Reading between the Lines: Utilizing RNA-Seq Data for Global Analysis of sRNAs in Staphylococcus aureus

Regulatory small RNAs (sRNAs) are a class of RNA molecules that are produced in bacterial cells but that typically do not encode proteins. Instead, they perform a variety of critical functions within the cell as RNA. Most bacterial genomes do not include annotations for sRNA genes, and any type of analysis that is performed using a bacterial genome as a reference will therefore overlook data for sRNAs. In this study, we reexamined hundreds of previously generated S. aureus RNA-Seq data sets and reanalyzed them to generate data for sRNAs. To do so, we utilized an updated S. aureus genome annotation file, previously generated by our group, which contains annotations for 303 sRNAs. The data generated (which were previously discarded) shed new light on sRNAs in S. aureus, most of which are unstudied, and highlight certain sRNAs that are likely to play important roles in the cell.

I n bacterial cells, the best-studied and most abundant RNA molecules (rRNA, tRNA, and mRNA) play prominent roles in protein synthesis. Members of a less-well-studied class of RNA molecules that perform a variety of functions in the cell are collectively referred to as small RNAs (sRNAs). Many of these sRNAs are regulatory RNAs that influence gene expression by interacting with mRNA transcripts and/or proteins (for a review of bacterial sRNAs, see reference 1). Although sRNAs play important roles in the cell, their study has lagged behind those of protein coding genes for a variety of reasons. One reason is that sRNA genes are difficult to identify by sequence analysis alone. Another factor confounding the study of sRNAs is the fact that annotations for sRNA genes are typically absent from genome annotation files. This has led to repeat identifications of sRNAs and a general lack of awareness of the genomic location of sRNA genes (2). Furthermore, as genome annotation files are used as a reference for transcriptome sequencing (RNA-Seq) studies, the resulting analyses typically do not contain expression values for sRNAs because sRNA genes are absent from these files.
In the Gram-positive bacterium Staphylococcus aureus, hundreds of sRNA molecules have been identified (3). To facilitate further analysis of sRNAs in this important human pathogen, we previously performed a study whereby we annotated the position of known sRNAs on the S. aureus genome (2). The resulting annotation files (created in three S. aureus genetic backgrounds) are a valuable resource and have allowed us to (i) identify new sRNAs (with confidence that they had not been already identified), and (ii) analyze global sRNA gene expression by RNA-Seq (2). Recently, we further demonstrated the benefit of these annotation files by investigating the role of an sRNA, Teg41, that is in a location adjacent to the genes encoding the alpha phenol-soluble modulins (␣PSMs), potent cytolytic peptides produced by S. aureus (4). Our study revealed that ␣PSM peptide production is positively influenced by the Teg41 sRNA (4).
During our sRNA annotation study, we included annotations for 303 sRNAs on the genome of the USA300 strain FPR3757, 39 of which we identified in the previous study (2). For clarity, as outlined in the annotation study, we used the term "sRNA" to refer to any RNA that had not been previously annotated (including riboswitches, untranslated regions [UTRs], and cis-acting and trans-acting sRNAs). The rationale behind this was to provide us with a more informative map of the S. aureus genome in which the locations of key RNA features were more apparent and to generate a tool that could be used for further studies. Undoubtedly, the list of "sRNAs" that we annotated is imperfect. sRNAs that are expressed only under specific conditions (not thus far tested) are likely to be missing from the list. Furthermore, it is also likely that some of the annotations added do not represent genuine sRNAs. As mentioned above, we did not restrict our annotation to trans-acting sRNAs; therefore, many of the annotated "sRNA genes" are riboswitches, UTRs, and cis-acting elements. Furthermore, we consider it likely that many of the annotations added in fact represent protein coding genes. Consistent with this idea, we noted that certain loci where sRNA annotations were added in USA300 FPR3757 correspond to loci where coding DNA sequence (CDS) genes had been annotated in other S. aureus strains (2). These may represent protein coding genes that were omitted from the USA300 FPR3757 annotation. Consequently, while we acknowledge that the created annotation files are imperfect, they are nonetheless a valuable resource and represent a first step toward a fully annotated S. aureus genome.
