Proteome Remodeling in Response to Sulfur Limitation in “Candidatus Pelagibacter ubique”

Cultures were first harvested as their density reached 1 × 10⁷ cells/ml—near the maximum density of sulfur-limited cultures but well below the maximum density of control cultures. Comparing exponential-phase samples from sulfur-limited cultures (n = 3) to exponential-phase samples from control cultures (n = 4) revealed that remodeling of the transcriptome and proteome began prior to entering stationary phase. Cultures treated with a limiting concentration of DMSP showed higher levels of osmC, bhmT, ordL, metY, mmuM, and csdB mRNA transcripts, with correspondingly larger amounts of OsmC, BhmT, and MetY proteins (Table 1; Fig. 2).

Using the previously established sulfur requirement of 6.67 attomoles per “Ca. Pelagibacter ubique” cell (1), the concentration of DMSP unaccounted for by the biomass of late stationary-phase cultures was 39 nM (standard deviation [SD], ±5.5 nM) for the limited treatment and 789 nM (SD, ±31 nM) for the control treatment. The levels of messenger RNA transcripts for ccmC, SAR11_1163, and SAR11_1164 were observed to be five- to eightfold higher in the sulfur-limited stationary-phase samples (n = 5) than in the control stationary-phase samples (n = 4) (Table 2). However, mass spectrometry was unable to identify peptides for these proteins at any time point. Descriptions of these three genes and an interpretation of their observed expression patterns are given in “Integral membrane proteins,” below.

Examining the relative expression levels of mRNA and protein in samples collected from the same culture across five time points revealed no systematic correlation between mRNA transcription and protein translation, with the exception that the most highly upregulated transcripts showed better correlation with the abundances of their protein products (Fig. 3C). The correlation coefficients for the genes were stochastically distributed in the range from −0.99 (inversely correlated) to 0.99 (directly correlated), indicating that the expression values were neither random (clustered near 0) nor interdependent (clustered near −1 or 1). This observation is in agreement with the pairwise comparisons of samples (Fig. 2B), in which the collective mRNA/protein ratios did not form a linear trend.

The four mRNA transcripts that were most upregulated in sulfur-limited exponential-phase samples (n = 3) relative to their levels in control exponential-phase samples (n = 4), osmC, bhmT, metY, and mmuM, are downstream from experimentally validated S-adenosyl methionine (SAM) riboswitches (50) (Fig. 2). These genes are all preceded by SAM-V class riboswitches that inhibit the translation of mRNA into protein by occluding the ribosome binding site when the concentration of SAM is plentiful; the K_D (equilibrium dissociation constant) is 15 µM for metY and 120 µM for bhmT (50). One locus is also under the control of a SAM-II riboswitch, previously described as a regulator of methionine and cysteine metabolism in Bacillus subtilis and other Gram-positive bacteria (47, 48), which terminates transcription upstream from bhmT when intracellular SAM concentrations are sufficient; the K_D for bhmT is 1.2 µM (50). Although tandem riboswitches are not uncommon, the SAM-II–SAM-V pairing is unique to “Ca. Pelagibacter ubique” among currently sequenced organisms and relatively rare in the global ocean survey metagenomic data set (53). A prior survey of the “Ca. Pelagibacter ubique” genome identified several intergenic regions similar to known riboswitches in length, GC content, and conservation (53), two of which are located immediately upstream from genes observed to increase in abundance in response to sulfur stress, ordL and SAR11_1164. Due to the close association between SAM riboswitches, mRNA expression, and protein expression, it is apparent that functional RNAs play a central role in this organism’s response to a low-sulfur environment.

The genes regulated by SAM-sensing riboswitches in “Ca. Pelagibacter ubique” are involved in the interconversion of the organosulfur compounds MeSH, methionine, SAM, homocysteine, and S-adenosyl-homocysteine (Fig. 1). Compared to their expression levels in the control exponential-phase samples (n = 4), three genes under the control of SAM-V riboswitches, encoding MetY, MmuM, and BhmT, were more highly expressed in the sulfur-limited exponential-phase samples (n = 3) as both mRNA and protein (Fig. 2). These three enzymes function independently to produce methionine (Fig. 1).

