The RpoS/σS sigma subunit of RNA polymerase is the master regulator of the general stress response in many Gram-negative bacteria. Extensive studies have been conducted on σS-regulated gene expression at the transcriptional level. In contrast, very limited information regarding the impact of σS on global protein production is available. In this study, we used a mass spectrometry-based proteomics approach to explore the wide σS-dependent proteome of the human pathogen Salmonella enterica serovar Typhimurium. Our present goals were twofold: (1) to survey the protein changes associated with the ΔrpoS mutation and (2) to assess the coding capacity of σS-dependent small RNAs. Our proteomics data, and complementary assays, unravelled the large impact of σS on the Salmonella proteome, and validated expression and σS regulation of twenty uncharacterized small proteins of 27 to 96 amino acids. Furthermore, a large number of genes regulated at the protein level only were identified, suggesting that post-transcriptional regulation is an important component of the σS response. Novel aspects of σS in the control of important catabolic pathways such as myo-inositol, L-fucose, propanediol, and ethanolamine were illuminated by this work, providing new insights into the physiological remodelling involved in bacterial adaptation to a non-actively growing state.
Salmonella enterica serovar Typhimurium (hereafter referred as Salmonella or S. Typhimurium) is a facultative intracellular bacterial pathogen capable of colonizing a wide range of hosts, including humans and many animal species. This serotype is ubiquitous in the environment, and its capacity to adapt to many environmental conditions depends on its ability to integrate various environmental signals to coordinate gene expression appropriately. The alternative sigma factor σS (also called σ38 or RpoS) of RNA polymerase (RNAP) controls a global adaptive response allowing many Gram-negative bacteria to survive nutrient deprivation and environmental stresses1,2,3. σS also contributes to virulence and biofilm formation of S. Typhimurium3,4,5. In contrast to the housekeeping sigma factor σ70, σS is almost undetectable in early exponential phase. Its expression is induced in stationary phase, or in response to various stresses, by a fine-tuned combination of transcriptional, translational and proteolytic controls2, 3. σS and σ70 bind to almost identical –35 and –10 promoter elements, but σS is more tolerant than σ70 with respect to deviations from the consensus promoter sequences, especially in the −35 region3, 6,7,8,9,10. The activities of the σS- and σ70- dependent RNAP can be modulated by additional DNA-binding regulatory proteins, which can also contribute to σ factor selectivity at a given promoter2, 3.
Regulatory action of σS has been extensively studied at the transcriptional level. Global transcriptomic analyses have revealed that σS controls, directly or indirectly, the expression of 10–20% of the genome of Escherichia coli K123, 7, 11,12,13,14,15. We have recently used RNA-sequencing to unravel the σS-dependent transcriptome in S. Typhimurium ATCC1402816. These studies have revealed a major effect of σS on the remodelling of metabolism and membrane functions, and have highlighted the importance of down-regulation of gene expression by σS 16, 17. Negative regulation by σS is an active process requiring σS binding to DNA17. Mechanisms of repression could be direct in some cases14, 17, 18, through competition with σ70 for promoter binding14, 17, or indirect, via transcription activation of repressor molecules3, 16, 17. In particular, we have unravelled σS-regulation of a large number of small RNAs (sRNAs)16, among which some might endow σS with repressor functions, by affecting mRNA stability and/or translation or protein stability.
Compared to transcriptomic data, very limited information regarding the effect of σS on global protein production is available. The proteome of wild type and ΔrpoS strains have been compared in a few studies by using bi-dimensional gel electrophoresis3, 19,20,21,22. Although the expression of dozens of proteins correlated with transcriptomic data, only a small fraction of the proteome was assessed in those studies, and the σS-dependent proteome was largely underestimated. Comparative proteomic profiling using mass spectrometry (MS) is a powerful tool for investigating differences in global protein abundance that occur in response to a mutation, or a specific condition. In bacteria, MS-based proteomics is under-represented when compared to transcriptomic studies and has not been used, to our knowledge, to characterize the σS-dependent proteome.
In this study, we performed label-free MS-based relative quantification of protein abundance in the wild type and ΔrpoS strains of S. Typhimurium. Our present goals were twofold: (1) to survey the protein changes associated with the ΔrpoS mutation and (2) to assess the coding capacity of σS-dependent small RNAs (sRNAs) revealed by our transcriptomic analyses16 (Supplementary Fig. S1). Our proteomics data, and complementary assays, unravelled the large impact of σS on the Salmonella proteome in stationary phase, and validated translation and σS regulation of more than twenty sRNAs. As expected, the majority of genes up-regulated by σS at the protein level were also up-regulated at the transcript level16. In sharp contrast, our data revealed a large number of genes down-regulated at the protein level, but not at the transcript level, suggesting that post-transcriptional regulation plays a larger role in σS gene regulation than previously recognized. In addition, this study unravelled new facets of σS in metabolism rewiring during bacterial adaptation to a non-actively growing state. The complexity of the σS regulatory network and its impact on cell physiology, revealed by this study, points to a key role of σS, at the transcriptional and post-transcriptional levels, in maintaining the delicate balance between cellular resistance during quiescence and re-growth potential, under diverse environmental conditions.