As mentioned above, the absence of sRNA annotations in genome reference files has meant that all previously published RNA-Seq studies in S. aureus have failed to analyze sRNAs. Therefore, in this study, we revisited publicly available RNA-Seq data sets deposited in the Gene Expression Omnibus (GEO) database, to mine them for sRNA data. Using our annotated sRNA genome file as a reference, we examine (i) RNA-Seq transcriptomic data sets to analyze sRNA gene expression levels in a variety of strain backgrounds and under a variety of environmental conditions; (ii) RNA-Seq data from rifampin-treated cultures to analyze the rates of sRNA decay and stability; (iii) a ribosome profiling (Ribo-seq) data set to determine which of the 303 previously annotated sRNAs are bound by ribosomes and therefore potentially encode peptides; and (iv) two recently published in vivo RNA-Seq studies to search for sRNAs that are upregulated during infection.
In 2016, Mäder et al. performed a comprehensive analysis of S. aureus gene expression using tiling microarrays (5). The results generated provide an analysis of gene expression in an isogenic S. aureus background, across a variety of conditions, and represent an immensely valuable research tool. The study also identified transcriptional units that contain sRNAs, examined their expression, and compared expression to that of protein coding genes. Results from that analysis also suggest that the number of trans-acting sRNAs in S. aureus is lower than previous studies had suggested. A recent publication by Liu et al. supports with this assessment and suggests that the number of bona fide trans-acting sRNAs in S. aureus is around 50 (6).
Unlike the study by Mäder et al., the results presented here do not represent a comprehensive global transcriptomic analysis of isogenic strains across different growth media. Consequently, we have restricted our analysis to compare only RNA-Seq data sets from studies performed by the same research group from projects submitted to GEO. This ensures that all comparisons are performed in the same S. aureus genetic background. We also perform the analysis using the 303 "sRNA annotations" added during our 2016 study while acknowledging the imperfect nature of this list. To validate our approach, we confirmed the stability profiles of two sRNAs by Northern blotting. We also demonstrate, on the basis of the results from the Ribo-seq data set, that a number of previously identified sRNAs have the ability to encode short peptides/ proteins and show that overproduction of one such peptide (Tsr37) in S. aureus led to increased "clumping" of the bacterial cells (i.e., rapid autoaggregation and settling of the cells at the bottom of the culture tube), which may indicate a potential biological role. While imperfect, these results were generated using freely available, previously discarded data and can serve as a valuable guide to inform future studies and facilitate improved analysis of sRNAs in S. aureus.

RESULTS AND DISCUSSION
Overview of study. Due to the fact that sRNA annotations are missing from S. aureus genome files, we reasoned (i) that all previously published RNA-Seq data sets/experiments from S. aureus would have generated reads corresponding to sRNAs but (ii) that such data would have been discarded during the data analysis step. Consequently, a reanalysis of existing data sets performed using our sRNA annotated genome files would likely give information on the expression of sRNAs that had previously been missed. A flowchart outlining the reanalysis procedure is provided in Fig. 1. We searched the Gene Expression Omnibus (GEO) online data repository for RNA-Seq data sets originating from S. aureus (see Text S1, S2, and S3 in the supplemental material for an overview of the criteria used to search for data sets). Through this search, we identified 185 individual RNA-Seq data sets from 22 studies for further analysis (Table 1; see also Data Set S1 in the supplemental material).
The data sets identified were generated in a variety of different S. aureus backgrounds and included data determined under different growth conditions as well as from a variety of wild-type and mutant strains. To streamline the analysis and generate results that could be easily interpreted, we elected to analyze all data sets using the same reference genome (USA300 FPR3757) and to perform only direct comparisons of sRNA gene expression between data sets generated in the same study (and therefore within the same genetic background). We hypothesized that the degree of sequence similarity between sRNAs from different S. aureus backgrounds would be sufficiently high that RNA-Seq reads generated in one strain (e.g., strain Newman) would map to the corresponding sRNA gene on a different S. aureus reference genome (i.e., USA300 FPR3757). Therefore, we elected to analyze all data sets using the USA300 FPR3757 genome (to which we previously added sRNA annotations [2]). The resulting data generated represent a series of pairwise expression analyses rather than a systematic analysis of sRNA expression in a variety of strains and under a variety of conditions. We acknowledge the imperfect nature of this analysis; however, the goal was to generate an overview of sRNA expression under various conditions using already available data and the results generated are meant to serve as a guide to inform future studies.
