Our previous analysis of the expression of the P. fluorescens SBW25 rim operon throughout rhizosphere colonisation [19] suggested that it was subject to control by uncharacterised signals from the rhizosphere environment and prompted us to investigate the rimABK operon in more detail. First, RT-PCR was used to show that the three rim genes are transcribed as a single polycistronic operon (S1A Fig), whose transcriptional start site was mapped to 28–30 bp upstream of the rimA ORF using 5’ RACE (S1B Fig). To examine the relationship between rim expression and environmental signals in more detail, we used qRT-PCR to quantify rimK mRNA abundance in SBW25 exposed to a variety of different nutrient conditions and abiotic stresses. Significant increases in rimK expression were observed for cells transferred to low temperature (8°C) and exposed to nutrient-limiting conditions (Fig 1), consistent with earlier findings that rim expression is reduced in the established (nutrient-replete) wheat rhizosphere [19]. Relative levels of rimA mRNA quantified under the same growing conditions do not substantially differ from those seen for rimK (S1C Fig), strongly suggesting that the identified promoter controls the expression of all three rim genes.

Our initial study suggested that a major element of RimK regulation is mediated post-transcriptionally, by interaction with RimA, RimB and the second messenger molecule cdG [19]. At this stage, how the RimABK-cdG system functions to control ribosomal protein modification remains unknown. To address this, we developed kinetic models of the system and conducted a series of biochemical experiments, measuring the effects of each regulator individually and in combination on RimK activity.

The effects of both RimA and cdG on RimK ATPase activity were dependent on a roughly equimolar ratio of protein/messenger with RimK (Fig 2A–2E). Combined with earlier observations [19], this strongly suggests that RimK activity is controlled by direct interaction in each case. A simple explanation for the change in RimK activity is that RimK can exist in two conformational states, one with basal activity and one with higher activity. To evaluate this hypothesis, we built models of protein binding based on mass action kinetics for the individual reactions (S1 Table). The two-state model captures the behaviour of each of the individual experiments, however, the predicted differences in the equilibrium binding constants are not consistent with existing data (S2 Fig). Using K_d = 1 μM for cdG binding to RimK [19] leads to 96% of RimK being in a cdG bound state under the experimental conditions and predicts a marginal ATPase activity increase upon addition of RimA for a two state model. Experimental validation qualitatively confirms this increase but quantitatively falsifies the two-state model, showing that the effects of RimA and cdG are approximately additive (Fig 2E). This suggests that RimK can exist in at least four different states with varying ATPase activities (S1 Table). This four-state model of RimK ATPase activity provides a good quantitative fit with the experimental data as well as reasonable K_d values (S2 Fig).

To test whether RimK modification of RpsF represents a rapid, active cellular response to environmental signals, we examined the short-term impact of RimK stimulation on the bacterial proteome. Our qRT-PCR data (Fig 1) indicates that rim mRNA abundance is promoted in cold, starved cells. Therefore, overnight cultures of WT SBW25 and ΔrimK cells grown at 28°C in LB media were abruptly transferred to 8°C in nutrient-limiting rooting solution (A), and 28°C in LB media (B) for 45 minutes. We then conducted a quantitative proteomic analysis experiment using isobaric labelling (iTRAQ) to examine how these conditions affected protein abundance in the WT and mutant backgrounds. When a 2D scatter plot was constructed for this data (Fig 7), proteins whose conditional abundance changes are unrelated to rimK deletion sit along a line with gradient +1 (x = y). Deviation from the x = y line indicates an effect of rimK deletion on protein abundance under the activating conditions (Fig 7, S2 Table), with a substantial fraction of the SBW25 proteome shifting towards higher ratios in ΔrimK (Fig 7, and boxplot in S5 Fig). Proteins with ratios differing significantly (p≤ 0.05) by a factor ≥2 between WT and ΔrimK (located above or below the dashed lines) are coloured in red (up) and blue (down). Substantial agreement with the previously determined RimK regulon [19] could be observed (indicated by the green dots in Fig 8 representing significant differences) confirming that RimK exerts rapid control of the SBW25 proteome under nutrient-limiting conditions and low temperature.

The timeframe for the proteomic changes observed in our iTRAQ experiment was much too rapid to be a result of the production of new ribosomes [23], suggesting that the proteomic changes we observe could represent an active response to modification of ribosomally-associated RpsF proteins.