Results and Discussion
STnc1330, STnc1110 and IsrI are coding sRNAs
Our recent RNA-sequencing data in S. Typhimurium ATCC1402816 revealed a number of transcripts showing high relative abundance and strong σS-dependency in stationary phase, including the σS-dependent sRNAs STnc1110, SdsR, STnc1330, IsrI, and SraL16 (Supplementary Fig. S1). Regulatory functions have been described for SdsR23, 24 and SraL25, but not yet for IsrI, STnc1110 and STnc1330. Surprisingly, open reading frames (ORFs) showing good potential ribosome binding sites were predicted in the sequences of IsrI26, and of STnc1110 and STnc1330 which originally were considered non-coding27 (Fig. 1a and Supplementary Fig. S2). Moreover, whereas these sRNAs have been reported to be specific to Salmonella28, 29, the predicted ORFs in STnc1110 and STnc1330 were actually orthologs of the E. coli yncL and yohP genes, respectively (Supplementary Fig. S2). In E. coli, yncL and yohP direct the synthesis of hydrophobic membrane proteins of 31 and 27 amino acids, respectively, and of unknown function30, 31. The predicted ORF in isrI, STM14_3199, encodes a product of 62 amino acids (Supplementary Fig. S3). IsrI26 is located on the Gifsy-1 prophage in S. Typhimurium ATCC14028, SL1344 and LT227. Interestingly, STM14_3199 is paralogous to STM14_1447 (75% identity at the amino acid sequence level, Supplementary Fig. S3). STM14_1447 is carried on the Gifsy-3 prophage, which is present in ATCC14028, but absent from strains LT2 and SL134432. Homologs of STM14_3199 were found in a few other enterobacterial species (Supplementary Fig. S3).
To assess the coding capacity of IsrI, STnc1110 and STnc1330, a DNA sequence encoding a 3xFlag epitope was inserted at the 3′ end of each of the predicted ORFs, and production of the flag-tagged proteins was followed by immunodetection, using an anti-flag antibody. A 3xFlag-tagged YohP protein was produced in high amounts in ATCC14028, and was almost undetectable in the ΔrpoS mutant (Fig. 1b). Production of YohP was also strongly dependent on the growth phase (Fig. 1b). Although no flag-tagged protein was detected for STM14_1932/yncL and STM14_3199, translational STM14_3199-lacZ and STM14_1932-lacZ fusions were expressed in ATCC14028, and their expression required σS (Fig. 1c). The STM14_3199 and STM14_1932 flag-tagged products may be unstable, or produced at levels too low to be detected under the experimental conditions used. Altogether, these data indicated that STnc1330, and likely STnc1110 and IsrI, are coding sRNAs.
Additional coding sRNAs tightly controlled by σS
Besides STnc1110, STnc1330 and IsrI, twenty-nine additional small transcripts, abundant in stationary phase and tightly controlled by σS, were assigned to small annotated ORFs (Supplementary Figs S1 and S4, Supplementary Table S1). To validate these data, we first examined the genomic context and sequence features of these ORFs. This analysis suggested that transcripts assigned to the small ORFs STM14_0419, STM14_1559, and STM14_5096, and the pseudogene STM14_1274, more likely correspond to long 5′ UnTranslated Regions (UTR) of the σS-dependent genes STM14_0421, STM14_1558, STM14_5097, and STM14_1275, respectively (Supplementary Fig. S4 and Table S1). This hypothesis is consistent with the non-canonical start codons and lack of ribosome binding sites for the putative ORFs STM14_0419, STM14_1559 and STM14_5096, and the location of transcription start sites and RNA reads in these regions (Supplementary Fig. S4). Of the remaining twenty-five small annotated ORFs, only six have been functionally characterized, at least to some extent (ecnB, yqaE, chaB, osmB, yahO and yciG, Supplementary Table S1 and references therein). Nineteen ORFs were putative or of unknown function (Table 1). Examination of their sequence features, and alignment of their predicted amino acid sequences with that of identified homologs in other bacterial genomes, prompted us to re-annotate the start codon of STM14_2239, STM14_2409, and STM14_5481 (Table 1, Supplementary Fig. S5, Supplementary Dataset S1). The uncharacterized ORFs ymdF and STM14_1829 are paralogous to yciG (Table 1, Supplementary Fig. S6). YciG33, 34 belongs to the group of “hydrophilins”, proteins defined by high glycine content and hydrophilicity index35. Interestingly, while some of the uncharacterized small ORFs appear to be restricted to bacteria of the Enterobacteriaceae family, others are more widely distributed (Table 1 and Supplementary Dataset S1) and might be involved in conserved biological processes. In the following part of the work, a LC-MS-based proteomics approach was used to validate production and σS regulation of these small proteins, and to assess, to our knowledge for the first time, global effects of σS at the protein level in Salmonella.
