Metabolomic Characterization of the Salt Stress Response in Streptomyces coelicolor

Strains and medium conditions.Escherichia coli strain ET12567/pUZ8002 was used for conjugation of cosmids into S. coelicolor (14). Salt shock or osmoadaptation experiments using the S. coelicolor M145 parent strain (27), LW35 (ectA::Tn [where Tn is transposon 5062]) (this study), LW36 (ectC::Tn) (this study), LW37 (ectD::Tn) (this study), the osaB::Tn strain (3), and the sigB::aac(3)IV strain (ΔsigB) (30) were performed with Difco nutrient broth (DNB), whereas solid growth experiments were performed with supplemented minimal medium (SMMS) with various amounts of NaCl or ectoines added (20). For the liquid cultures, a high inoculum of spores was used to reduce the formation of clumps.

Ectoine mutant construction and characterization.Three cosmids containing Tn5062 transposon insertions in ectA (SCO1864), ectC (SCO1866), and ectD (SCO1867) were used to disrupt the ectoine biosynthesis cluster genes in S. coelicolor. Conjugation to S. coelicolor was performed as previously reported (3). Allelic replacements were screened by selecting for Apr^r and Kan^s clones. To confirm the genomic replacement of the ectA, ectC, and ectD genes with the Tn5062-containing copies, genomic DNA from each mutant was isolated and used as a template for PCR using one of the transposon primers EZR1 and EZL2 (3) and ect gene primers which anneal to the start and stop codons of ectA, ectC, and ectD (for primer sequences, see Table SB in the supplemental material). PCR conducted with wild-type strain M145 total DNA as a template gave rise to amplified products which corresponded in size with the predicted products of 633 (ectA), 1,250 (ectB), 414 (ectC), and 888 (ectD) bp by use of the ect gene primers (see Fig. SA.b, lanes 1 to 4, in the supplemental material). Due to the insertion of the large transposon, no PCR product was amplified when only the ect gene primers were used with genomic DNA from the mutants (see Fig. SA.b, lanes 7, 10, and 13). However, the transposon primers in combination with the ect gene primers yielded products corresponding in size with the predicted distance from the ect start and stop codons to the location of the transposon (see Fig. SA.b, lanes 5, 6, 8, 9, 11, and 12). These data confirmed the genomic disruption of the ectA, ectC, and ectD genes. This was further validated by a complementation analysis using a drop dilution assay. Cells were grown on SMMS with 1 M NaCl, with and without an additional 20 μM ectoine and hydroxyectoine, and monitored for growth (Fig. 2).

• Open in new tab
• Download powerpoint

FIG. 2.

Ectoine biosynthesis disruption mutants display a salt-sensitive phenotype, which is complemented by the extracellular addition of ectoines. S. coelicolor spore stocks of the M145, ectA::Tn, ectC::Tn, ectD::Tn, ΔsigB, and osaB::Tn strains were diluted so that a spot of 2.5 μl contained approximately 10⁵, 10⁴, 10³, 10², 10¹, or 10⁰ spores. Spots were plated on SMMS without any addition (A, left) or with the addition of 1 M NaCl (A, right), 1 M NaCl and 20 μM ectoine (B, left), 1 M NaCl and 20 μM hydroxyectoine (B, right), or 1 M NaCl, 20 μM ectoine, and 20 μM hydroxyectoine (C). Plates were incubated at 30°C for 3 days in the absence of salt and for 6 days in the presence of salt.

Growth conditions and metabolite extraction.Flasks containing 50 ml DNB were inoculated with 1 × 10⁸ S. coelicolor spores of the M145, ectA::Tn, ectC::Tn, ectD::Tn, osaB::Tn, or ΔsigB strain, and cultures were grown for 24 h, at which point 5 ml of 5 M NaCl or 5 ml of MilliQ water was added for salt shock samples. Cells exposed to continuous salt stress were inoculated and grown in 50 ml DNB supplemented with 0.5 M NaCl. Samples of 5 ml (optical density at 450 nm [OD₄₅₀] of 1.0) or an equivalent thereof needed to obtain an identical cell mass were harvested at 0, 4, 8, and 24 h after addition of salt for the salt shock cultures or at 24 and 48 h for the continuously salt-stressed cultures. Cells were collected by centrifugation for 10 min at 4,500 rpm and 4°C, the supernatant was removed completely, and the cell pellet was resuspended in 500 μl methanol. After incubation on ice for 10 min with occasional mixing, cells were spun down at 4°C, 13,000 rpm, and the supernatant was collected and stored at −80°C.