Counterintuitively, given the expectations for a sulfur limitation response, the abundance of MetK, ProX, and OpuAC, DmdA, DmdB, and DmdC proteins remained constant across all samples. MetK hydrolyzes ATP to convert methionine to SAM, in opposition to the highly upregulated MmuM, which catalyzes the reverse reaction. ProX, OpuAC, DmdA, DmdB, and DmdC import and degrade DMSP to MeSH, the organosulfur substrate for MetY. Accordingly, one might pose the question, “why would a sulfur-limited bacterium endlessly cycle methionine to and from SAM and not allocate more resources to importing sulfur-containing DMSP?” We speculate that “Ca. Pelagibacter ubique” has adapted to exploit DMSP resources in an environment where reduced sulfur is rarely limiting. Under this model, a decrease in SAM would indicate that the cell should draw on the abundant DMSP pool to make more methionine, thereby providing MetK with the substrate needed to generate more SAM. This riboswitch-mediated response is likely a homeostatic mechanism that maintains a constant intracellular supply of SAM by redistributing sulfur between different organosulfur molecules.

The second-most-upregulated gene after osmC is bhmT, encoding betaine-homocysteine methyltransferase. This enzyme catalyzes the demethylation of glycine betaine to dimethylglycine. The methyl group is transferred to homocysteine to form methionine. Two other upregulated genes also catalyze the synthesis of methionine: MmuM transfers a methyl group from SAM onto homocysteine to make methionine and S-adenosyl homocysteine, while one of the functions of MetY is to replace the acetyl group on O-acetyl homoserine with MeSH to form methionine and acetate (Fig. 1). The channeling of multiple organosulfur compounds into methionine appears to be the central strategy for “Ca. Pelagibacter ubique” when the pool of bioavailable reduced sulfur is nearing depletion.

Though common in eukaryotes, the bhmT gene is rarely found in bacteria. Another bacterium utilizing bhmT is the actinobacterium Brevibacterium aurantiacum, which in addition to MetY and other species-specific sulfur acquisition genes, upregulates BhmT (BL2496) during sulfur limitation (37). This conserved stimulus for bhmT between phyla, together with the gene’s 34-fold mRNA upregulation in “Ca. Pelagibacter ubique,” highlights the odd absence of bhmT from all but a few bacterial genomes. Interestingly, the bhmT gene from “Ca. Pelagibacter ubique” HTCC1062 is more closely related to orthologs in Actinobacteria and Firmicutes than to bhmT in “Ca. Pelagibacter ubique” strain HTCC7211 (see Fig. S2 in the supplemental material).

In contrast to genes for methionine biosynthesis, genes dedicated to cysteine biosynthesis, such as cysK, cysE, and metC, were not upregulated in response to sulfur stress. MetY and MetC in “Ca. Pelagibacter ubique” have high sequence similarity (E values of 6e−50 and 2e−45, respectively) to Rv1079, a gene in Mycobacterium tuberculosis that has been suggested to act as a cystathionine gamma-lyase (EC 4.4.1.1) to catalyze the reversible reaction from cystathionine to cysteine (55). Therefore, we hypothesize that MetY, in addition to synthesizing methionine, may also synthesize cysteine. However, the pathway by which cystathionine is generated in “Ca. Pelagibacter ubique” remains unclear. An alternative explanation for observing no difference in expression for cysteine synthesis proteins is that these genes may be constitutively expressed and are reliant on BhmT, MetY, and MmuM to maintain an adequate supply of methionine to be converted into cysteine.

