constitutively active promoter, and p5UTR -yeaG MUT contained the same with the CG at residues −28 / 27 relative to the AUG mutated to GC.The required combinations of plasmids were transformed into sdsR MG1655 E. coli and overnight cultures were grown in LB containing 50 μg / ml kanamycin, 100 μg / ml ampicillin, 15 μg / ml chloramphenicol and supplemented with 0.4% glucose to repress expression of SdsR from pSdsR.All bacterial strains were then inoculated into 0.2 ml LB containing 50 μg / ml kanamycin, 100 μg / ml ampicillin, 15 μg / ml chloramphenicol, with 0.1 mM IPTG for induction of SdsR expression, at a starting OD 600nm of 0.05, in a blackwalled 96-well plate.Cultures were then grown in a SPEC-TROstar OMEGA plate reader at 37 • C for 10 h.OD 600nm and GFP fluorescence were then measured.

Lag time and doubling time
Lag time was calculated by generating a straight-line graph of the exponential growth section of each growth curve, and was defined as the intersection of the straight-line graph with the x axis.The doubling time was determined from the slope of logarithmic growth function.

RNA-sequencing and analysis
WT and sdsR bacteria were harvested at N-24.RNA was extracted using the RNAsnap protocol ( 41 ).Three biological replicates of each strain were taken and mixed with a phenol:ethanol (1:19) solution at a ratio of 9:1 (culture:solution) before harvesting the bacteria immediately by centrifugation.Pellets were resuspended in RNA extraction solution (18 mM EDTA, 0.025% SDS, 1% 2-mercaptoethanol, 95% formamide) and lysed at 95 • C for 10 min.Cell debris was pelleted by centrifuged.RNA was precipitated by addition of 1 / 10 volume of 3 M sodium acetate (pH 5.2) and 3 volumes of 100% ethanol, followed by incubation at −80 • C for at least 1 h.RNA was washed with 75% ethanol and resuspended in water.Analysis of extracted RNA was performed following depletion of ribosomal RNA molecules using a commercial rRNA depletion kit for mixed bacterial samples (Lexogen, RiboCop META, #125).The ribo-depleted RNA samples were first fragmented using ultrasound (4 pulses of 30 s at 4 • C).Then, an oligonucleotide adapter was ligated to the 3 end of the RNA molecules.First-strand cDNA synthesis was performed using M-MLV reverse transcriptase with the 3 adapter as primer.After purification, the 5 Illumina TruSeq sequencing adapter was ligated to the 3 end of the antisense cDNA.The resulting cDNA was PCR-amplified using a high-fidelity DNA polymerase and the barcoded TruSeq-libraries were pooled in approximately equimolar amounts.Sequences were generated on an Illumina NextSeq 2000 system in single-end mode using 101 cycles read length.Raw sequencing reads were subjected to quality and adapter trimming via Cutadapt ( 42 ) version 2.5 using a cutoff Phred score of 20 and discarding reads without any remaining bases (parameters:-nextseq-trim = 20m1-aA GATCGGAA GA GCA CA CGTCTGAA CTCCA GTCA C).Afterwards, all reads longer than 11 nt (-l 12) were aligned to the E. coli K12 MG1655 reference genome (RefSeq assembly accession: GCF_000005845.2) using the pipeline READemption ( 43 ) version 0.4.5 with segemehl version 0.2.0 ( 44 ) and an accuracy cut-off of 95% (-a 95).READemp-tion gene_quanti was applied to quantify aligned reads overlapping genomic features by at least 10 nt (-o 10) on the sense strand (-a) based on RefSeq annotations (CDS, ncRNA, rRNA, tRNA) for assembly GCF_000005845.2 from 11 March 2022.The number of reads mapping to each gene was calculated, and a matrix of read counts was generated.The matrix was analysed using the DESeq2 BioConductor package (1.38.3) for differential gene expression analysis, with log-fold change shrinkage applied with lfcShrink ( Supplementary Data 5 ) ( 45 ).
The relative abundance of SdsR at N-, N-24 and N-24 + 2 growth states was obtained from RNA-seq datasets of WT bacteria from the corresponding growth states.The RNA for these RNA-seq datasets were obtained as described above but the extracted RNA was sequenced by Vertis Biotechnologies, Germany.

Results
RIL-seq reveals an extensive and dynamic RNA-RNA interactome in N starved E. coli RIL-seq allows isolation and sequencing of ligated RNA-RNA pairs through immunoprecipitation of Hfq-RNA complexes (see Materials and Methods).To immunoprecipitate Hfq, we used an E. coli strain, which contained a 3xFLAGtag sequence fused C-terminally to hfq at its normal chromosomal location.Control experiments revealed that the growth dynamics ( Supplementary Figure S1 A) and survival ( Supplementary Figure S1 B) during growth arrest of E. coli strains without and with 3xFLAG were indistinguishable, indicating that the modification to Hfq did not compromise its activity under our experimental conditions.We then performed RIL-seq in N+, N-, N-24 and N-24 + 2 E. coli , grown in a defined minimal growth medium with a limiting ( ∼3 mM) amount of ammonium chloride as the only N source.N + bacteria were sampled from the exponential phase of growth at an OD 600 of ∼0.3 (Figure 1 A).Growth arrest occurs as soon as the ammonium chloride in the media runs outs ( 26 ,46-48 ) and we considered bacteria ∼20 min after N runout to be in the N-state.The N-24 bacteria were sampled ∼24 h following N run-out.Growth recovery was induced by supplying ∼3 mM ammonium chloride to N-24 bacteria and N-24 + 2 bacteria were collected following 2 h.Details on the sequencing depth and mapping statistics are listed in Supplementary Data 1 .The data was analysed as described in by Matera et al., ( 15 ) ( Supplementary Data 2 & 3 ).Statistically significant chimeras from the Hfq precipitated showed an abundance of sRNAs and mRNAs over the control sample (Figure 1 B and Supplementary Figure S1 C).Principal component analysis (PCA) showed that Hfq-mediated RNA-RNA interactomes from the individual replicates from each growth stage clustered together, underscoring the stringency and reproducibility of the sampling and experimental approach (Figure 1 C).We note that chimeras from N + and N-24 + 2 bacteria were closely clustered, contrasting the chimeras from N-and N-24 bacteria, which were represented by two distinct clusters, suggesting that the Hfq-mediated RNA-RNA interactome dramatically differ between growing and growtharrested bacteria and in bacteria in different stages of growtharrest (see later).Only RNA pairs represented by at least 30 chimeric fragments in at least two replicates were considered for downstream analysis, giving ∼315, ∼354, ∼379, ∼279 in-teractions in N+, N-, N-24 and N-24 + 2 bacteria, respectively ( Supplementary Data 1 & 2 ).Most of these corresponded to sRNA-mRNA chimeras (mRNA collectively defined here as coding sequences [CDSs], 5 UTRs, 3 UTRs and intergenic regions within transcripts [IGTs]) ( Supplementary Figure S1 D).As with previous Hfq-mediated RNA-RNA interactomes (see Introduction), we also detected numerous interactions between two sRNAs (i.e.sponging interactions), consistent with the idea that Hfq-mediated post-transcriptional regulation extends far beyond its canonical role in mRNA regulation ( 12 ,49 ) (Figure 1 D and Supplementary Figure S2 ).