LC-MS.Data were acquired with a Luna 3-μm-particle-size HILIC 200A high-pressure liquid chromatograph (HPLC) (150 by 2.0 mm; Bester, Rotterdam, Netherlands) coupled to an LTQ Orbitrap XL mass spectrometer (Thermo Fisher Scientific, Bremen, Germany) in positive-ionization mode at a resolution of 30,000. The liquid chromatography-mass spectrometry (LC-MS) system was run in binary gradient mode with a flow rate of 400 μl/min; a volume of 20-μl methanol samples was injected. The solvents used were 90/10 acetonitrile with 5 mM ammonium acetate (solvent A) and 50/50 acetonitrile with 5 mM ammonium acetate (solvent B). The gradient used was set to hold 90% acetonitrile isocratic for 2.5 min, after which it was moved to 50% for 7.5 min and finally held isocratic for 2.5 min. The same conditions were maintained for the tandem MS (MS-MS) measurements. Analysis of the fragmentation patterns was done with XCalibur 2.0.5. Extracts were supplemented before injection with solvent A at a ratio of 1:10, improving the LC peak shape. Even though molecules were diluted 10-fold, the peak intensity levels of compounds of interest were not affected negatively, due to sharper peaks. Ectoine and hydroxyectoine standards (Fluka, Zwijndrecht, Netherlands) were used as controls.

Data processing.Data processing was performed using a configurable software pipeline capable of extracting and automatically matching peaks from multiple measurements into peak groups (40). Retention time alignment of the extracted mass chromatograms was done using the COW-CoDA algorithm (9). Noise resulting from the greedy peak picking approach was reduced by application of a relative standard deviation (RSD) filter set to 35% for matched replicates, as quantification is expected to be at least 20% accurate over multiple runs (41). The selected mass chromatograms were putatively identified by matching the masses progressively to those from metabolite-specific databases. First, ScoCyc (2), LIPID MAPS (12), and a contaminant database (25) were used. Unidentified peak groups were then matched to KEGG (24) and the remaining unidentified ones to METLIN (42) and the Human Metabolome Database (47). This iterative process was used in order to restrict the number of potential matches to the most likely (39). As a last step, all peak groups related to another peak group (e.g., isotopes, ion adducts, and fragments) were removed from the set based on a correlation analysis of the intensity patterns and peak shape (46).

Hierarchical clustering was performed to identify metabolites that showed similar dynamic changes after salt shock with various mutants. To achieve a robust clustering, the distance score was calculated for data with sufficient variation (RSD of >0.2) based on the Pearson correlation of the discretized slopes of the z-score-normalized time courses, rather than on the absolute values. By this approach, instead of determining whether the original signal intensities correlate, we determined the correlation of the slopes, i.e., the pattern of changes over time, and in order to reduce the influence of noise on the slopes, they were discretized to three possible values (−1, 0, or 1, respectively) depending on whether signal intensities were decreasing, stable, or increasing. The resulting hierarchical tree was consequently cut into 4 clusters. For each cluster, a quality score was calculated based on the average distance of each observation for each strain to the average profile of the cluster; this values indicates the coherence of expression profiles within the cluster (smaller values indicate better quality). For convenient visualization of the time courses for individual metabolites, the intensity values are displayed as fold changes relative to the initial time point (i.e., for each strain, the values are divided by the intensity value of the first time point for that strain).

Validation of metabolite quantification.The relative quantification accuracy of the mass spectrometry data was assessed by quantifying the S. coelicolor antibiotic undecylprodigiosins, which are visibly present in the extracts (the extracts are dark red when large amounts of antibiotics are produced). The traditional method of measuring the abundance of undecylprodigiosins by OD₅₃₃ measurement (44) was used as a second quantification method. Figure SB in the supplemental material shows the total detected signal for undecylprodigiosin (mass of 391.26236 Da for butylcycloheptylprodigiosin, metacycloprodiginine, and methyl-cyclo-decyl-prodiginine; mass of 393.27801 Da for undecylprodiginine) compared to the OD₅₃₃ values. The Pearson correlation is 0.95, indicating that the relative quantification achieved by the HILIC-Orbitrap combination is surprisingly precise, especially considering that these compounds do not bind well to the HILIC column and are flushed out with a large number of other compounds (mostly lipids), making exact quantification challenging due to ionization suppression effects (23). For the large number of more polar compounds in our metabolome screen, which are better separated by the HILIC column, we can therefore expect similarly reliable quantitation.