Global effects of σS on protein abundance in Salmonella
A comprehensive quantitative proteomic analysis was performed using the wild type and ΔrpoS strains of Salmonella ATCC14028, grown in nutrient-rich LB medium to late stationary phase, i.e. the growth conditions previously used to characterize the σS-transcriptome16. Three biological replicates of wild type and ΔrpoS strains were grown, and subjected to proteome-wide label-free quantification. After harvesting, cells were lysed, proteins were digested with rLys-C and trypsin, and the resulting peptides were analysed by nLC-MS/MS. A statistical analysis of relative changes in protein abundance between the wild type strain and the ΔrpoS mutant was performed using Perseus, the companion software of MaxQuant (see Supplementary Methods for details). A protein was considered as being “present” in a strain, if observed in at least two of the three replicates. Proteins detected in less than two replicates of one strain and in at least two replicates of the other strain, were designated as “exclusive” to that latter strain. Two-sided T-tests of the base-2 logarithm (log2) transformed intensity values were employed, using three False Discovery Rates (FDR of 5%, 1% and 0.1%) to identify differentially abundant proteins with different degrees of statistical significance. A complete list of all proteins, along with their relative abundance pattern and sequence coverage is provided in Supplementary Dataset S2. The (log2) fold change of the level of each protein in the ΔrpoS mutant, with respect to that in wild type strain, is shown. Positive values indicate higher abundance in the mutant than the wild type. Negative values represent lower levels in the mutant than the wild type.
A total of 2444 Salmonella proteins were identified (Fig. 2a, Supplementary Dataset S2), which corresponds to about 46% coverage of the ATCC14028 proteome. The remaining proteins might not be expressed under the growth condition employed in this study, or they might be expressed at levels too low to be detected. Also, hydrophobic integral membrane proteins are difficult to identify by MS-based proteomics36.
Expression of σS was associated with the significant increase in the level of 401 proteins and a decrease in abundance of 538 proteins (Fig. 2a, Supplementary Datasets S2–S3). For these two groups, the change in the level will be referred to as “up- and down-regulation” regardless of the specific mechanism (or combination of mechanisms) leading to such change. The log2 ratios of fold changes were at least 0.68 (representing a minimum fold change value of 1.5). Of note, the log2 fold changes in protein abundance were the highest when the FDR was low, and were higher for up-regulated than for down-regulated proteins (Supplementary Dataset S3). The set of 401 up-regulated proteins contained 54 proteins exclusively found in the wild-type strain (Fig. 2b). The set of 538 down-regulated proteins contains 79 proteins found exclusively in the ΔrpoS mutant (Fig. 2c).
σS affected directly or indirectly the abundance of proteins involved in a variety of processes including metabolism, membrane trafficking, regulation, and stress resistance, and of many uncharacterized proteins (Supplementary Dataset S3). These data confirmed at the proteome level the large impact of σS in bacterial adaptation to a non-growing lifestyle.
Small uncharacterized proteins in the σS-network
Examination of the proteome data sets allowed identification of protein products for 80% of genes showing strong σS-dependency at the transcript level (Supplementary Table S1). The twelve proteins not identified by LC-MS all contain lysine and arginine residues, and are thus likely sensitive to protease digestion used to generate peptides for LC-MS. However, nine are likely associated with the membrane (Table 1 and Supplementary Table S1) and might not be soluble enough to be detected by this approach.
Interestingly, proteins were identified for more than half of the uncharacterized σS-dependent small ORFs (Table 1 and Supplementary Table S1). Of note, peptides identified by MS allowed discrimination between the yciG, ymdF and STM14_1829 gene products, despites their high level of sequence conservation (Supplementary Fig. S6). The coding capacity of the uncharacterized small ORFs, for which no protein was identified by LC-MS (STM14_2173, STM14_2189, STM14_5292, STM14_5469, STM14_5479, STM14_5481), was assessed by immunodetection of the corresponding 3xFlag-tagged proteins. Proteins were detected for all of them, except STM14_5469 and STM14_5479 (Fig. 1b). In the case of STM14_5469, canonical translation initiation signals are present (Supplementary Fig. S5), and an STM14_5469-lacZ translational fusion was expressed and activated by σS (Fig. 1c). Thus, in this case, the 3xFlag tag may interfere with the cellular localization and/or the stability of the protein, explaining why it was not detected. No significant expression of the STM14_5479-lacZ translational fusion was detected in the growth condition used (Fig. 1c). STM14_5479 does not seem to have a canonical ribosome-binding site (Supplementary Fig. S5), suggesting that the putative ORF might not be translated. We cannot exclude, however, that its translation requires a specific condition. In contrast to STM14_5469, STM14–5479 is conserved in only a few genomes of Salmonella (Table 1 and Supplementary Dataset S1).
With the exception of STM14_5292/ytfK, a good correlation was found for changes in expression of the uncharacterized σS-dependent small ORFs at the RNA and protein levels (Fig. 1b and Supplementary Table S1). This finding is consistent with transcription activation of these genes by σS. However, many genes transcriptionally activated by σS can also be transcribed by σ70 during exponential growth, either from a unique promoter recognized by the two sigmas, or from different promoters3, 8. To assess growth-phase regulation, expression of the 3xFlag-tagged small proteins was examined in logarithmic phase of growth (Fig. 1b). With the exception of YtfK, production of the proteins was strongly dependent on stationary phase, suggesting a tight control by σS under these conditions. Consistent with this hypothesis, promoter sequences of these genes28, 37 showed typical features of canonical σS- promoters in the −10 region3, 8, 10, and no conservation of the σ70 consensus in the −35 region (Fig. 1d). E. coli K12 orthologs of the small genes also showed σS activation in transcriptomic analyses7, 11,12,13,14 (Table 1), including ytfK, which displayed in our studies regulation by σS at the RNA level only. The impact of σS on YtfK protein abundance in E. coli K12 remains to be determined.