Our data suggests that a large proportion of the observed rimK regulon is controlled through RimK-induced shifts in the abundance of other translational regulators, particularly Hfq [19]. To understand the contribution made by RimK glutamation to ribosome function, we must first understand the relationship between Hfq abundance and RimK. To assess the contribution of Hfq to the RimK phenotype, we produced rimK mutant strains in which the chromosomal hfq gene was flag-tagged at the C-terminus. We then quantified Hfq abundance using a Wes Simple Western system (Bio-Techne) in different media conditions. As expected based on previous results [19], Hfq abundance decreased markedly in the ΔrimK background for cells grown in M9-pyruvate media. This reduction in Hfq levels was observed for cells grown to early exponential or to stationary phase. Interestingly, the relationship between RimK and Hfq was reversed for cells grown in M9-pyruvate supplemented with Cas-aminoacids. Under these conditions, rimK deletion led to a noticeable increase in Hfq abundance in early exponential phase (Fig 8). The results obtained for stationary phase growth in M9-pyr-Cas followed the same trend, with one ΔrimK sample showing an increase in Hfq and the second showing little change. This may reflect the depletion of one or more amino-acid sources in the second sample.

Based on these data, we expect that much of the effect of rim gene deletion on the SBW25 translatome would be reversed for cells grown with Cas-aminoacids as opposed to with pyruvate alone. To test this, we conducted a Ribo-seq analysis on SBW25 WT and ΔrimBK cells grown to early exponential phase in M9-pyr-Cas media. The translational activity of 806 genes was significantly (log₂>1) affected by rimBK deletion (S3 Table), with more affected genes showing reduced (564) than increased (242) mRNA abundance in the ΔrimBK mutant background. The upregulated rimBK regulon was dominated by ribosomal proteins, while the downregulated genes contained a large proportion of ABC amino-acid transporter subunits. We observed a large degree of overlap between this experiment and our published data for the rimK proteome [19], with many upregulated genes in the initial (M9-pyr, stationary phase) experiment downregulated in our Ribo-seq dataset and vice versa (S3 Table).

Our data indicate that RimK produces an indirect, nutrient-dependent effect on the SBW25 translatome by changing the intracellular abundance of Hfq. To test whether RpsF glutamation also exerts a direct effect on SBW25 gene translation, we produced an additional ΔrimBK mutant strain in which the rpsF gene was extended by 4 glutamate residues. The ΔrimBK rpsF_4glu mutant proved extremely difficult to produce, with the eventual mutant strain displaying a complete loss of UV fluorescent siderophore secretion compared to WT SBW25 (S3 Fig). Subsequent whole genome sequencing of this mutant revealed additional SNPs in the pvdI and pvdJ pyoverdin biosynthetic loci. To investigate the potential significance of these pvdIJ mutation we grew SBW25 WT, ΔrimBK and ΔrimBK rpsF_4glu strains on KB Congo Red plates containing excess FeCl₂ to repress pyoverdin production. Under these conditions the ΔrimBK rpsF_4glu mutant displayed enhanced Congo Red binding compared to WT and ΔrimBK, suggesting that the rpsF_4glu mutation exerts a rimBK/pvdIJ-independent effect on SBW25 physiology (S3 Fig).

Both ΔrimBK rpsF_4glu and ΔrimBK rpsF_10glu mutants showed a pronounced swarming motility defect compared with WT SBW25 and ΔrimBK, while the ΔrimBK rpsF_4glu strain also displayed a significant increase in Congo Red binding. Growth of neither strain was affected in rich (KB) or poor (M9-pyr) media (S4 Fig). To investigate the molecular basis of the rpsF_4glu/10glu phenotypes, we conducted Ribo-seq assays for ΔrimBK rpsF_4glu/10glu and compared these data to the translatome of ΔrimBK. Translational activity of 159 genes was significantly (>1.0 log₂) affected by the rpsF_4glu mutation (Fig 9, S4 Table), with roughly equal numbers of up- (79) and downregulated (80) genes. Consistent with the observed preference of the RimBK system to produce RpsF_4glu ribosomal variants (Fig 4), the rpsF_10glu mutation affected considerably fewer gene targets, with ribosomal activity differing significantly from the ΔrimBK background for only 45 loci (S7 Fig, S4 Table). By comparing the significantly affected genes for ΔrimBK/WT against the results for ΔrimBK rpsF_4glu/ΔrimBK, we were able to identify three distinct subgroups of loci within the rpsF_10glu/4glu translatomes. The first of these were genes whose translation is altered by rimBK deletion, but not by RpsF glutamation (Class 1, Fig 9B). These genes comprised the majority of the rimBK regulon, and overlapped substantially with our previously published Hfq translatome dataset [7]. Translation of the second set of genes (Class 2, Fig 9C) was shifted by RpsF glutamation and rimBK deletion in the same direction (usually downregulated). This set contained numerous small, uncharacterised proteins alongside several stress response proteins and amino-acid ABC transporters/ metabolic genes (S5 Table). We observed a large degree of overlap with the equivalent dataset for RpsfF_10glu, suggesting that both mutations produce similar effects on this gene class.