Two of the proteins encoded in the “Ca. Pelagibacter ubique” genome that are implicated in sulfur metabolism are AprB and AprA (56). These subunits form the holoenzyme AprBA, which mediates the bidirectional transfer of sulfite onto AMP to form adenosine 5′-phosphosulfate (APS). In other microorganisms, this complex is a component of the assimilatory sulfate reduction pathway through which inorganic sulfur is incorporated into organic molecules. However, the absence of other enzymes needed for this pathway (cysDNCHIJ, sat, and phsABC genes) from the “Ca. Pelagibacter ubique” genome led prior investigations to conclude that “Ca. Pelagibacter ubique” is unable to utilize the assimilatory sulfate reduction pathway and might instead rely upon AprBA to detoxify sulfite accumulating in the cytoplasm as a by-product of organic sulfur compound degradation (1, 3). In support of this hypothesis, we observed aprB and aprA mRNA transcripts to be 3.13- and 5.88-fold lower in sulfur-limited stationary-phase samples (n = 5) than in control exponential-phase samples (n = 4). Furthermore, peptides for the AprA subunit were also slightly, though significantly (P = 0.00025) different between stationary-phase conditions, with a sulfur limited/control ratio of 6:5.

The oxidoreductase ordL stands out in this study for increasing 17-fold in transcript abundance while the abundance of its protein product did not change significantly. Although changes in transcriptomes are commonly not mirrored in proteomes, a discrepancy of this magnitude is unusual and therefore suggestive of a posttranscriptional control mechanism. A conserved 117-bp UTR has been previously noted in ordL’s upstream intergenic region (53), but no function or secondary structural fold was proposed for it. Our findings suggest that this 5′ UTR structure (see Fig. S3 in the supplemental material) might play a role in regulating ordL translation.

The location of ordL immediately downstream from bhmT and osmC suggested that the biological roles of these three proteins might be related. OrdL belongs to a family of deaminating oxidoreductases that includes PuuB, an Escherichia coli enzyme that deaminates γ-l-glutamylputrescine to γ-glutamyl-γ-aminobutyraldehyde in the putrescine degradation pathway (57). However, the absence of other putrescine degradation pathway enzymes in the “Ca. Pelagibacter ubique” genome suggests that OrdL may be responsible for catalyzing a different reaction in this organism. The tertiary structure prediction program I-TASSER (58) identified structural similarities between “Ca. Pelagibacter ubique” OrdL and sarcosine oxidase motifs. This metabolic activity could allow OrdL to function analogously to BhmT, transferring methyl groups to homocysteine from an unknown donor molecule when glycine betaine is not present. We postulate that the conserved UTR upstream from OrdL is a riboswitch that binds to this unknown donor molecule, activating OrdL translation when this alternate methyl group donor is available. Alternatively, the BhmT reaction product dimethylglycine (DMG) may be a substrate for OrdL. If this interpretation of BhmT function is correct, we speculate that under sulfur-limiting conditions, DMG would be produced at a decreased rate and less OrdL would be required for metabolizing DMG; a DMG-sensing riboswitch could precisely regulate the translation of OrdL in response to fluctuations in DMG concentration and account for the discrepancy in OrdL expression observed in this study. While testing these models was beyond the scope of this study, they imply functions for a novel riboswitch that is likely to be a subject for further research.

We observed upregulation of the “Ca. Pelagibacter ubique” OsmC protein. This protein is structurally similar to OsmC in Escherichia coli, which has been described as a peroxidase that favors organic hydroperoxides but also acts on inorganic hydrogen peroxide (59). As its name implies, osmC is induced by osmotic stress in E. coli and other species (60). The genome of “Ca. Pelagibacter ubique” also encodes the peroxidase rubrerythrin (rbr), but as Rbr relies on an iron-sulfur center (61–63), OsmC may be better suited to responding to the loss of osmolytes during low-sulfur conditions. A previous study of the proteome of “Ca. Pelagibacter ubique” in natural seawater medium also observed an increase in OsmC protein as cells entered stationary phase (64), an observation that was not repeated in limitation studies using iron (65) or nitrogen (66) as the limiting nutrient. Therefore, the expression of OsmC appears to be dependent on both the limiting condition and stationary-phase remodeling.

Previous studies have consistently found that bacterial responses to sulfur limitation involve increased production of sulfur compound transporters (36–38, 40), and iron limitation in “Ca. Pelagibacter ubique” was observed to result in a 27-fold increase in the periplasmic iron-binding protein SfuC (65). Therefore, it was unexpected that proteins for transporting organosulfur compounds were not observed at a higher abundance during sulfur-limited conditions, particularly the periplasmic binding components of the glycine betaine ABC transporters (ProX and OpuAC), which also bind and transport the organosulfur molecule DMSP (67).