Comparison of the global proteomic and transcriptomic σS profiles
One of the main issues addressed in this study was to determine to which extent changes induced by the ΔrpoS mutation at the transcriptome and proteome levels are correlated. Our proteomics approach identified a protein for about half of genes showing detectable expression by RNA-seq16 (Supplementary Dataset S3). The other genes might be poorly transcribed and/or translated, or their products might be unstable or hardly detectable by MS.
For proteins up-regulated by σS, a high correlation was observed for changes in gene expression at the transcript and protein levels, especially for proteins selected with a FDR of 0.1% and 1% (Fig. 2b, Supplementary Dataset S3). Half of the up-regulated proteins selected with a 5% FDR were encoded by genes showing no significant (p-value > 0.05) regulation by σS at the transcriptional level. This apparent discrepancy could result from some proteins being encoded by genes activated by σS post-transcriptionally, or genes with a low level of transcription that excluded them from the comparative transcriptomic analysis. It is also possible that some proteins are damaged and degraded in the ΔrpoS mutant as a result of increased endogenous stress (i.e. endogenous oxidative stress may target proteins that are very sensitive to carbonylation and favour their degradation).
Strikingly, a different picture emerged for proteins down-regulated by σS (Fig. 2c, Supplementary Dataset S3). Most of these proteins were encoded by genes showing no significant regulation by σS at the RNA level (p > 0.05). Furthermore, about 5% of down-regulated proteins were encoded by genes activated by σS at the RNA level, suggesting an inverse correlation between regulation by σS at the protein and RNA levels. Most interestingly, these unexpected regulatory patterns were observed independently of the FDR value. These findings suggest that the observed changes at the protein level were the result of post-transcriptional regulatory effects by σS. It is also possible that some genes were selected in the proteomics approach, and not by RNA-seq, due to differences in the sensitivity of the two methods.
Overall, the proteomics data confirmed the positive effect of σS on transcription of genes involved in central energy metabolism (glycolysis and the pentose phosphate pathway, mixed acid fermentation), glycogen and trehalose metabolism, arginine degradation, putrescine synthesis and degradation, and antioxidant pathways (catalases, superoxide dismutase, glutaredoxins, ferritins and Fe-S repair proteins) (see details in ref. 16 and Supplementary Dataset S3). Genes down-regulated by σS at both the protein and RNA levels encode porins, TCA cycle enzymes and proteins of unknown functions16, 17 (Supplementary Dataset S3).
Regulatory effects of σS at the RNA level, not transferred to the protein level
The abundance of 1504 proteins was found unchanged in the ΔrpoS mutant, with respect to the wild type strain (Fig. 2a, Supplementary Dataset S3). Surprisingly, a fraction of the corresponding genes (309 genes, Fig. 2d) was differentially expressed in RNAseq experiments, among which 135 showed highly significant regulation16 (p < 0.001, Supplementary Dataset S3). Thus, for these genes, σS regulation at the RNA level was not transferred to the protein level. This regulatory profile was similar to that of STM14_5292 (ytfK), one of the small genes strongly activated by σS at the RNA level16 (Supplementary Table S1) but not at the steady state level of the protein (Fig. 1b). Consistent with these data, the ΔrpoS mutation strongly impaired expression of a transcriptional STM14_5292-lacZ fusion, but had a minor (and positive) impact on the expression of the translational STM14_5292-lacZ fusion (Fig. 1c). The LacZ protein was fused at the C- terminus of YtfK to account for translation and turnover of the protein. Since protein stability usually exceeds transcript stability in bacteria38, σS activation of STM14_5292 transcription might be masked by a long half-life of YtfK produced during the exponential phase of growth (Fig. 1b). Alternatively, translation and/or stability of the protein might be improved in the absence of σS, compensating the defect in gene transcription. More generally, it is possible that, in some cases, transcription activation by σS contributes to proteostasis by compensating for impaired protein translation or stability in stationary phase, due to accumulation of σS itself or to other signals/regulators. Alternatively, when σS function is restricted, other signals/regulators may be activated to override at the protein level a decrease in transcription of a number of σS-dependent genes.