The third set of genes (Class 3, Fig 9D); those inversely affected by RpsF glutamation and rimBK deletion, are candidates for the direct RpsF-glutamation regulon. For RpsF_4glu, these genes predominantly split into four main classes (S5 Table): ABC transporters for fatty/amino-acids, metabolic pathways for metabolism of amino acids, and genes associated with the initial stages of attachment to surfaces (e.g. pili and curli biosynthesis loci) were upregulated by glutamation. As predicted based on the pvdIJ mutation, downregulated loci also included the ferripyoverdine receptor fpvA [24]. RimBK regulation of these key groups (fatty/amino acid transport and metabolism, initial surface attachment and secretion of macromolecules) was also seen when we extended our analysis to include genes that were less than 2-fold different in ΔrimBK versus WT, but more than 2-fold inversely affected in rpsF_4glu (S5 Table, Fig 9E). This group contained loci for pectate lyase and phytase secretion (both upregulated by glutamation), alongside loci linked to regulation of motility and surface attachment [25]. We saw little overlap between the RpsfF_4glu and RpsfF_10glu regulons for this group of genes, with the smaller, less well-defined 10E regulon containing a few ABC transporter components alongside several hypothetical proteins (S5 Table).

Plasmids and strains are listed in S6 Table. Primers are listed in S7 Table. Unless otherwise stated all P. fluorescens strains were grown at 28˚C and E. coli at 37˚C in lysogenic broth (LB) [37] solidified with 1.3% agar where appropriate. Gentamycin was used at 25 μg/ml, carbenicillin at 100 μg/ml, and tetracycline (Tet) at 12.5 μg/ml. For inducible plasmids, IPTG was added to a final concentration 0.2 mM (SBW25) or 1 mM (E. coli), unless otherwise stated.

Cloning was carried out in accordance with standard molecular microbiology techniques. The pETM11-rpsF_10glu, pETM11-rimA_E47A and E. coli rpsF purification vectors were produced by ligating PCR fragments (amplified with primers 17/18, 11/12/15/16 and 13/14 respectively) between the NdeI and XhoI sites of plasmid pETNdeM-11 [38] as appropriate. Purification vectors for rimA, rimB and rimK were as described in [19] using primers 7–12. P. fluorescens SBW25 site-specific mutants were constructed using a modification of the protocol described elsewhere [39]. Up- and downstream flanking regions to the target modification site (rimBK, hfq or rpsF) were amplified using primers 19–25, 26–31 and 32–37 respectively. PCR products in each case were ligated into pTS1 between KpnI/EcoRI, NdeI/BamHI and XhoI/KpnI respectively. The resulting vectors were transformed into the target strain, and single crossovers were selected on Tet and re-streaked for single colonies. Cultures from single crossovers were grown overnight in LB medium, then a dilution series plated onto LB plates containing 5% w/v sucrose to enable sacB counterselection. Individual sucrose-resistant colonies were patched onto LB plates ± Tet, and Tet-sensitive colonies tested for successful gene modification by colony PCR followed by Sanger sequencing. Reverse transcriptase PCR (RT-PCR) and 5’ RACE were conducted using primers (1/2 and 3/4 respectively). qRT-PCR was performed as described in [19] using primers 5/6. qRT-PCR experiments were repeated three times independently and data are presented +/- standard error.