The insoluble integral membrane protein CcmC has been well characterized as catalyzing the transfer of heme c groups to cytochrome c (68, 69). Because heme c molecules contain two sulfur atoms, they are likely to be less abundant in the cell during sulfur-limited conditions. It appears that the fivefold-higher expression of ccmC transcripts is a mechanism to compensate for this deficiency and maintain a constant supply of this essential cofactor to cytochrome c. The absence of peptide detections for CcmC is not unusual, as proteins such as CcmC are rarely detectable with the mass spectrometry sample preparation techniques used in this study.

Two proteins of unknown function, SAR11_1163 and SAR11_1164, were transcribed at levels 6 and 11 times higher, respectively, in sulfur-limited stationary-phase samples (n = 5) than in control stationary-phase samples (n = 4) (Table 2). The proteins encoded by these transcripts were never detected by mass spectrometry, indicating that these genes may be translationally inhibited or localized to the membrane. SAR11_1164 has previously been annotated as a putative lipoprotein due to a predicted transmembrane domain (3, 70). On the chromosome, SAR11_1163 and SAR11_1164 are set apart from neighboring genes by 248 nucleotides downstream and 445 nucleotides upstream. Intergenic distances of this size are particularly conspicuous given that the median intergenic spacer in “Ca. Pelagibacter ubique” is only 3 nucleotides, and the upstream region was noted by Meyer et al. as sharing characteristics of known riboswitches (53). Searching for homologs to these genes using amino acid alignment and structural alignment techniques revealed that both genes are unique to “Ca. Pelagibacter ubique” and, therefore, may represent a previously unknown class of proteins used to relieve sulfur stress. The search also revealed that they are regulated by a novel sulfur-related riboswitch.

Hypervariable regions (HVRs) are sections of a genome having variable gene content between closely related strains, likely arising from horizontal gene transfer. These genomic loci commonly encode a high proportion of novel hypothetical proteins and proteins known to confer increased fitness in a particular environment (71). HVRs are prevalent in many microorganisms, often encoding nitrogen fixation, iron and sucrose uptake, toxin and antibiotic resistance, and other genes for adapting to specific environmental stressors (72). Studies of differences in the HVR gene contents of closely related strains of the dominant marine phototroph Prochlorococcus found that the presence of nitrogen and phosphate assimilation genes correlated well with the availability of those macronutrients in the environment (73–77). In members of the SAR11 clade, the presence of genes relating to the metabolism of phosphate (77, 78) and glucose (79) have similarly been correlated with environmental conditions.

Wilhelm et al. identified four “Ca. Pelagibacter ubique” HVRs by aligning the Global Ocean Sampling (80) metagenomic sequences to the “Ca. Pelagibacter ubique” HTCC1062 genome. Three of these HVRs (HVR1, HVR2, and HVR4) were characterized by a predominance of genes for cell surface properties, transport, and secretion (70). However, HVR3, comprised of bhmT, osmC, ordL, SAR11_1163, SAR11_1164, and seven other genes of unknown function, had a less apparent role. In this study, increased expression of the five above-named genes suggests that HVR3 is responsive to changes in the availability of sulfur.

Comparing the genome of the coastal Oregon isolate “Ca. Pelagibacter ubique” strain HTCC1062 used in this study to that of “Ca. Pelagibacter ubique” strain HTCC7211, isolated from the Sargasso Sea, revealed the extent of variability at this HVR. Absent from the Sargasso Sea strain were homologs for osmC, SAR11_1163, and SAR11_1164, encoding the peroxidase and novel proteins of unknown function. A 2009 metaproteomic study found SAR11 phosphate transporters to be the most abundant proteins in Sargasso Sea surface waters (81), consistent with the theory of phosphate-limited productivity in the North Atlantic Gyre. Differences in the commonly limiting nutrients of HTCC1062 and HTCC7211 may be reflected by the composition of their HVRs, furthering the hypothesis that osmC, SAR11_1163, and SAR11_1164 provide an advantage to “Ca. Pelagibacter ubique” strains present in the phosphate- and nitrogen-rich coastal waters of the Pacific Northwest.