Interestingly, a few genes, such as the prpBCDE genes, showed antagonistic regulation by σS at the RNA and protein levels (Fig. 3a). The prpBCDE operon encodes enzymes of the methyl citrate cycle involved in propionate catabolism39 (Fig. 4). The prp genes were activated by σS at the RNA level16 and down-regulated at the protein level (Fig. 3a). Consistent with these data, σS activated a transcriptional prpE-lacZ fusion, but had a negative impact on the translational prpE-lacZ fusion (Fig. 3b). The negative effect of σS on the abundance of the Prp proteins is consistent with the higher capability of the ΔrpoS mutant, compared to the wild type strain, to growth on propionate as a sole carbon source40 (Fig. 3c). As expected, a prpB mutation prevented the growth of the ΔrpoS mutant, and the rpoS gene on pSTK4 complemented the ΔrpoS mutation (Fig. 3c). In E. coli K12, the prp genes are also positively controlled by σS at the RNA level13. However, since rpoS disruption did not improve E. coli K12 growth on propionate41, prp regulation may be different in E. coli and Salmonella. This may be related to the finding that the prp operon of E. coli, in contrast to that of S. Typhimurium, is interrupted by repetitive elements (Supplementary Fig. S8 and references therein).
The complex regulation of the prp genes by σS in Salmonella might account for the dual effects of propionate catabolism on bacterial growth39. Short chain fatty acids, like propionate, have antibacterial activities and are used as preservatives in food industry39. Salmonella may be frequently exposed to propionate in the anaerobic environment of the gut, this exposure reducing its capacity to invade and colonize intestinal epithelial cells42. Thus, in such environment, activation of propionate catabolism may provide a competitive advantage to Salmonella42. However, catabolism of propionate is a risk because 2-methylcitrate is a potent inhibitor of cell growth39. Salmonella may thus be under constant pressure to maintain levels of both propionate and 2-methylcitrate low enough to avoid the negative effects caused by accumulation of these compounds. Under the aerobic conditions used here, where the activity of the TCA cycle is down-regulated by σS, excessive propionate catabolism may have an inhibitory effect due to accumulation of 2-methylcitrate (Fig. 4). Lowering abundance of the Prp proteins under these conditions may thus confer a fitness advantage. A dual and antagonistic control of prp expression would allow σS to monitor balanced Prp expression, and to shift rapidly the balance according to the cell needs under changing environmental conditions.
Novel physiological effects of σS revealed by proteomic analyses
Among genes negatively controlled by σS at the protein level, we noticed several products of the iol, pdu, eut and fuc genes, involved in the catabolism of myo-inositol, propanediol, ethanolamine and L-fucose, respectively (Fig. 4 and Supplementary Dataset S3). Western blot analysis of relative levels, in the ΔrpoS and wild type strains, of products of the fuc, iol, pdu and eut genes validated the proteomics data (Fig. 5a). Also consistent with these results, the ΔrpoS mutant grew better than the wild type strain at the expense of myo-inositol, fucose, propanediol and ethanolamine as sole carbon sources (Fig. 5b) and the ΔrpoS mutation was complemented by a plasmid carrying rpoS (Supplementary Fig. S9). Production of the MgsA protein that converts dihydroacetone phosphate (DHAP), a product of myo-inositol and fucose degradation, into methylglyoxylate was also down-regulated by σS (Figs 4 and 5a and Supplementary Dataset S3).
To our knowledge, this is the first report of positive effects of a ΔrpoS mutation on catabolism of myo-inositol, L-fucose, ethanolamine and propanediol. Transcript levels of the corresponding genes were not significantly affected by the ΔrpoS mutation16 (Supplementary Dataset S3), suggesting that σS down-regulates production of the gene products post-transcriptionally. The underlying molecular mechanisms are unknown and will be the subject of more detailed studies. Interestingly, the Salmonella prp and pdu genes are down-regulated, at the protein level only, by the RNA-binding protein Hfq43 (Fig. 6a). Increased abundance of the EutE protein was also observed in the Salmonella Δhfq mutant, compared to the wild type (Fig. 6a). It is thus possible that σS down-regulates expression of the prp, pdu and eut genes using σS-dependent sRNAs16, 28, together with Hfq. However, since Hfq favours rpoS translation2, 3, 44 (Fig. 6a), effects of Hfq on expression of these genes may be indirect via σS regulation. An effect of Hfq via σS would be consistent with the observed activation by Hfq of several σS-activated genes43, 44 (Supplementary Table S1).
As mentioned above, transcript levels of the pdu, iol, eut, fuc, and mgsA genes were not significantly increased in the ΔrpoS strain, compared to the wild type strain16. However, their low transcript abundances (Supplementary Dataset S3) might have excluded them from the list of σS-activated genes during the comparative transcriptomic analysis16. To address this issue, regulation of some of these genes was assessed in strains into which lacZ was inserted downstream of the flag-tagged genes to create transcriptional fusions (Fig. 6b). The lacZ insertion did not affect σS regulation of the flag-tagged proteins (compare data in Figs 5a and 6b). No or moderate effects of the ΔrpoS mutation were observed on the expression of lacZ fusions in the mgsA and pdu genes (Fig. 6b). In sharp contrast, expression levels of lacZ fusions in the fucU, fucA and STM14_5317 genes were strongly decreased in the ΔrpoS mutant, compared to the wild type strain, indicating that these genes belong to the class of genes showing antagonistic regulation by σS. These data suggest that the number of genes transcriptionally activated by σS might have been under-estimated in the RNA-sequencing analysis for genes with low transcript abundances16. As a consequence, the percentage of genes showing antagonistic regulation by σS might exceed 5% (Fig. 2c). Considering the extensive overlapping between σ factor binding sites and the similarity in the −10 box sequence motifs of σ70 and σS promoters3, 9, 14, it is possible that, in stationary phase, σS provides basal transcript levels for genes otherwise transcribed by other σ, while controlling their expression post-transcriptionally.