Protein purification was conducted after [19]. Briefly, overnight cultures of E. coli BL21-(DE3) pLysS were used to inoculate 1.0 litre overexpression cultures. These were then grown at 30°C to an OD₆₀₀ of 0.6, before protein expression was induced for 2 hours with 1mM IPTG. Cells were lysed using a cell disruptor (Avestin) and His₆-tagged proteins purified by NTA-Ni chromatography. 1 ml HiTrap chelating HP columns (Amersham) were equilibrated with 25 mM KH₂PO₄, 200 mM NaCl, pH 8.0 (SBW25 RimA/B/K), 50 mM Tris-Cl, 2.5% glycerol, pH 8.0 (SBW25 RpsF) or 50 mM Tris, 300 mM NaCl, 10 mM imidazole, pH 9.0 (E. coli RpsF), and loaded with cell lysate. Following protein immobilization, proteins were eluted with a linear gradient to 500 mM imidazole over a 15 ml elution volume.

ATPase activity was measured indirectly by monitoring NADH oxidation in a microplate spectrophotometer (BioTek Instruments) at 25°C. The reaction buffer consisted of 100 mM Tris-Cl (pH 9.0), 20 mM MgSO₄. Each 100 μL reaction contained 0.4 mM NADH, 0.8 mM phosphoenolpyruvic acid, 0.7 μL PK/LDH (Sigma) and was initiated by the addition of 10 μL ATP. Protein concentrations were as shown in the figure legends. Enzyme kinetics were determined by measuring A₃₄₀ at 1-minute intervals. Kinetic parameters were calculated by plotting the specific activity of the enzyme (nmol ATP hydrolyzed/min/mg protein) versus ATP concentration and by fitting the non-linear enzyme kinetics model (Michaelis-Menten) in GraphPad Prism. 25 μM cyclic dinucleotides were included where appropriate, as noted in the text.

The glutamation assay was adapted from [22]. Briefly, purified RpsF and Rim proteins at the concentrations indicated in the figure legend were incubated at room temperature for the indicated times in reaction buffer comprising 100 mM Tris-HCl pH 9.0, 20 mM L-glutamate (unless otherwise indicated), 20 mM ATP and 20 mM MgSO₄·7H₂O. Reactions were supplemented with cdG (Sigma) as indicated in the text. Protease assays were performed in 100 mM TRIS-HCl pH 9.0, 20 mM L-glutamate, 20 mM ATP, 20 mM MgSO₄·7H₂O with RimB concentrations as stated in the figure legend. For both experiments, samples were boiled in SDS loading buffer and analyzed by SDS-PAGE following completion of the reaction.

Bacterial growth was measured in a microplate spectrometer (BMG Labtech FLUOstar Omega) using 2 biological duplicates each providing 2 experimental replicates and are presented as mean +/- standard error. 150 μL of the indicated growth medium in each case was added to the internal 60 wells of clear-bottomed, black-walled 96-well microplates (Sigma). Carbon sources were added to a concentration of 0.4% w/v in each case. Growth was initiated by the addition of 5 μL of overnight cell culture (LB media, 28°C, shaking) normalized to an OD₆₀₀ of 0.01. Plates were incubated statically at 28°C and agitated immediately prior to each data acquisition step. Optical density was measured at 600nm and fluorescence was monitored at 460nm with an excitation wavelength of 355nm. Experiments were conducted at least twice independently.

To assess the colony morphology of the SBW25 rim/rpsF mutants, Kings B (KB) plates [40] were prepared containing 0.8% w/v agar, 1% w/v NaCl, 0.004% Congo Red dye, +/- 100 μM FeCl₂. Plates were allowed to dry for 45 min in a sterile flow chamber. Meanwhile SBW25 strains were grown over-day at 28°C in LB media with shaking. Cell densities were adjusted to give an optical density (λ 600nm) of 0.2 then 1.5 μL of each culture was spotted onto the media surface. Six replicates were prepared from two biological duplicates (three plates prepared from each). Plates were then incubated at 28°C for the time indicated in the figure legends before photographing, with a representative image shown in each case.

To quantify swarming diameter, Kings B (KB) media plates containing 0.3% agar w/v, 1% w/v NaCl, 0.004% Congo Red dye were produced and allowed to set and dry for 45 min in a sterile flow chamber. SBW25 strains were grown overnight at 28°C in LB media with shaking. The following morning, cell densities were adjusted to an optical density (λ 600nm) of 0.2 then 5 μL of each culture was spotted onto the media surface. Six replicates were prepared from two biological duplicates (three plates prepared from each). Plates were then incubated over-day at 28°C then overnight at 4°C before photographing, with a representative image shown in each case. All phenotypic experiments were conducted at least twice independently.