Circos plot representations of the sRNA-mRNA, sRNA-sRNA, sRNA-tRNA and other (e.g.sRNA interacting with intergenic region between two transcripts or antisense RNA) showed that the Hfq-mediated RNA-RNA interactome is dominated by sRNA-mRNA and sRNA-sRNA interactions across all growth states (Figure 1 D, Supplementary Figure S2 ).As expected, the Hfq-mediated RNA interactomes substantially changed upon transition from N + to N-(225 new at chimeras at N-) and N-24 to N-24 + 2 (187 new chimeras at N-24 + 2) growth states (Figure 1 E).Notably, the Hfqmediated RNA-RNA interactome also substantially differed in bacteria between the two different growth-arrested states, i.e. between N-and N-24 states ( ∼266 new chimeras at N-24; Figure 1 F).Conversely, we noted that several sRNA-mRNA ( ∼39), sRNA-sRNA ( ∼4), sRNA-tRNA ( ∼3) and other ( ∼5), interactions were present across all four growth states, which we termed the 'core' Hfq-mediated RNA interactome in our condition (Figure 1 G; see Discussion).However, the abundance of some of the core RNA-RNA interactions differed at each growth state ( Supplementary Figure S1 E).The number of Hfq-mediated sRNA-mRNA and sRNA-sRNA interactions that are unique to or common between each growth state is summarised in Figure 1 H. Overall, we conclude that the Hfq-mediated RNA interactome is extensive and highly dynamic in E. coli experiencing N starvation.
An elaborate post-transcriptional regulatory basis of the adaptive response to N-starvation in E. coli E. coli cells adapt to N starvation by activating the nitrogen regulation ( 50 ) stress response, resulting in the expression of 20 operons ( 25 ,26 ).The master transcription regulator of the Ntr response is NtrC of the NtrBC two-component system.The activation of the Ntr response results in a large scale reprogramming of gene expression at the transcriptional level because NtrC activated genes include Nac, a global transcription factor affecting the transcription of ∼80-100 genes ( 51 ) and RelA, the enzyme responsible for the synthesis of guanosine tetraphosphate (ppGpp), which acts directly on the RNA polymerase and alters its activity and promoter specificity ( 52 ).In contrast, the extent to which post-transcriptional regulation contributes to the adaptive response to N starvation is not fully understood.Recently, the sRNA GlnZ was shown to promote cell survival by regulating genes linked to N and carbon flux in N starved E. coli in an Hfq-dependent manner ( 53 ,54 ).In MS2-affinity purification and RNA sequencing (MAPS) experiments using GlnZ as bait, conducted in short-term N-starved E. coli (i.e. a condition comparable to N-), GlnZ interacted with numerous mRNA targets ( 54 ).In Supplementary Figure S3 A, we show the GlnZ regulon obtained by RIL-seq in E. coli bacteria in N+, N-, N-24 and N-24 + 2 states.The mRNAs of the 11 genes, which were en-     riched in the MAPS experiments are shown in green.Interestingly, RIL-seq analysis revealed 5 new potential GlnZ targets, which also included an interaction with the nac mRNA at N-.This suggests that the NtrC and Nac regulon are potentially also coupled at the post-transcriptional level (see Discussion).Further, we interrogated the RIL-seq data from all four growth states for sRNAs interacting with mRNAs of genes that belong to the Ntr regulon (i.e.genes directly activated by NtrC).The results in Supplementary Figure S3 B reveal that the expression of a total of 13 of the 20 NtrC dependent operons (shown in blue) are potentially affected by 16 different sRNAs (shown in red) in N+ (3 operons), N-(12 operons) and N-24 states (4 operons) and N-24 + 2 (2 operons) bacteria.Overall, the results highlight that post-transcriptional regulation of gene expression is likely to be an important, yet unexplored, facet of the adaptive response to N starvation in E. coli .