Like propionate, ethanolamine, propanediol, L-fucose and myo-inositol are abundant compounds in food and in anaerobic environments of the gut, and their utilization may confer an advantage to Salmonella within natural environments45,46,47. It is noteworthy that myo-inositol catabolism confers a fitness advantage to Legionella pneumophila in macrophages and amoebae, and that σS has a positive effect on expression of the iol operon and myo-inositol uptake48. It is possible that expression of the Salmonella prp, iol and fuc genes is activated in vivo by σS and provides fitness advantages to the pathogen. However, as hypothesized above for the prp genes, in the growth conditions used in the present study, the amount of the fuc, pdu, iol, and eut products may be adjusted according to the activity of the TCA cycle (Fig. 4). In addition, catabolism of myo-inositol and L-fucose produces DHAP that can be converted by the MgsA protein into methylglyoxal, a cytotoxic and mutagenic product affecting DNA and proteins49. DHAP can also be produced from glyceralhedyde-3 phosphate via excessive carbon flux through glycolysis. Since σS activates expression of genes in the glycolytic pathway (Supplementary Dataset S3 and ref. 16), an increased carbon flux in the first half of glycolysis, and myo-inositol and L-fucose degradation, might give rise to excess DHAP and its conversion to methylglyoxal by MgsA, a process detrimental for the cell. Therefore, down-regulation of mgsA by σS might confer a fitness advantage to the cell.
From the few examples discussed here, it is tempting to speculate that a number of negative effects of σS on protein production, in the growth conditions used, aim at buffering toxic effects that metabolic rewiring by σS might generate in stationary phase cells. Since several genes appear to be down-regulated by σS only or mainly at the protein level in Salmonella (Supplementary Dataset S3), their study might pinpoint to yet uncovered σS post-transcriptional mechanisms maintaining the delicate balance between cellular preservation during quiescence and re-growth potential, under various conditions.
Our global proteomic analyses provide new clues about physiological and regulatory mechanisms controlled by σS in non-actively growing Salmonella. σS, directly or indirectly, modulates the expression of 38% of the observed S. Typhimurium proteome, including a broad spectrum of Salmonella proteins needed for various biological processes, and proteins that have not been functionally characterized. Western blotting to a number of proteins and growth assays have validated these results.
One part of the study focused more specifically on σS-dependent sRNAs coding for uncharacterized small proteins that might play as-yet-unidentified roles in Salmonella fitness during the quiescence phase. Our data suggest that a number of annotated sRNAs identified in global transcriptome analyses are coding sRNAs, notably STnc1330 and probably also IsrI and STnc1110. Some coding sRNAs, such as IsrI, associate with Hfq44 (Supplementary Table S1) and might thus have a dual function, as small and messenger RNA.
Study of small proteins is an emerging research topic in bacteria and eukaryotic cells50. Due to their small size, these proteins usually act by modifying the activity of larger proteins/complexes or RNAs, via physical interactions with them. A fraction of the small σS-dependent proteins identified here could interact with membrane proteins or complexes, and modulate important features of the membrane, such as permeability and transport, or have a stabilizing role, contributing to the known role of rpoS in membrane resilience in stationary phase51. For their part, small soluble σS-dependent proteins could act as chaperones, facilitate protein synthesis or protein degradation and autophagy upon stress exposure, or they may have a toxic function. The main function of these small proteins may be to act as regulators, to increase plasticity and dynamics of adaptive functions. Such “modulatory activities” might be difficult to unravel, which could explain why a very few number of small proteins have been functionally characterized so far. In addition, some proteins may have partially redundant functions, such as those encoded by the paralogous genes identified in this study, which might complicate the phenotypic analyses of mutants. Identification of potential biological interactions involving small σS-dependent proteins in future studies could provide insights into their functions.
Until recently, the σS response was believed to be predominantly transcriptional. In this study, direct dependency for gene expression between transcript levels and protein levels was observed for a large majority of proteins up-regulated by σS, while the majority of genes encoding down-regulated proteins showed no changes in mRNA levels. The observation that only 20–25% of protein changes in this category can be matched to significant transcriptional changes was unexpected, but can be explained by altered translation or turnover rates of proteins, irrespective of mRNA levels. Such modifications can result from σS-dependent regulatory mechanisms, involving proteins with metabolic functions, RNA-binding or chaperones-like proteins, and regulatory sRNAs controlled by σS 16, 28. Additionally, the lack of functional σS protein likely favours the accumulation of cellular damages, such as mis-folded proteins, and might reduce the efficiency of protein turnover or recycling/autophagy mechanisms, resulting in an increase in the abundance of some proteins in the ΔrpoS mutant.