Metabolomics of continuous salt exposure.To validate our methodology, we first focused on a targeted metabolomics screen of the wild type and the ectoine biosynthesis mutants of S. coelicolor grown under continuous salt stress. Under these conditions, we expect very characteristic abundance profiles for ectoine, hydroxyectoine, and their precursors; these critical osmoprotectants should be absent in the mutants that show a clear growth defect under salt stress conditions (Fig. 2). This was indeed the case in our measurements, as the intensity levels that we found closely matched the expected phenotype of the knockouts (Fig. 3). (i) The early precursor metabolite l-2,4-diaminobutyrate (DABA) was found to accumulate in the ectA::Tn strain, where its conversion to N-γ-acetyl-l-2,4-diaminobutyrate (ADABA) is deficient, and was detected nowhere else. (ii) The later precursor metabolite ADABA was found to accumulate strongly in the ectC::Tn strain and, to a lesser degree, in the ectD::Tn strain. The presence of this metabolite in the ectD::Tn strain can be explained by the reversibility of its conversion to ectoine. (iii) Ectoine was found to accumulate strongly in the ectD::Tn strain, where its conversion to hydroxyectoine is deficient. Low intensity levels of ectoine were also detected for the M145 strain, suggesting that ectoine acts as an intermediate in S. coelicolor and is converted almost quantitatively to hydroxyectoine. Very low levels of the same mass and similar retention times for ectoine were also detected for the ectA::Tn and ectC::Tn strains; this could indicate the existence of an alternative, low-yield biosynthesis pathway for ectoine or the presence of an unrelated isomer of low abundance. In both cases, the detected levels were so low that we do not expect this to be of biological relevance for osmoprotection. (iv) Finally, hydroxyectoine was found to accumulate strongly in the M145 strain, indicating that it acts as the prime osmoprotectant. Retention time and tandem MS fragmentation patterns in comparison with the standards confirmed the identities of the detected compounds as ectoine and hydroxyectoine in the ectD::Tn and M145 strains, respectively (see Fig. SC and SD in the supplemental material).

• Open in new tab
• Download powerpoint

FIG. 3.

Levels of ectoine and related compounds detected under continuous salt stress reflect the ectoine mutants. The presence of DABA, ADABA, ectoine, and hydroxyectoine was determined in wild-type cells and in mutants lacking ectA, ectC, or ectD grown under continuous salt stress. DABA was detected only in the etcA::Tn strain, which is expected to be deficient in the conversion of DABA into ADABA and therefore accumulates the precursor. ADABA, another intermediate compound, accumulates in the ectC::Tn strain and to a lesser degree in the ectD::Tn strain, as expected. The etcD::Tn strain, which is deficient in the conversion of ectoine to hydroxyectoine, accumulated large amounts of ectoine, while hydroxyectoine accumulated in the M145 strain, as expected.

Metabolomics of the salt shock response. (i) Global metabolome screen.To obtain a global metabolomic characterization of the salt shock response, we measured time series for metabolite profiles of salt-shocked cells, using the same S. coelicolor parent strain (M145) and mutant strains (ectA::Tn, ectC::Tn, ectD::Tn, osaB::Tn, and ΔsigB strains). A total of 1,247 distinct peak groups were quantified and were combined into 363 peak groups of related peaks (each corresponding to a potential metabolite). Of the peak groups, 229 could be assigned putative identities based on exact mass matching (for an overview of the identifications, see Fig. 4 and Table SA in the supplemental material). This compares well to the predicted amount of metabolites based on genome annotations reported by Borodina et al. (4).

• Open in new tab
• Download powerpoint

FIG. 4.