To assess whether similar variability exists among the genes for the acquisition of reduced sulfur compounds among SAR11 strains, we examined two complete SAR11 genomes (HTCC1062 and HTCC7211) and five incomplete SAR11 genomes (HTCC1002, HTCC9565, HIMB5, HIMB59, and HIMB114) for genes involved in sulfur metabolism, consistent with the observations reported above that the genes for sulfur acquisition are in HVR3. We observed considerable variability between strains (Table 3). Despite the presence of AprA in many strains, there was no evidence of complete operons for assimilatory sulfate reduction within any member of this group of organisms.

Our findings demonstrate that sulfur acquisition systems in “Ca. Pelagibacter ubique” strain HTCC1062 are constitutively expressed but that enzymatic steps leading to the essential sulfur-containing amino acid methionine are regulated by a unique array of riboswitches and genes, many of which, surprisingly, are encoded in a rapidly evolving genome region. “Ca. Pelagibacter ubique” exhibits two distinct responses to sulfur limitation. The observations support the model that, during exponential phase in sulfur-limited cultures, SAM-sensing riboswitches increase the mRNA and protein expression of the genes bhmT, metY, and mmuM, which act to synthesize methionine via separate pathways (Fig. 1 and 3). As the sulfur supply becomes exhausted in sulfur-limited stationary phase, the aforementioned genes are repressed and transcripts for ccmC, SAR11_1163, and SAR11_1164 are upregulated. Additional CcmC helps maintain a steady supply of sulfur-containing heme c to cytochrome c, while the two latter genes are hypothesized to be regulated by a novel sulfur-sensing riboswitch and localize to the cytoplasmic membrane.

This study was designed to provide a better understanding of SAR11 sulfur metabolism. “Ca. Pelagibacter ubique” is a key driver of marine biogeochemistry cycles and a model for understanding the cell biology of streamlined, free-living, anucleate organisms. A number of examples of successful planktonic microorganisms that have streamlined genomes have been described recently (82, 83), focusing attention on the importance of simpler cell models to microbial ecology (84).

The loss of genes for assimilatory sulfate reduction across the SAR11 clade, an apparent example of selection for genome simplicity, implies that these cells might periodically face the environmental stress of sulfur limitation. This raises the question, “have they evolved adaptations to compensate for stress?” To reconstruct unusual rearrangements of subcellular systems found in these successful minimalists, it has been necessary to apply functional genomics approaches, which provide insight when systems do not easily fit within known schemes of metabolic and regulatory organization (66). Unexpectedly, the response we observed was partially localized to a variable genome region that is rich in SAM riboswitches, including tandem riboswitch configurations, and also includes multiple motifs that likely represent new riboswitch types of unknown function. These findings are consistent with other reports of reduced transcriptional switching and further describe the preservation of posttranslational control mechanisms in streamlined genomes (82, 85). Organosulfur requirements and sulfur limitation responses in SAR11 represent an interesting case of the tradeoffs associated with genome reduction and now seem likely to be an interesting special case of how cells use hypervariable genome regions to adapt to variation across environmental regimes.

Artificial seawater (ASW) medium was made using previously established protocols (54). Water, salts, and metals were added to 10 20-liter polycarbonate carboys as detailed in Table S1 in the supplemental material and then autoclaved for 10 h. After cooling to room temperature, the carboys were sparged with CO₂ for 20 h and brought up to 20 liters using sterile water to compensate for evaporation due to autoclaving. Vitamins and nutrients were added to each carboy from a stock solution containing either large or small amounts of DMSP and then sparged overnight on air while cooling to 16°C. The final nutrient concentrations were as follows: 10 µM glycine, 500 nM glycine betaine, 500 µM pyruvate, and 100 nM or 1 µM DMSP. The medium was then inoculated with “Candidatus Pelagibacter ubique” HTCC1062 (53) growing exponentially in low-sulfur medium. Following inoculation, the cultures were incubated at 16°C with constant air sparging.