Post-transcriptional regulation would allow σS to rapidly decrease expression of a subset of genes by actively destroying the mRNAs or proteins, which may be particularly relevant for genes with long mRNA or protein half-lives. The fast reversibility of post-transcriptional mechanisms would confer to the cell the ability to rapidly fit its physiology to changing conditions. Post-transcriptional regulatory mechanisms might endow σS with repressor functions for fine-tuning expression of target genes for which transcription rates are higher in the presence of σS (for example the prp, iol, and fuc genes, Figs 3 and 6). Such dual and antagonistic regulatory circuits might be well adapted to control expression of genes showing antagonistic phenotypic pleiotropy (i.e. their expression provides a fitness advantage in some environmental conditions and a fitness cost in others). They might be pivotal to favour cell survival in stationary phase and rapid outgrowth from dormancy under varying conditions. The complexity of the σS effects probably fits the cell physiology to the trade-off between cellular maintenance during the quiescence state and re-growth potential. Though challenging questions remain open, this work provides solid basis for deeper exploration of novel regulatory and physiological aspects in quiescent cells.
Bacterial strains, bacteriophage, plasmids and growth conditions
Strains and plasmids are listed in Supplementary Table S2. Bacteriophage P22HT105/1int was used to transfer mutations and lacZ fusions between Salmonella strains by transduction52. Green plates, for screening for P22-infected cells or lysogens, were prepared as described previously53. Bacteria were routinely grown in Luria-Bertani medium (LB)54 at 37 °C under aeration. M63 minimal medium55 was used to assess Salmonella growth at the expense of various carbon sources: glucose (20 mM), propionate (20 mM), L-fucose (25 mM), myo-inositol (55.5 mM), ethanolamine (25 mM) and propanediol (25 mM). For growth on propanediol and ethanolamine, the medium was supplemented with cobalamine (200 nM). For growth assays on plates, stationary phase cultures in LB were washed, resuspended in phosphate-buffered saline (PBS)54 to OD600 of 1.0, and 5 μl of serial dilutions were spotted onto plates that were incubated at 37 °C. Antibiotics were used at the following concentrations (in μg per ml): carbenicillin (Cb), 100; chloramphenicol (Cm), 15 for the chromosomal resistance gene and 30 for the plasmid resistance gene; kanamycin, (Km) 50; and tetracycline (Tet) 20.
Global proteomics analyses
Experiments were performed in biological triplicates. The wild type and ΔrpoS Salmonella strains (VF6910 and VFC331, respectively) were grown in LB at 37 °C for 18 h. Fifty ml of cultures were centrifuged and pellets were resuspended in 10 ml Tris-HCl 100 mM (pH 7.4), Urea 8 M. The cell suspension was lysed at 4 °C by a Cell disrupter (Constant System Ltd), centrifuged for 15 min at 4 °C and 4,500 rpm, and the soluble fraction was immediately freezed in liquid nitrogen and stored at −80 °C. The amount of proteins in the cell lysates was determined using the DC Protein Assay kit (Bio-Rad). Integrity of the samples was checked by SDS-PAGE. This was done by resolving 10 μg of all lysates on an SDS-polyacrylamide gel, followed by Coomassie staining. Absence of protein degradation and uniform intensity of major bands across all lysates were considered as indicators of sample integrity and accuracy of protein quantification.
Cell lysates were sonicated 2 × 1 min on ice with a Hielscher Ultrasound Technology UP200St equipped with the Vialtweeter Sonotrode (parameters of the AMPL mode: Amplitude 80%/Cycle 80%), centrifuged at 4 °C for 30 min at 14000 g. Subsequently, 50 μg of total proteins were reduced in 50 mM TCEP (Sigma − 646547) for 1 h, and alkylated in 50 mM iodoacetamide (Sigma - I114) for 1 h in dark. Proteins were digested with 1 μg rLys-C (Promega - V1671) for 3 h at 37 °C, and then with 1 μg of Sequencing Grade Modified Trypsin (Promega - V5111) for 16 h at 37 °C. The digestion was stop with 4% formic acid, and peptides were desalted on reversed phase C18 Sep-Pak Cartridge (Waters - WAT054955). Peptides were eluted with 2× Acetonitrile 50%/Formic acid 0.1% and 1× Acetonitrile 80%/Formic acid 0.1%. Finally, samples were dried in vacuum centrifuge and resuspended with Acetonitrile 2%/Formic acid 0.1%.
Online chromatography was performed with a Thermo EASY-nLC 1000 UHPLC system (Thermo Fisher Scientific, Bremen, Germany) coupled online to the Q Exactive HF instrument with a nano-electrospray ion source (Thermo Fisher Scientific). For each samples, 1 μg of peptides was injected onto a 50 cm column (EASY-Spray column, 50 cm × 75 µm ID, PepMap C18, 2 µm particles, 100 A pore size - ES803 - Thermo Fisher Scientific) and separated with a multi-step gradient from 2 to 23% acetonitrile in 135 min and 23 to 45% acetonitrile in 20 min, at a flow rate of 250 nL/min over 190 min. Column temperature was set to 50 °C. MS data were acquired using Xcalibur software, using a data-dependent top10 method with a survey scans (300–1700 m/z) at a resolution of 60,000, and a MS/MS scans (fixed first mass 100 m/z) at a resolution of 15,000. The AGC target and maximum injection time for the survey scans and the MS/MS scans were set to 3 E6, 100 ms and 1 E5, 45 ms, respectively. The isolation window was set to 1.6 m/z and normalized collision energy fixed to 28 for HCD fragmentation. We used an underfill ratio of 2.0% for an intensity threshold of 4.4 E4. Unassigned precursor ion charge states as well as 1, 8 and >8 charged states were rejected and peptide match was disable. Exclude isotopes was enabled and selected ions were dynamically excluded for 45 seconds.