The conserved sRNA SdsR has a substantial post-transcriptional regulatory influence in E. coli experiencing long-term N starvation
To understand post-transcriptional control mechanisms that underpin gene expression in long-term growth arrested bacteria, we focused on the Hfq-mediated RNA-RNA interactome in N-24 E. coli .Figure 2 A shows the top 10 sRNAs that contribute to the Hfq-mediated RNA-RNA interactome across all four growth states (also see Supplementary Data 4 ), with the SdsR sRNAs contributing to almost a third ( ∼31.5%) of chimeras at N-24.Indeed, we observed that SdsR abundance increases 13-fold when bacteria transit from the Nto the N-24 state but decreases 140-fold when the cells recover growth at N-24 + 2 ( Supplementary Figure S5 A).SdsR, which is ∼100 nucleotide long, is one of the most highly conserved enterobacterial sRNAs ( Supplementary Figure S4 A).Its transcription is dependent on the RNAP containing the general stress response promoter specificity factor RpoS and as such SdsR is produced when cells encounter stress and / or growth-arrest ( 28 ,55 ).In previous RIL-seq experiments, the Hfq-mediated interactome of SdsR consisted of ∼92 interactions in E. coli from stationary phase of growth in lysogeny broth (i.e. a growth state comparable to N-) and only one Hfqmediated RNA-RNA interaction involving SdsR was detected in E. coli from the exponential phase of growth (i.e.comparable to the N + bacteria used here) ( 17 ).Consistent with previous results, only 4, 31 and 6 chimeras involving SdsR (representing ∼0.3, ∼4.9 and ∼0.9% of the Hfq-mediated RNA interactome) were detected in N+, N-and N-24 + 2 bacteria, respectively.In N-24 bacteria, SdsR interacted with 124 mRNA targets (Figure 2 B).To establish how many of these interactions were productive (i.e. had a direct regulatory effect on the target mRNA), we compared the total proteomes of wild-type and sdsR bacteria at N-24 (Figure 2 C).Neither the growth dynamics of wild-type and sdsR ( Supplementary Figure S5 B) nor the proportion of viable cells in the N-24 population of wild-type and sdsR ( Supplementary Figure S5 C) were substantially different.We identified ∼50% (2156 / 4328) of total E. coli proteins in the proteomes of wild-type and sdsR bacteria.Notably, the absence of SdsR resulted in the differential expression of ∼70% ( P -value ≤ 0.05) of the identified proteins in sdsR compared to wild-type bacteria.Of the 124 potential mRNA targets of SdsR, the proteins of 90 were detected in the proteomics data, of which 69 were designated as significantly differentially expressed in the sdsR proteome (shown red in Figure 2 C).Analysis of chimeras involving SdsR in the RIL-seq dataset (see Materials and Methods) revealed that of these 90 identified proteins, SdsR bound to 28 of the corresponding mRNAs at or overlapping the RBS (defined here as 40 bp upstream to 15 bp downstream of the AUG codon), 44 mRNAs internally or in the 3 UTR, 5 mR-NAs in the 5 UTR not overlapping the RBS and 9 mRNAs to multiple binding sites within the mRNA.The chimeric alignments and RNA duplex predictions for the five most up and downregulated targets of SdsR are shown in Supplementary Figure S6 .In sum, the data suggest that SdsR can influence the post-transcriptional fate of its mRNA targets in several different ways (see Introduction) in long-term N starved E. coli .
We expected that proteins of mRNAs to which SdsR bound at or overlapping the RBS to be upregulated in the sdsR proteome compared to the wild-type proteome, because sRNA interference at the RBS usually reduces translation (see Introduction).Consistent with this view, ∼79% (22 / 28) of the detected proteins where SdsR was bound to sites at or overlapping the RBS of their corresponding mRNAs were found to be upregulated in the sdsR proteome (Figure 2 D, red dots).We expected that proteins of mRNAs to which SdsR bound internally or at the 3 UTR to be either downregulated in the sdsR proteome (if SdsR action at these sites results in the protection of the targeted mRNA from degradation), or upregulated (if SdsR action leads to enhanced degradation of the targeted mRNA (see Introduction).Consistent with this view, ∼50% (22 / 44) and ∼30% (13 / 44) of the detected proteins where SdsR was bound internally or in the 3 UTR of their corresponding mRNA were found to be downregulated and upregulated, respectively, in the sdsR proteome (Figure 2 D, blue dots).Overall, we conclude that in N-24 E. coli (i) the regulatory influence of SdsR extends far beyond its direct regulatory targets and thereby indirectly affects diverse and broad range of cellular processes ( Supplementary Figure S5 D), ( 35 ) a large proportion of the interactions between SdsR and its mRNA targets have a productive regulatory effect and (iii) SdsR has both a broad positive and negative regulatory effect on gene expression.

SdsR is required for optimal growth recovery from long-term N starvation
Although SdsR affects the expression of genes associated with diverse processes in N-24 bacteria, its absence, surprisingly, has little impact on the growth or long-term survivability of sdsR bacteria ( Supplementary Figure S5 B and C).Therefore, we considered whether SdsR had any influence on the ability of N-24 bacteria to recover from growth-arrest.We discovered that when N-24 sdsR bacteria were inoculated into fresh culture media, they consistently displayed an increased lag phase (here defined as the time from inoculation to the doubling of the OD 600 reading of the culture) by ∼57 min longer than wild-type bacteria (Figure 3 A, left).The increased lag phase of sdsR bacteria could be reverted to that of wildtype bacteria when plasmid-borne SdsR was expressed from its native promoter in sdsR bacteria (Figure 3 A, inset).Although, we noted that harbouring the pBR322 plasmid further increases the lag phase of sdsR bacteria by ∼45 min compared to sdsR bacteria without any plasmid.This increased lag phase was not detected when N-sdsR bacteria were recovered in fresh media (Figure 3 A, right), demonstrating that    Statistical analysis performed by Welch's T-test.( B ) Scanlag analysis of colony appearance time (in min) and growth rate (in pixels 2 / h (px 2 / h)) for wild-type (blue) and sdsR (red) strains experiencing N starvation for 24 h and plated onto Gutnick agar with 3mM NH 4 Cl.Black circles represent a v erage population growth rate and appearance time with mean values indicated.( C ) Graph of the proportions of wild-type and sdsR following co-inoculation of equal amounts of wild-type and sdsR bacteria from the N-and N-24 growth states into fresh growth media.Proportions were determined by CFU measured at an OD 600nm of 0.05, 0.4 and 0.8 in the co-culture, following plating on LB agar with (to select only for sdsR bacteria) and without kanam y cin (to select for total number of bacteria).The experimental approach is shown schematically at the top.Statistical analysis perf ormed b y Bro wn-F orsyth and W elch's ANO V A. ( D ) As in (C), but equal proportions of N-24 wild-type and sdsR bacteria were resuspended in Gutnick media without any NH 4 Cl, incubated for up to 120 h, with proportion of wild-type and sdsR bacteria determined at regular intervals.Statistical analy sis perf ormed b y Bro wn-F orsyth and W elch's ANO V A. (**** P < 0.0 0 0 1; ** P < 0.0 1; * P < 0.05).
the increased lag phase is a specific property of long-term N starved sdsR bacteria.