Overview of the putatively identified metabolites detected in the global metabolome screen. A total of 229 metabolites could be assigned putative identities based on database matching. By far the largest class of detected metabolites were lipids (107 metabolites) and in particular a large number of short- and medium-chain fatty acids. The second-largest class contained amino acids (AA) and their derivatives (66 metabolites), including a large number of di- and tripeptides, many of which contained proline and glycine residues.

(ii) Dynamic responses to salt exposure.The dynamic response to salt exposure was visualized by unsupervised clustering of the metabolite levels for the time series of the salt-stressed cultures. Two coherent clusters of metabolites were revealed, one with consistently increasing abundance and one with consistently decreasing abundance (Fig. 5, clusters 1 and 4, respectively; for the detected masses, see Table SC in the supplemental material). All other clusters were much less coherent, as indicated by their quality scores. This coherence shows that the global metabolomic response to salt stress is clearly dominant in our samples and involves a reasonably large number of 52 peak groups (33 increasing and 19 decreasing). Comparison of the time trends for all putatively identified metabolites from the salt shock cultures to those for the control, non-salt-shocked cultures revealed reproducible results for 15 metabolites accumulating in response to the salt shock and 3 metabolites with the reverse response (Table 1). The remaining metabolites either did not give reproducible results or showed equal behaviors in the salt shock and control cultures. Figure 6 shows exemplary time courses with and without salt shock, including the putative molecular identities of the compounds involved. A large number of the upregulated compounds were also contained in the medium used, and further analysis is required to determine whether they are synthesized or taken up by the organism. The most important accumulating metabolite was proline, which is a well-known osmoprotectant that was previously reported for Streptomyces (11). The strong levels of accumulation of proline after the salt shock were the same for all of the strains, while in the nonshocked cultures proline levels remained stable or decreased slightly over time. The strong and immediate change indicates that proline is used for acute osmoprotection. Interestingly, this response is independent of the major regulators OsaB and σ^B. In addition to this classical salt-responsive metabolite, the global metabolome screen identified other potential osmoprotectants. Arginine, phenylalanine, methionine, tryprophan, and (iso)leucine showed the same strong accumulation as proline and thus could also play a supplementary role in acute osmoprotection. Like all amino acids, the listed compounds are zwitterions that could plausibly act as osmoprotectants. A similar dynamic, seen for guanosine, adenosine, adenine, and hypoxanthine nucleotide derivatives, is less obvious to explain. While affecting a large number of amino acids, the accumulation in response to the salt shock is not a general phenomenon for all amino acids, as can be seen in the time courses for glutamate and valine.

• Open in new tab
• Download powerpoint

FIG. 5.

Clustering analysis uncovers the salt stress response. Clustering analysis has been used to group metabolites showing similar temporal behaviors into clusters. For each of the 4 clusters, the mean behaviors are shown, with error bars depicting the standard deviations. Cluster 1 contains compounds that accumulate after the salt shock, including the well-known osmoprotectant proline. Cluster 4 contains compounds that decrease after the salt shock. Interestingly, these clusters are the most homogenous of all clusters, as indicated by the quality score, showing that the salt-responsive metabolites react as a single group.

• Open in new tab
• Download powerpoint

FIG. 6.

Comparison of time trends for putatively identified salt stress-responsive metabolites. The level of the major osmoprotectant proline, as well as those of several other putatively identified amino acids, shows an increase in response to salt shock. The detected proline/glycine-containing di- and tripeptides also accumulate in response to salt shock (ArgGlyPro is shown as a representative example), as do a few nucleotide-related compounds, including adenosine. 5′-Methylthioadenosine is one of the few compounds reproducibly showing the opposite pattern, an accumulation under low-salt conditions only. Levels of histidine and its catabolite urocanate decrease in all of the cultures, showing no response to the salt shock. For an overview of all 20 reproducible metabolites, see Fig. SE in the supplemental material. Results are shown for samples taken at 0, 4, 8, and 24 h after salt shock.

In addition to the amino acids, we detected five di- and tripeptides that accumulated during salt stress (Fig. 4). All of these peptides contained a proline and a glycine residue. These proline/glycine-containing peptides were significantly enriched compared to the protein compositions for both S. coelicolor and Saccharomyces cerevisiae (P value of <0.01, based on random sampling of peptide sequences). Glycine, glutamine, and alanine individually were observed at similar significance levels, while other amino acids were represented at the same frequencies in proteins and the accumulating peptides. This indicates that proline/glycine-containing peptides are either produced by specific proteolysis or taken up preferentially from the peptone-containing medium. A possible candidate for a di- and tripeptide uptake system is SCO3064, which has homology with DtpT from Lactococcus lactis (18). This is in agreement with earlier studies of other organisms that also showed an involvement of proline-containing peptides in the salt stress response (1, 32, 37).