Raw data were analysed using MaxQuant software version 220.127.116.116, using the Andromeda search engine57. The MS/MS spectra were searched against the Salmonella Typhimurium strain 14028 s UniProt database containing 5,369 proteins, and against the contaminant file included in MaxQuant. The digestion mode was set to trypsin, and a maximum of two missed cleavages were allowed. N-terminal acetylation and Methionine oxidation were set to variable modifications and Cysteine Carbamidomethylation as fixed modification. Identification of protein required at least one unique peptide per protein group, and every peptide were used only once in the protein identification process by the Razor protein FDR parameter. The minimum peptide length was fixed to 7 amino acids, and the required false discovery rate was set to 1% at the peptide and protein level. The main search peptide tolerances was set to 4.5 ppm and to 20 ppm for the MS/MS match tolerance. Second peptides was enabled to identify co-fragmentation events and match between runs accepted a match time window of 0.7 min for an alignment time window of 20 min. Quantification was performed using the XIC-based LFQ algorithm, with the Fast LFQ mode as described in ref. 58. Unique and razor peptides, included modified peptides, with at least 2 ratio counts were accepted for quantification.
Output protein group file was integrated into Perseus59, the companion software of MaxQuant, to perform data filtering and statistical tests. First, contaminants, reverse identifications, and proteins only identified by site were excluded from further data analysis and a categorical annotation was applied to create two sample groups according to the two types of bacterial strain in triplicate. Second, LFQ intensities were log2 transformed. A protein filtering was set for the validation process, such as a protein was integrated in the final list only if the protein was identified in at least two replicates of one sample group. Third, statistical analysis of the proteome adaptation between the two bacterial strains was performed on the 2444 filtered proteins. To this effect, we decided to analyse and compare our dataset with (SI approach) and without (AI approach) missing values. Missing values for LFQ intensities were imputed and replaced by random LFQ intensities that were drawn from a normal distribution at the low detection level (Supplementary Fig. S10). Yellow indicated imputated values in Supplementary Datasets S2 and S3. In both cases, two-sided T-tests of the log2 transformed LFQ intensities with a permutation-based FDR calculation at 5%, 1%, 0.1% and S0 = 160 were employed to determine different degrees of statistically significant proteins. This statistical process is the base of the proteomic comparison between the two bacterial strains, which is represented by the two Volcano-Plots, plotting the protein difference values against negative log10 transformed p-values of the two-sided T-test (Supplementary Figs S11 and S12). Proteins detected in less than two replicates of one strain and in at least two replicates of the other strain, were designated as “exclusive” to that latter strain. With the SI approach, “exclusive” proteins were considered as significant proteins (Supplementary Dataset S2). The SI approach yielded a list of 299 exclusives proteins (116 in the wild type strain and 183 in the ΔrpoS mutant), and three statistically significant sets of differentially abundant proteins. RpoS was found exclusively in the wild type strain, consistent with the deletion of the rpoS gene in the mutant, and was thus excluded from the final list of σS-regulated proteins (Fig. 2). For further analyses, changes in protein abundance were considered significant only when meeting the threshold of p-value 0.05 (log10 > 1.3, Supplementary Dataset S3). This yielded a final list of 806 significant proteins showing differential abundance in the wild type and ΔrpoS strains, among which 400 proteins were selected only with a 0.5% FDR, 223 were selected with 0.5 and 0.1% FDR, and 183 were selected with 0.5, 0.1 and 0.01% FDR (Fig. 2, Supplementary Dataset S3). The AI approach, used to evaluate the significance of the 299 exclusive proteins, yielded a final list of 134 statistically significant (p-value < 0.05) exclusive proteins, including RpoS itself (Fig. 2, Supplementary Dataset S3). In total, the abundance of 939 proteins (133 exclusives and 806 differentially abundant) was regulated by σS (Fig. 2, Supplementary Dataset S3).
Methods for strains construction, DNA manipulation, immunoblot analysis of proteins, enzymatic assays and sequence analyses are described in Supplementary Methods.
Data availability statement
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE61 partner repository with the dataset identifier PXD005256.
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We are very grateful to Nara Figueroa-Bossi for helpful comments about this manuscript and for the gift of strains MA7224 and MA7791. We thank all members of the LSMS Laboratory and BIM Unit for their kind support. This work was supported by the French National Research Agency (grant ANR-11-BSV3-009 to FN) and by grants from the Institut Pasteur and the Centre National de la Recherche Scientifique.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.