We considered whether the difference in lag phase between wild-type and sdsR bacteria upon recovery from N-24 state was due to the presence of a subpopulation(s) of bacteria in the sdsR population that resumed growth slowly.To investigate this, we used a method called ScanLag, which allows the measurement of the appearance time of individual colonies on a solid growth medium as a function of time following inoculation ( 40 ).We considered that if the N-24 sdsR population consisted of one or more slow growing subpopulations compared to the wild-type population, then the appearance of individual colonies of sdsR bacteria would occur at different times than that of wild-type bacteria.As shown in Figure 3 B, when N-24 sdsR bacteria and wild-type bacteria were plated on solid media containing ammonium chloride as the sole N source, most of the sdsR colonies appeared at the same time, but their appearance time differed by ∼280 min compared to that of the wild-type colonies.Although, we noted that the appearance time of individual sdsR colonies were inherently more heterogenous than that of wild-type colonies.
As bacteria exist in polymicrobial environments, we considered whether the subtle increase in lag time to growth recovery of N-24 sdsR bacteria would lead to a competitive disadvantage when recovered in a mixed population with wild-type bacteria.To investigate this, we inoculated the same number (determined by counting colony forming units [CFU]) of sdsR and wild-type cells from N-24 into fresh media and enumerated the CFU of sdsR bacteria as a percentage of total bacteria in the culture.Following initial inoculation at OD 600 ∼0.05, when sdsR and wild-type bacteria were present at approximately equal proportions, at OD 600 ∼0.4 and ∼0.8, the proportion of sdsR bacteria decreased to that of ∼20% of the total population (Figure 3 C).Such a difference was not detected when N-sdsR and wild-type bacteria were co-inoculated (Figure 3 C), consistent with the results shown in Figure 3 A (right) when no difference in lag time between sdsR and wild-type bacteria was observed during growth recovery.The competitive disadvantage of the sdsR bacteria was largely specific to growth recovery from the N-24 state, as it was absent when N-24 sdsR and wild-type were co-inoculated into fresh but N depleted media, which did not support growth (Figure 3 D).However, we note that following 120 h of incubation, the population of sdsR bacteria started to drop (by ∼15% of total population).This suggests that the absence of SdsR can confer a subtle competitive disadvantage or adversely affect viability following prolonged ( > 120 h; Supplementary Figure S5 C) N starvation.In sum, we conclude that the post-transcriptional regulon of SdsR primarily contributes to optimal growth recovery of long-term N-starved bacteria.

Optimal growth recovery from long-term N starvation requires translational downregulation of the conserved yeaGH operon by SdsR
We interrogated the proteome of sdsR bacteria to identify genes post-transcriptionally regulated by SdsR that might be responsible for the lag in the growth recovery of N-24 sdsR bacteria.We focused on proteins whose translation was directly affected by SdsR and transcription of their mRNAs was activated by NtrC.We identified HisQ, YeaG and YeaH as such proteins, which were ∼1.3, ∼1.8 and ∼2.1-fold, respectively, more abundant in N-24 sdsR bacteria than in wildtype bacteria.As we previously implicated YeaG and YeaH in the adaptive response to N starvation ( 47 ,48 ), we focused our analysis on these two proteins.
In previous work, we reported that the yeaGH operon is activated by NtrC when E. coli experiences N starvation (i.e. at N-) and that YeaG and YeaH functionally cooperate and, by a yet to be characterised mechanism, contribute to quiesce metabolism in N-starved E. coli ( 47 ,48 ).The products of the yeaGH operon are conserved in several enterobacteria ( Supplementary Figure S4 B and C) and the yeaGH operon is expressed (in addition to N starvation) when bacteria encounter diverse stresses such as low pH ( 56 ), high osmolarity ( 57 ), stationary phase ( 56 ), low sulphur ( 58 ).YeaG is a Hank's type eukaryote-like serine / threonine kinase ( 59 ,60 ) but the biological function of YeaH remains unknown.The absence of y eaG or y eaH compromises the long-term survivability of mutant bacteria under N starvation ( 47 ,48 ).Notably, the growth recovery phenotype of y eaG , y eaH and y eaGH bacteria contrasts that of the sdsR bacteria: The y eaG , y eaH and yeaGH bacteria recover from the N-24 state with a shorter lag phase (by ∼30 min) compared to wild-type bacteria (see later and ( 47)).
As YeaG is experimentally more tangible than YeaH, we were previously able to purify and raise antibodies against YeaG to demonstrate that YeaG protein levels peak during first 6-9 h from the onset of N starvation but decrease by N-24 ( 48 ), suggesting a potential post-transcriptional regulatory mechanism underpinning yeaGH gene expression.Our new data have now revealed that yeaGH expression is subjected to translational downregulation by SdsR, underscoring the view that managing the intracellular levels of YeaG (and YeaH) is important for optimal growth recovery from longterm N starvation.Consistent with this view, immunoblotting of YeaG in N-24 wild-type and sdsR bacteria revealed a substantially increased abundance of YeaG in sdsR bac-teria compared to in wild-type bacteria (Figure 4 A).However, the transcription of yeaGH operon can be driven by two different promoters that depend on the RNAP containing the promoter specificity factor RpoS or RpoN.Inspection of the proteome of the sdsR bacteria at N-24 reveals that RpoS and RpoN proteins are ∼34% and ∼13%, respectively, more abundant in sdsR bacteria than in wild-type bacteria ( Supplementary Figure S7 A).Therefore, it is possible that the increased abundance of YeaG and YeaH in sdsR bacteria is potentially due to increased abundance of yeaGH mRNA.However, the transcriptome sdsR bacteria revealed that the abundance of yeaGH mRNA did not significantly differ from that in wild-type bacteria (differentially expressed genes were defined as those with a false discovery rate adjusted P < 0.05), suggesting that the increased abundance of Y eaG and Y eaH in N-24 sdsR bacteria is due a post-transcriptional influence of SdsR on yeaGH mRNA ( Supplementary Figure S7 B).