5’ RACE (Life Technologies) was performed on mRNA prepared from SBW25 cells grown overnight in M9 media with pyruvate (0.4% w/v) as sole carbon source. RNA was purified by column capture (Qiagen RNeasy Mini Kit). Purified RNA was subjected to additional DNase treatment (Turbo DNase, Ambion). The final PCR product from the 5’ RACE procedure was excised from a 1% agarose gel and purified by column capture (Macherey-Nagel Nucleospin mini kit). The purified product was then Sanger sequenced (Eurofins Scientific) to reveal the position of the transcriptional start site.

cDNA from SBW25 cells was prepared as an independent step prior to PCR. Total RNA was extracted from cells grown in M9 media containing pyruvate (0.4% w/v) as sole carbon source by column capture (Qiagen RNeasy Mini Kit). Purified RNA was subjected to additional DNase treatment (Turbo DNase, Ambion). The resulting RNA was then subjected to amplification (Merck WTA2 Amplification Kit) to produce cDNA that was subsequently purified by column capture (Qiagen QiaQuick PCR purification kit) and eluted in RNase-free water. PCR was performed on the cDNA using high-fidelity DNA polymerase (New England Biolabs Phusion polymerase) using a forward primer targeting the start of rimA (PFLU0263) and a reverse primer targeting the end of rimK (PFLU0261). The resulting PCR product was purified by column capture (Macherey-Nagel Nucleospin mini kit) and sequenced (Eurofins Scientific) to confirm the presence of the three rim genes.

50ml LB cultures were grown overnight at 28°C. Cultures were pelleted by centrifugation and resuspended into either 35ml LB media at 28°C or Rooting Solution (RS) at 4°C. Samples were subsequently incubated in the stated condition for 45 minutes. Crude ribosomes were prepared according to (64). In brief, cells were pelleted at either 28°C or 4°C according to the temperature at which they had been incubated. Pellets were resuspended into 15 ml Buffer A (20 mM Tris pH 7.5, 300 mM NH4Cl, 10 mM MgCl2, 0.5 mM EDTA, 6 mM β-mercaptoethanol (β-ME), 10 units/ml SuperASE-In (Ambion) + 0.1mM PMSF pre-equilibrated to the appropriate temperature. Cells were lysed using a cell disruptor (Avestin). Samples were made up to 35 ml using Buffer A and centrifuged at 30,000 x g for 50 minutes. 27 mls of sample was carefully withdrawn and centrifuged at 100,000 x g for 1 hour. The resulting crude ribosome pellet was rinsed with Buffer A and resuspended into 100 μL 50 mM Tris pH 8.0, 50mM NaCl.

Biological duplicate SBW25 cultures were grown overnight in LB media at 28°C. These cultures were used to inoculate 50ml volumes of M9-pyruvate minimal media, with and without Cas-aminoacids. Pyruvate and Cas-aminoacids were used at a concentration of 0.4% w/v. When cultures reached an optical density of 0.3 (measured at 600 nm), 1 ml of culture was removed, centrifuged and the resulting pellet resuspended into 100 μL supernatant + 100 μL x2 SDS loading buffer. The remaining cultures were allowed to grow overnight to reach stationary phase. 150 μL of a WT culture grown in M9-Pyr-CAS was taken and a proportionate volume of all other cultures (according to differences in optical density measured at 600nm) was taken. Pellets were resuspended as above. Prior to immunodetection, samples were briefly sonicated to reduce viscosity. For the detection of Hfq, 0.5 μL of each sample was run on WES ProteinSimple automated immunodetection system (Bio-Techne) according to the manufacturer’s instructions. For the detection of RpsF, following ribosome preparation, 1μL sample derived from cells grown in LB media at 28°C and 2μL sample derived from cells grown in RS at 4°C was run on the same instrument.