One of the few putatively identified metabolites that showed a slight but consistent decrease in abundance during salt shock but not under control conditions was 5′-methylthioadenosine. This metabolite acts as an inhibitor of polyamine biosynthesis in vitro and in vivo (19, 36). As polyamines are known to protect against salt stress in plants (48), active degradation of the inhibitor 5′-methylthioadenosine could be part of the salt response in Streptomyces as well. The expression of the pfs gene, encoding 5′-methylthioadenosine/S-adenosylhomocysteine nucleosidase (10), is increased by salt stress in E. coli (29). Alternatively, the decreased levels of 5′-methylthioadenosine could indicate a decreased activity of the polyamine biosynthesis pathway under salt stress. Distinguishing these two alternatives requires targeted measurements of polyamines, which were not detected in our experimental setup.

Other metabolites that decreased in abundance under high-salt conditions also did so under control conditions. For instance, the amino acid histidine and its catabolite urocanate decreased in both groups of cultures; these were the only amino acid-related compounds which showed this behavior.

Not surprisingly, ectoine and hydroxyectoine were absent from the core salt shock response cluster, due to the disruption of the ectoine pathway in several of the strains. The detected compounds are thus those that show a consistent immediate response to salt shock independent of genotype, indicating that the salt shock metabolome extends far beyond the ectoines. The only putative osmoprotectant that we detected but which showed no response to salt stress was alanine, which was previously described as an osmoprotectant for Streptomyces (11). The time series for all strains showed stable behavior of this metabolite for the salt-shocked samples and the non-salt-shocked control cultures, indicating that it is in fact not used as an osmoprotectant under the conditions tested.

Conclusions.Streptomyces species are known to be able to withstand considerable levels of osmotic stress (34) and to accumulate ectoine and hydroxyectoine under salt adaptation (7). We have studied the global metabolic response of S. coelicolor to salt stress by using a high-resolution LTQ Orbitrap mass spectrometer. We did not aim for a complete coverage of the entire metabolome but rather intended to achieve an unbiased, comprehensive assessment of the moderately polar osmolytes most similar to ectoine, in particular, amino acids, nucleotides, and their derivatives. Sugars and polyols, the remaining major group of osmoprotectants (which were not detectable under the liquid chromatography-mass spectrometry conditions used), will therefore be the target of future studies using complementary analytical technologies with higher selectivity toward these compounds. It is very likely that among the described compounds, we have detected all that are osmotically relevant, as osmoprotectants need to accumulate in large amounts to be effective. However, the mass spectrometry data do not provide absolute quantification of the compounds involved. Instead, our results focus on the dynamic changes of the metabolome. This relative quantification, which is standard practice in metabolomics studies, clearly identifies the temporal trends for a large number of potentially osmoprotective compounds.

The current study leads to the putative identification of potential novel osmoprotectants in S. coelicolor. Strikingly, we did not observe a major difference between the metabolomic responses to salt stress in wild-type cultures and in the two mutants with mutations in the major osmotic stress regulatory genes osaB and sigB (31). This indicates a considerable robustness of the metabolome, compared to the much more volatile transcriptome, in agreement with recent studies of plants (15). Robustness of metabolic flux under environmental and genetic perturbations has been found previously (13, 45) and seems to be a general feature of microbial metabolic networks (43). Our findings extend this observation to the robustness of steady-state metabolite levels. This robustness could be due to redundancy in the regulatory input leading to metabolic adaptation and has obvious evolutionary benefits. Elucidating the cellular networks that enable this metabolic robustness will require additional experimentation at multiple molecular levels, including transcriptomics, proteomics, and, most importantly, the salt-dependent kinetics of the enzymes involved.

Furthermore, the strong accumulation of ectoine in the ectD::Tn strain is an indication that S. coelicolor could potentially be used as a producer of ectoine in (large-scale) bacterial milking (38), but the absolute levels of production would need to be quantified to determine whether this is an attractive option.