Therefore, to independently validate the SdsR-mediated translational downregulation of yeaGH , we used a GFP reporter system in which SdsR and its target sequence were co-expressed from different compatible plasmids ( 27 ).As the predicted base-pairing interactions between SdsR and 5 UTR of yeaG mRNA occur between positions −31 and −22 relative to the translation start site (AUG; A is +1; Figure 4 B), we fused the nucleotide sequence comprising positions -93 to +63 to the amino terminus of GFP and placed the fusion construct under a constitutive promoter in pXG10 ( 27 ) to create p5UTR-yeaG .The SdsR sequence was placed under an IPTGinducible promoter in pKF68 ( 28 ) to create plasmid pSdsR.Plasmid pJV300, which expresses a ∼50 nucleotide nonsense RNA derived from rrnB terminator region ( 27 ), served as the control vector for pSdsR (hereafter referred to as pCONT).E. coli containing p5UTR-yeaG and either pCONT or pS-dsR were grown in lysogeny broth to stationary phase in the presence of IPTG.As shown in Figure 4 C, SdsR expression resulted in a decrease in GFP signal by ∼50% compared to bacteria containing pCONT.We created p5UTR-yeaG MUT in which the SdsR interaction sequence in the 5 UTR was altered to compromise efficient SdsR binding (Figure 4 B).The overall GFP signal from p5UTRyea G MUT was lower than that from p5UTR-yeaG , suggesting that the changes we introduced are adversely affecting the adjacent RBS sequence.Nonetheless, as expected, we did not detect the decrease in GFP signal upon expression of SdsR (Figure 4 C).However, when we introduced a compensatory mutation into SdsR (pSdsR COMP ) we detected a significant decrease in GFP levels compared to cells containing pSdsR ( ∼19%, Figure 4 C) and pCONT ( ∼29%, Figure 4 C).In sum, the reporter system independently confirms that RIL-seq and proteomics data and supports our prediction that yeaGH mRNA translation is regulated by SdsR.
Next, we considered whether the increased lag time to growth recovery from N-24 displayed by the sdsR mutant (Figure 3 A) would be absent if YeaG could not be overexpressed.Hence, we created a sdsR yeaG double mutant E. coli strain and compared its lag time to growth recovery form N-24 to that of the yeaG and sdsR single mutant and wild-type bacteria.As previously reported by us ( 48 ), the yeaG bacteria displayed a decreased lag time (by ∼47 min) to growth recovery from N-24 compared to wild-type bacteria (Figure 5 A, compare green and blue lines, respectively).However, the increased lag time to growth recovery from N-24 displayed by the sdsR mutant (relative to wild-type bacteria) was barely present in the sdsR yeaG double mutant relative to yeaG bacteria (Figure 5 A, compare brown and green lines, respectively).
Finally, we considered that overexpression of YeaG from an l -arabinose inducible plasmid (pBAD-yeaG ), in which the native 5 UTR of yeaG is absent and SdsR thus cannot translationally downregulate its expression, would mimic a scenario analogous to the one in sdsR bacteria resulting in an increased lag time to growth-recovery from N-24.We grew wild-type bacteria containing pBAD-yeaG and pBADempty to N-and induced YeaG expression with l -arabinose.As expected, the lag time to growth recovery from N-24 wild-type bacteria containing pBAD-yeaG was increased by ∼28 min compared to that of bacteria containing pBAD-empty (Figure 5 B left, blue and dark blue lines, respectively) and resembled that of sdsR bacteria containing pBAD-empty (Figure 5 B left, red line).Clearly, the overexpression of yeaG does not completely mimic the absence of SdsR, likely suggesting that SdsR has additional role in growth recovery beyond its regulatory effect on yeaG .The lag time to growth recovery, when YeaG was overexpressed in sdsR bacteria, mirrored that of sdsR bacteria containing pBAD-empty (Figure 5 B, right, orange line).It thus seems that overexpressing YeaG does not further increase the lag time to growth recovery of the sdsR bacteria.To confirm that the increased lag-time to growthrecovery of wild-type bacteria containing pBAD-yeaG was due to the biological activity of YeaG, and not to YeaG overexpression per se , we overexpressed a catalytically deleterious YeaG variant containing a K426A mutation in the kinase domain of YeaG ( 47 ).As expected, the lag-time to growthrecovery was absent in wild-type and sdsR bacteria containing pBAD-yeaG (K426A) (Figure 5 B right, compare dark blue and red lines with orange dashed and cyan dashed lines, respectively).In sum, we have uncovered that the translational downregulation of the conserved operon yeaGH by SdsR is required for optimal growth recovery of long-term N starved E. coli .

Discussion
The past few years have seen a surge in new methodologies to identify post-transcriptional regulatory networks in bacteria at a global scale, which have underscored the depth and breadth of their contribution to gene expression ( 61 ).However, our current knowledge of post-transcriptional regulatory networks is restricted to growing bacteria, bacteria undergoing growth transitions (e.g. from the exponential to station- ary phase of growth) or have been exposed to iron limitation ( 4 , 17 , 20 ).The impetus for this study was to understand to what extent the Hfq-mediated post-transcriptional regulatory network, i.e. the RNA-RNA interactome, changes during growth arrest and underpins recovery from growth arrest.In addition, as we used N starvation to induce growth arrest, this allowed us to simultaneously gain deeper insights into the post-transcriptional basis of the adaptive response to N starvation in E. coli .