Following growth in M9-Pyr-CAS, the concentration of the soluble protein fraction from SBW25 cells was normalized. 140 μg total protein from each sample was loaded in a 25 μL volume onto one of two SDS-PAGE gels. Gels were blotted onto polyvinylidene difluoride (PVDF) membranes (Millipore). Membranes were incubated overnight in blocking solution (1x PBS pH 7.4, 0.01% Tween 20, 5% milk powder). Protein was subsequently detected with either 1/2000 anti-polyglutamate chain (AdipoGen) or 1/200 anti-RpsF (Dundee Cell Products) antibodies and 1/6000 anti-rabbit secondary antibody (Sigma). Bound antibody was visualized using ECL chemiluminescent detection reagent (GE Healthcare).

SDS gel slices containing protein samples of interest were washed, treated with DTT and iodoacetamide, and digested with trypsin according to standard procedures. Peptides were extracted from the gels and analysed by LC-MS/MS on an Orbitrap-Fusion mass spectrometer (Thermo Fisher, Hemel Hempstead, UK) equipped with an UltiMate 3000 RSLCnano System using an Acclaim PepMap C18 column (2 μm, 75 μm x 500mm, Thermo). Aliquots of the tryptic digests were loaded and trapped using a pre-column which was then switched in-line to the analytical column for separation. Peptides were separated using a gradient of acetonitrile at a flow rate of 0.25 μl min-1 with the following steps of solvents A (water, 0.1% formic acid) and B (80% acetonitrile, 0.15 formic acid): 0–3 min 3% B (trap only); 3–4 min increase B to 7%; 5–40 min increase B to 50%; 40–45 min increase B to 65%; followed by a ramp to 99% B and re-equilibration to 3% B.

Data dependent analysis was performed using HCD fragmentation with the following parameters: positive ion mode, orbitrap MS resolution = 120k, mass range (quadrupole) = 300–1800 m/z, MS2 top20 in ion trap, threshold 1.5e⁴, isolation window 1.6 Da, charge states 2–7, AGC target 2e⁴, max inject time 35 ms, dynamic exclusion 1 count, 15 s exclusion, exclusion mass window ±5 ppm. MS scans were saved in profile mode and MS2 scans were saved in centroid mode.

To generate recalibrated peaklists, MaxQuant 1.6.0.16 was used, and the database search was performed with the generated HCD peak lists using Mascot 2.4.1 (Matrixscience, UK). The search was performed on a Pseudomonas fluorescens SBW25 protein database (Uniprot, 20140727, 21,935 sequences) to which copies of the RpsF sequence with 1–10 additional glutamate residues (E) at the C-terminus was added. For the search a precursor tolerance of 6 ppm and a fragment tolerance of 0.6 Da was used. The enzyme was set to trypsin/P with a maximum of 2 allowed missed cleavages. Oxidation (M), deamidation (N, Q), and acetylation (N-terminus) were set as variable modifications, carbamido-methylation (CAM) of cysteine as fixed modification. The Mascot search results were imported into Scaffold 4.4.1.1 (www.proteomsoftware.com) using identification probabilities of 99% for proteins and 95% or 0% for peptides, as discussed in the results. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE [41] partner repository with the dataset identifier PXD017233.

50 ml overnight cultures of SBW25 WT and ΔrimK were grown at 28°C in LB medium with shaking. Cultures were then pelleted by centrifugation and resuspended in equal volumes of A) carbon-free rooting solution [19], pre-cooled to 8°C, and B) LB medium, pre-heated to 28°C. Cultures were incubated for 45 minutes with shaking at 8 or 28°C as appropriate, then cellular activity was halted by addition of 30 ml of RNAlater [saturated (NH₄)₂SO₄, 16.7 mM Na-Citrate, 13.3 mM EDTA, pH 5.2] containing protease inhibitors. Cell samples were then pelleted, washed three times with 10 mM HEPES pH 8.0 + protease inhibitors, before re-suspension to a final volume of 200 μL. 700 μL pre-cooled RLT + β-mercaptoethanol buffer (RNeasy Mini Kit, QIAGEN) was added and samples lysed with two 30 s Ribolyser pulses at speed setting 6.5. Supernatant was removed, and the soluble fractions separated by ultracentrifugation (279,000 g, 30 min, 4°C).