Culture growth was tracked daily by staining cells with SYBR green and counting on a Guava EasyCyte flow cytometer (86). Samples for microarray and proteomic analysis were taken from each culture at three time points: exponential growth, exponential-to-stationary transition, and stationary phase. At these time points, 5 × 10¹⁰ cells were removed to a separate vessel and amended with 10 mg chloramphenicol, 100 µl 500 mM EDTA, and 100 µl 100× Halt protease inhibitor cocktail (catalog number 78438; Thermo Scientific) per liter of culture. Tangential flow filtration against a Pellicon 2 mini-ultrafiltration module 30-K membrane (catalog number P2C030C01; Millipore) reduced the volume of culture to less than 150 ml, which was subsequently centrifuged for 1 h at 20,000 rpm and 0°C. The pelleted cells were resuspended in 1 ml Tris-EDTA and centrifuged again in a single 1.5-ml tube for 40 min at 40,000 rpm and 10°C. After decanting the supernatant, the pellet was stored at −80°C until proteomic analysis. In parallel, 80 ml of culture was collected by centrifugation (1 h at 20,000 rpm and 0°C) for microarray analyses. Pellets were resuspended in 1 ml of RNAprotect bacterial reagent (catalog number 76506; Qiagen) and then centrifuged again in a 1.5-ml tube for 40 min at 40,000 rpm and 0°C. After decanting the supernatant, microarray samples were placed at −80°C until analysis.

Messenger RNA was processed using the same protocol described previously (66). Total RNA was extracted using an RNeasy MinElute cleanup kit (catalog number 74204; Qiagen) and then amplified using the MessageAmp II bacterial RNA amplification kit (catalog number AM1790; Ambion) with biotin-11-UTP (catalog number AM8451; Ambion) according to the MessageAmp “Improved Protocol” Handbook. Ten micrograms of amplified RNA per sample were hybridized to Pelagibacter-specific Affymetrix microarray chips.

Protein expression was measured using capillary liquid chromatography-mass spectrometry as previously described (66). Briefly, samples were sonicated and barocycled to lyse cells and then digested with trypsin. Peptide separation was performed using a high-performance liquid chromatography column in line with an LTQ (linear trap quadrupole) Orbitrap Velos mass spectrometer. High-resolution mass spectrometry spectra were collected from duplicate runs for each biological sample and matched to entries in a “Ca. Pelagibacter ubique” accurate mass and time tag database (87). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (88) partner repository with the data set accession numbers PXD003672 and 10.6019/PXD003672.

Fold changes in protein abundance were calculated as described previously (66) and are detailed in Text S1 in the supplemental material. Briefly, peptides from the same treatment and time point (defined in the next section) were averaged together and divided by the average from the exponential-phase control samples (n = 4) or other specified reference set, and then peptide fold changes were averaged together to arrive at a protein fold change.

Harvest time points were selected to provide a minimum of 2 × 10¹⁰ cells for mass spectrometry. As a result, the sulfur-limited exponential-phase samples were harvested close to stationary phase. In order to ensure that samples categorized as exponential phase are biologically accurate, the levels of expression of rpsCEGHJLNS, rplBCDEFNOPRVWX, and fusA mRNA were taken into account. These 21 genes were selected because they are nearly contiguous loci of ribosomal proteins which all decreased significantly in mRNA expression (P < 0.01) between the exponential-phase control samples (n = 4; all ≤50% maximum cell density) and stationary-phase control samples (n = 4) marked in Fig. S1 in the supplemental material. The transcript RNA expression values from the microarray chips for each gene were first log₂ transformed and then normalized to a range of 0.0 to 1.0. A sample’s “growth state” was then calculated by averaging all 21 genes’ normalized expression values in that sample. For example, a growth state of 1.0 would indicate that all 21 genes were expressed at their maximum observed abundances. The samples from this study clustered into three distinct growth states, 0.91 to 0.80, 0.58 to 0.39, and 0.29 to 0.08, which corresponded well to the exponential, transitioning, and stationary labels. This qualitative assessment reclassified two sulfur-limited exponential-phase samples as transitioning.

Microarray data collected by this study are available in the NCBI GEO database under accession number GSE31630 .