Our results have revealed that the Hfq-mediated RNA-RNA interactome undergoes large scale reprogramming, not only between growth and growth arrested states (i.e.N + to Nand N-24 to N-24 + 2 states), but also throughout the course of growth arrest (i.e.N-to N-24 state) (Figure 1 ).One way this reprogramming manifests itself is via the different sRNA that govern the Hfq-mediated RNA-RNA interactome at different growth states.For instance, we detected SdsR in ∼30% of chimeras in the Hfq-mediated RNA-RNA interactome of N-24 bacteria.However, it is only present in < 5% of chimeras in the Hfq-mediated RNA-RNA interactome of bacteria from N-, and in < 1% of chimeras in bacteria from N+ and N-24 + 2 (Figure 2 A & Supplementary Figure S8 ).This is consistent with previous RIL-seq results showing increased prominence of SdsR in the Hfq-mediated RNA-RNA interactome of bacteria from the stationary phase of growth ( 17 ).Conversely, ArcZ is an sRNA that governs the Hfq-mediated RNA-RNA interactome at all growth states studied here (Figure 2 A), but its target suite differs substantially between growth states ( Supplementary Figure S8 ), with only 11% of its targets being found in its chimeras under all growth states.Notably, the largest proportion ( ∼37%) of all chimeras involving ArcZ in Hfq-mediated RNA-RNA interactome in N-bacteria is with the sRNA RybB, suggesting that regulation of sRNA activity by sponging interactions can be growth state specific phe-nomenon.In sum, although the Hfq-mediated RNA-RNA interactome is clearly plastic, it is worth exploring how much of this plasticity is driven by changes in the intracellular concentrations of specific RNAs or if the changes are due to a more active mechanism of selection by Hfq.
We identified a number of RNA-RNA interactions that were detected across all growth states -we collectively referred to this as the 'core' Hfq-mediated RNA-RNA interactome (Figure 1 G).The number of chimeras that constitute the core Hfq-mediated RNA-RNA interactome unsurprisingly fluctuate between the four growth conditions used here ( Supplementary Figure S1 E), which underscores the inherent plasticity of post-transcriptional regulatory networks in bacteria.Notably, 49% of the core Hfq-mediated RNA-RNA interactions identified under our conditions were also detected in E. coli during exponential growth and in the early stationary phase in lysogeny broth ( 17 ) ( Supplementary Figure S9 ) suggesting that a portion of the Hfq-mediated posttranscriptional regulatory network may act in a 'housekeeping' capacity, rather than in a condition specific regulatory capacity.However, we suggest that concept of the core Hfqmediated RNA-RNA interactome must be viewed cautiously as the extent of this interactome can be highly influenced by the analysis parameters of any RIL-seq study (e.g. in this study we have only considered interactions identified in more than 30 chimeras, whereas other studies ( 17 ) have used a different threshold and would produce a different core Hfq-mediated RNA-RNA interactome).Hence, we recommend that if the concept of core Hfq-mediated RNA-RNA interactome is to be accepted by the field, we suggest a consensus threshold is defined.
Conversely, the RNA-RNA interactions identified in growing bacteria in this study (i.e. at N+ and N-24 + 2 states) also differs from the Hfq-mediated RNA-RNA interactome identified by Melamed and colleagues in exponentially growing E. coli in lysogeny broth ( 17 ).For example, of the 315 and 279 unique RNA-RNA interactions identified in the Hfqmediated RNA interactome of N + and N-24 + 2 bacteria, respectively, only 110 and 84 of them are also detected in the Hfq-mediated RNA-RNA interactome of exponentially growing E. coli in lysogeny broth.This implies that the Hfq-mediated RNA-RNA interactome is highly sensitive to changes in growth conditions even if they permit active growth.Consistent with this view, the Hfq-mediated RNA-RNA interactomes of N+ and N-24 + 2 bacteria are similar but not identical (Figure 1 C and Figure 1 D).Overall, the sensitivity and specificity to growth conditions may underpin the inherent physiological purpose of post-transcriptional regulatory networks, where they might be considered to fine tune (see below), rather than to act in a binary fashion, to regulate the flow of genetic information to meet cellular needs.
Our study also suggests how a single sRNA can couple two regulons to potentially fine tune an adaptive response.This is exemplified by GlnZ whose expression is activated by NtrC and is thus a member of the NtrC regulon.The regulatory targets of GlnZ have been explored in a number of previous studies ( 53 ,54 ), and many of those identified previously have also been identified in the currently study ( Supplementary Figure S3 A).However, our data additionally uncovered that GlnZ interacts with the mRNA of nac in the N-growth state, which encodes a global transcription factor and is a member of the NtrC regulon.The interaction site of GlnZ is in the CDS of nac mRNA ( Supplementary Figure S10 ) and at this stage we are uncertain whether GlnZ has a positive or negative regulatory effect on Nac expression.The interaction between GlnZ and nac mRNA is only detected in the N-state and it is absent in the N-24 state ( Supplementary Figure S3 A).Thus, we propose that GlnZ functions to adjust Nac proteins levels to fine tune the adaptive response to N starvation as N starvation condition ensues.For example, GlnZ could have a positive regulatory effect on nac mRNA and thereby increase Nac proteins levels in N-bacteria to repress the transcription of sucA .As sucA encodes a subunit of 2-oxoglutarate dehydrogenase enzyme, the cooperative action of GlnZ and Nac could allow efficient balancing of carbon and nitrogen metabolism at the onset of N starvation ( 53 ,54 ).Further, our analysis of RNA-RNA chimeras involving NtrC activated genes from different stages of N starvation in E. coli underscores that posttranscriptional regulation is clearly an important, yet relatively underexplored, facet of the adaptive response to N starvation in E. coli ( Supplementary Figure S3 B).