After determination of protein concentration, the soluble proteins were precipitated with chloroform-methanol and subjected to iTRAQ 4-plex quantification. The experiment was performed with 2 biological replicates. In each replicate the samples were labelled with an iTRAQ 4 plex kit (Sciex) as follows: 114 = WT LB28, 115 = ΔrimK LB28, 116 = WT RS8, 117 = ΔrimK RS8. The labelled samples were combined and fractionated using the Pierce High pH Reversed-Phase Peptide Fractionation Kit producing 7–8 fractions. After analysis on an Orbitrap Fusion (Thermo) using MS3 synchronous precursor selection based on methods described previously [7], the raw data from both replicates were combined and processed in Proteome Discoverer 2.4 (Thermo) with the following main parameters: protein sequence database: P. fluorescens SBW25 (Uniprot, Feb/2016, 6388 entries); variable modifications: oxidation (M), deamidation (N,Q); Percolator strict FDR target: 0.01; reporter ion quantifier: most confident centroid, 20 ppm, HCD, MS3; consensus workflow for statistical analysis: replicates with nested design; use unique peptides for protein groups only, co-isolation threshold 50%, imputation: low abundance resampling, ratio based on pairwise ratios, hypothesis test: t-test (background based) generating adjusted p-values according to Benjamini-Hochberg. The protein results table was exported from Proteome Discoverer and used to generate the final protein expression tables and plots (Fig 7, S5 Fig and S2 Table) in The R Project for Statistical Computing.

Ribosomal profiling was conducted as described in [7]. SBW25 cultures were grown at 28°C in M9-pyr-CAS medium to the late exponential phase, then cells were harvested by rapid filtration as described in [42]. Collected cells were flash frozen in liquid nitrogen and cryogenically pulverized by mixer milling (Retsch), then thawed and clarified by centrifugation. Resulting lysates were digested with MNase, quenched with EGTA and resolved by sucrose density gradient ultracentrifugation. Ribosome-protected mRNA footprints were processed as previously described [7, 42] and sequenced by Illumina HiSeq2000. Sequencing reads in fastq files were adaptor trimmed using a Perl script that implemented the procedure described in [43]. Next, ribosomal RNA sequences were filtered out by aligning them against a Bowtie2 index of the SBW25 rRNAs. The remaining reads were then aligned to the SBW25 genome to produce SAM files, which were used to calculate the centre-weighted coverage at each nucleotide position in the genome. For this, we used a Perl script to select alignments of between 23 and 41 nucleotides in length and counted for nucleotide positions after trimming 11 nucleotide positions from either end of the alignment. This was done separately for reads aligning to the forward and reverse genome strands and the centre-weighted coverage was stored in separate files for each strand. A separate Perl script was used to calculate RPKM values for each gene based on strand specific centre-weighted coverages along the genome. The limma function plotMDS was then used to make PCA plots. The ribosome profiling data has been submitted to ArrayExpress, with accession number E-MTAB-5408.

For each model, the chemical equations were translated into ordinary differential equations using mass action kinetics (S1 Table). Thermodynamic constraints were considered to ensure consistency between reaction rates via different routes, thereby reducing the number of free kinetic parameters. Kinetic forward rate constants were set to 10⁹ M^-1s^-1 (diffusion limited) and equilibrium dissociation constants, K_d, were–based on previous estimates–set initially to 1 μM and then optimized to fit with experimental data (see below). We used the ODE solver Vern7 from the Julia Differential Equations library to solve the chemical kinetics equations, which is suitable for high accuracy, non-stiff systems. Initial concentrations for cdG, RimK, RimA, and RimB were set to 1 μM unless specified otherwise, for instance when mimicking the conditions of a specific experiment as in S2 Fig. where [cdG] was 25 μM. Solution of the ODE allowed us to determine steady state values for all RimK species for each model. These steady state concentrations were used to calculate the enzymatic modification of RpsF. K_m and k_cat values for each species of RimK and K_d values between Rim proteins were optimized to fit the model simulations to available experimental data, S2 Fig., using the Julia BlackBoxOptim library. Contributions of each RimK species to the kinetics with multiple species present were determined via optimization using all available data set simultaneously using Michaelis-Menton equations for the enzymatic reactions. Glutamation of the population was computed using a stochastic simulation that consisted of the addition of glutamate unit, the removal of a unit, or doing nothing. These probabilities are dependent on the concentration of the various RimK species and RimB. Average behaviour was computed from 1,000,000 independent stochastic simulations. All code was written in the free and open source scientific programming language Julia.