Although many of the new global approaches provide a comprehensive overview of post-transcriptional regulatory interactions at the transcriptome wide level, the major challenge for many investigators is to ascertain their impact on gene expression and ultimately cellular physiology.This is further compounded by the fact that deletion of an sRNA of interest often does not result in discernible phenotypic changes.We have demonstrated that coupling RIL-seq with global comparative proteomics of wild-type and a deletion mutant of an sRNA of interest can not only validate whether the interactions made by the said sRNA are productive, i.e. lead to changes in gene expression, but also differentiate between the direct and indirect effects of how the sRNA contributes to gene expression.We have shown that the absence of SdsR, the sRNA that governs the Hfq-mediated RNA-RNA interactome in N-24 bacteria, results in a large scale perturbation of the proteome of E. coli , extending far beyond its direct regulatory targets (Figure 2 C).However, this interpretation must be viewed cautiously as the observed large scale changes to the proteome could be the accumulative consequence of dysregulated post-transcriptional events that occurred in the N+ and N − growth states.Nonetheless, at N-24, we observed that the expected regulatory outcomes of SdsR targeted mRNAs correlated surprisingly well with the levels of proteins encodes by the targeted mRNAs (Figure 2 D).The target spectrum of SdsR in N-24 bacteria is functionally diverse, containing genes associated with almost every major cellular function ( Supplementary Figure S5 ).We noted an enrichment of targets involved in energy metabolism (e.g.gltA and aceA ) carbon metabolism (e.g.tktA and suhB ) and nitrogen metabolism (e.g.aroF , thrA, serC, solA, asd ) amongst those targets as potentially positively regulated by SdsR ( Supplementary Figure S5 and S11 ).This implies a role for SdsR in regulating cellular metabolism in N-24 bacteria.This view is consistent with SdsR regulating a number of targets associated with energy metabolism in stationary phase Salmonella bacteria ( 28 ).
We also observed a number of proteins (TolC, TamA, MepK, FumC) associated with antibiotic susceptibility are upregulated in the proteome of sdsR bacteria ( Supplementary Figure S11 ), suggesting that SdsR acts to post-transcriptionally downregulate the expression of them.Consistent with this view, Parker and Gottesman ( 62 ) previously reported that SdsR downregulates expression of the outer membrane channel TolC, and that this has a subtle effect on susceptibility to novobiocin and erythromycin.Further, mRNAs of genes encoding TamA, a component of the TAM translocation and assembly module that has been linked to antimicrobial stress in K. pneumoniae ( 63 ), MepK, a regulator of alternate peptidoglycan biosynthesis that has been linked to beta-lactam resistance ( 64 ,65 ) and FumC, the alternative fumarase that has been linked to susceptibility of E. coli to certain bactericidal antibiotics ( 66 ) also belong to the regulon of SdsR and their expression is potentially negatively affected by SdsR.In contrast, the d -ala −d -ala ligase DdlB, which is directly targeted by the antibiotic cycloserine ( 67 ), was downregulated in the proteome of sdsR bacteria ( Supplementary Figure S11 ), suggesting that SdsR acts to post-transcriptionally upregulate DdlB expression.Overall, our results support the widening body of evidence that sRNAs, and the RNA-RNA interactomes defined by them, may contribute to modulating the antibiotic susceptibility of bacteria (reviewed in ( 68 )).
Despite the broad involvement of SdsR in diverse cellular processes in growth arrested bacteria ( Supplementary Figure S11 & ( 28 )), under N starvation, the absence of SdsR, surprisingly, neither effects growth nor viability.However, we discovered a role for SdsR in affecting growth recovery from longterm N starvation and uncovered that this process is underpinned by two new members, yeaG and yeaH , of the SdsR regulon in N-24 bacteria.In previous work, we identified the highly conserved yeaGH operon as part of the NtrC regulon and showed that Y eaG and Y eaH function together to quiesce metabolism upon sensing N starvation ( 69 ).Therefore, the absence of YeaG and YeaH are detrimental to cell viability as a function of time under N starvation, but conversely yeaG, y eaH and y eaGH bacteria recover growth faster from N starvation following inoculation into fresh media.Put simply, Y eaG and Y eaH can be considered to function as a metabolic brake.YeaG protein levels peak during the first 9 h into N starvation but subsequently drops as N starvation ensuesperhaps suggesting that YeaG is only required in the early stages of N starvation to quiesce metabolism.Although the mechanism by which YeaG and YeaH induce metabolic quiescence will be a topic of a future study, the discovery that yeaG mRNA is subjected to post-transcriptional regulation by SdsR in long-term N starved bacteria (Figures 2 D and 4 C) underscores the importance of controlling YeaG (and YeaH) levels as N starvation ensues.The management of YeaG (and YeaH) levels seems important to balance quiescing metabolism under growth non-permissive conditions and to restart metabolism to allow optimal growth recovery when conditions become growth permissive.Consistent with this view, failure to do so adversely effects the ability of bacteria to optimally recover from long-term N starvation induced growth arrest.
In sum, our study has underscored the importance of studying RNA interactomes of growth arrested bacteria as they can be dynamic and extensive as those in actively growing bacteria and, importantly, can contribute to uncovering new biological processes that govern bacterial growth.As the yeaGH operon and SdsR are conserved in many bacteria, including ESKAPE pathogens (Figure 4 A & Supplementary Figure S4 ), our study has uncovered a conserved post-transcriptional regulatory axis that governs optimal growth recovery from longterm N starvation.We speculate that the yeaGH -SdsR regulatory axis might contribute to optimal growth recovery from diverse conditions that induce growth arrest and thus identify-ing ways to interfere with this regulatory axis might provide options to manage bacterial growth and colonisation.Similarly, the core Hfq-mediated RNA-RNA interactome offers several opportunities for antibacterial intervention.In this regard, our study provides a treasure chest of RNA-RNA interactions in a prototypical bacterium at four different states of growth that can be exploited to gain valuable new insights into the post-transcriptional regulatory basis of bacterial growth and growth arrest.

Figure 1 .
Figure 1.RIL-seq re v eals an e xtensiv e and dynamic Hfq-mediated RNA interactome in N starved E. coli .( A ) Schematic representation of experimental design, indicating growth states used throughout this study.Shown in orange are the ammonium chloride levels and in black the growth curve.( B ) R elativ e frequencies of each RNA type found in chimeric fragments, in individual replicates across all time points for hfq-FLAG datasets.( C ) Principal component analysis of complete chimeric fragment datasets from each replicate.The value in the brackets indicates the percentage contribution of each component to the total variance.( D ) Circos plots of RIL-seq interactions that are represented by at least 30 chimeric fragments in two individual replicates, mapped to the E. coli K12 MG1655 genome.The thickness of each connection is proportional to the a v erage number of chimeras detected for a given interaction across the three replicates.sRNA:mRNA, sRNA:sRNA, sRNA:tRNA and other interactions are represented by blue, orange, green and red lines respectively.( E ) Circos plots of the RIL-seq interactions detected at N-but not N+ (left), and N-24 + 2 but not N-24 ( 23 ).( F ) Circos plot of the RIL-seq interactions detected at N-24 but not N-.( G ) Circos plot of RIL-seq interactions detected at all time points.Thickness of connections are not w eighted b y the number of chimeras.T he RNA in v olv ed in each connection are sho wn. ( H ) Venn diagrams sho wing the condition specificity of sRNA:mRNA and sRNA:sRNA interactions.Diagrams on the left show the number of interactions, diagrams on the right are weighted by the number of detected chimeras for each interaction, shown as a percentage of the entire dataset.

Figure 2 .
Figure 2. The conserved SdsR has a substantial post-transcriptional regulatory influence in E. coli experiencing long-term N starvation.( A ) Graph showing the top ten most represented sRNAs in the RIL-seq data across all time points and with their relative abundance at each time point shown.( B ) Circos plot of interactions in v olving SdsR at N-24 that are represented by at least 30 chimeric fragments in two individual replicates.The thickness of each connection is proportional to the a v erage number of chimeras detected for a given interaction across the three replicates.sRNA:mRNA, sRNA:sRNA, sRNA:tRNA and other interactions are represented by blue, orange, green and red lines respectively.( C ) Volcano plot of differential protein le v els in N-24 sdsR bacteria shown as a log 2 change from wild-type bacteria.Proteins differentially expressed more than 1 log 2 (i.e. a greater than 2-fold change) are indicated, and those proteins whose mRNA interacted with SdsR in the RIL-seq dataset are coloured orange.( D ) As above, showing only proteins whose mRNA interacted with SdsR, coloured by the SdsR binding site as determined though RIL-seq.Interaction regions overlapping the RBS (defined here as 40bp upstream to 15bp downstream of the AUG) are shown in red, interactions internal to the coding sequence or in the 3 UTR are shown in blue, interactions in the 5 UTR not overlapping the RBS are shown in green, targets with multiple binding sites within the mRNA are shown with split colour and poorly defined binding sites and binding sites within intergenic regions in the same transcript are shown in grey.Schematic representation of the categories of sRNA binding site, and the expected regulatory outcome on protein levels in the wild-type and sdsR bacteria are schematically shown.

Figure 3 .
Figure 3. SdsR is required for optimal growth recovery from long-term N starvation.( A ) Graphs showing recovery growth of wild-type and sdsR bacteria f ollo wing sub-culturing of 24 h (L eft) and 20 min (Right) N-starv ed bacteria into fresh culture media.Inset sho ws reco v ery of wild-type and sdsR bacteria containing either empty pBR322 or pBR322 expressing sdsR from its native promoter, following sub-culturing of N-24 bacteria into fresh culture media.Error bars represent standard deviation ( n = 6).Additional graph shows lag-time and tables show lag-time (LT) and doubling-time (DT).Statistical analysis performed by Welch's T-test.( B ) Scanlag analysis of colony appearance time (in min) and growth rate (in pixels 2 / h (px 2 / h)) for wild-type (blue) and sdsR (red) strains experiencing N starvation for 24 h and plated onto Gutnick agar with 3mM NH 4 Cl.Black circles represent a v erage population growth rate and appearance time with mean values indicated.( C ) Graph of the proportions of wild-type and sdsR following co-inoculation of equal amounts of wild-type and sdsR bacteria from the N-and N-24 growth states into fresh growth media.Proportions were determined by CFU measured at an OD 600nm of 0.05, 0.4 and 0.8 in the co-culture, following plating on LB agar with (to select only for sdsR bacteria) and without kanam y cin (to select for total number of bacteria).The experimental approach is shown schematically at the top.Statistical analysis perf ormed b y Bro wn-F orsyth and W elch's ANO V A. ( D ) As in (C), but equal proportions of N-24 wild-type and sdsR bacteria were resuspended in Gutnick media without any NH 4 Cl, incubated for up to 120 h, with proportion of wild-type and sdsR bacteria determined at regular intervals.Statistical analy sis perf ormed b y Bro wn-F orsyth and W elch's ANO V A. (**** P < 0.0 0 0 1; ** P < 0.0 1; * P < 0.05).

Figure 5 .
Figure 5.Optimal growth recovery from long-term N starvation requires translational downregulation of the conserved yeaGH operon by SdsR.( A ) R eco v ery gro wth of wild-type, sdsR , y eaG and sdsR y eaG N-24 bacteria f ollo wing sub-cult uring into fresh cult ure media.Error bars represent standard deviation ( n = 3).Additional graph shows lag-time and table shows lag-time (LT) and doubling-time (DT).Statistical analysis performed by Brown-Forsyth and Welch's ANO V A. ( B ) R eco v ery gro wth of N-24 wild-type and sdsR bacteria o v ere xpressing either wild-type (pBAD18-y eaG ) or a catalytically defective mutant (pBAD18-yeaG K426A) YeaG following sub-culturing of 24 h N-starved bacteria into fresh culture media.Data is split onto two axis for clarity but was performed at the same time.Expression of yeaG was induced with 0.125% (v / v) L -arabinose at N-. Error bars represent standard deviation ( n = 6).Additional graph shows lag-time and table shows lag-time (LT) and doubling-time (DT).Statistical analysis performed by Brown-Forsyth and Welch's ANO V A (**** P < 0.0 0 01; ** P < 0.01).