Data acquisition and global sRNA expression analysis. From the 22 studies identified, 185 RNA-Seq data sets were downloaded from GEO and processed to calculate expression values for each of the 303 annotated sRNAs (Fig. 1). We next examined the conditions under which each experiment was performed to identify pairwise analyses that could be made using the data within each study. Only a single pairwise analysis was possible from the data within a certain study in some cases (e.g., studies 4, 10, and 11), while multiple pairwise analyses could be made in others (e.g., 13 pairwise analyses were performed using the 30 RNA-Seq data sets from study 1). In total, 64 pairwise analyses were performed from the 22 studies utilized. A complete list of the pairwise comparisons resulting from each study is provided in Table 2. For each of the 64 pairwise analyses performed, we determined (i) the expression values for each of the 303 annotated sRNAs under the two conditions being tested and (ii) the fold change in expression. The results obtained for each sRNA in each of the 64 pairwise analyses performed are provided as Data Set S2.
To obtain a broad overview of the data generated, we first examined every pairwise comparison for each of the 303 sRNAs to identify conditions in which a Ͼ3-fold change in expression was observed (highlighted in yellow in Data Set S2). Table 2 lists how many of the 303 sRNAs showed Ͼ3-fold changes in expression in each of the 64 pairwise comparisons. The largest numbers of changes (Ͼ100 of 303 sRNAs) were observed when S. aureus cells growing as a biofilm were treated with the antibiotic fosfomycin. The large degree of variation in sRNA expression is similar to the global alteration in expression observed for all S. aureus genes upon fosfomycin treatment (7,8). We next examined each of the 303 sRNAs to identify in how many of the 64 experiments a Ͼ3-fold expression change was observed for each of them. This analysis identified 11 sRNAs that showed Ͼ3-fold alterations in gene expression in 20 or more of the 64 pairwise analyses, suggesting that these are highly variably expressed sRNAs (Table 3). To our knowledge, none of these 11 variably expressed sRNAs have been studied in detail in S. aureus.
Since the vast majority of the 303 annotated sRNAs have unknown function, we posit that meta-analysis of the data generated (Data Set S2) would be less meaningful than individual analysis of the results generated for an sRNA of interest. To demonstrate this principle, we investigated the data for the sRNA Teg41. Recent work by our group demonstrated a role for Teg41 in S. aureus virulence by which it influenced production of the alpha phenol-soluble modulins (␣PSMs) (4). Although we demonstrated a role for Teg41 in ␣PSM production, expression of Teg41 itself was not examined. To gain some insight into the conditions that might influence Teg41 expression, we examined the data generated for Teg41 expression in our global sRNA expression analysis (Data Set S2). Teg41 expression was altered Ͼ3-fold in 13 of the 64 pairwise analyses ( Table 4). The greatest change in expression (25.71-fold) was identified in comparing the results seen with S. aureus grown anaerobically at 0 and 240 min. These data suggest that Teg41 expression may be increased during anaerobic growth or might have a role in metabolism. Expression of Teg41 also increased 13.65-fold when the endoribonuclease RNaseY was deleted, suggesting that RNaseY may degrade Teg41 RNA when present. These results demonstrate the utility of the data presented here for gaining an insight into the conditions under which expression of a specific sRNA might be influenced.
Global analysis of sRNA stability and degradation. RNA stability and degradation are components of an underappreciated method of genetic regulation whereby the  rate of RNA turnover can be altered in response to genetic and environmental factors (9,10). While degradation of mRNA transcripts is being increasingly studied, the stability and degradation of sRNA molecules are less so. One of the RNA-Seq studies used in the sRNA expression analysis described above (11) included a series of samples where wild-type S. aureus was treated with rifampin to arrest transcription and RNA was isolated both before the addition of rifampin (i.e., t ϭ 0) and at 2.5, 5, and 10 min after rifampin treatment. These data sets were originally used to examine the degradation and stability of mRNA transcripts; however, sRNAs were not examined. We reanalyzed these data sets, comparing the levels of stability of the 303 sRNAs by normalizing the total RNA abundances over time to that of the housekeeping gene hup, as outlined in the original study (11). We then determined the relative stability of each sRNA by taking the abundance value after 10 min (t 10 ) and dividing by the initial value measured at t 0  Sau -6745  2713518  2714477  25  SAUSA300s296  tsr32  2337921  2338072  25  SAUSA300s080  rsaOV  2696303  2696430  23  SAUSA300s078  rsaOT  2608637  2608828  22  SAUSA300s303  tsr39  2811277  2811330  22  SAUSA300s248  sRNA330  2257272  2257416  21  SAUSA300s289  tsr25  1442861  1443042  21  SAUSA300s168  Sau-76  643211  643332  20  SAUSA300s270  tsr6  120784  120897  20  SAUSA300s298  tsr34  2591031  2591131  20 (Data Set S3). Transcripts with a value greater than 0.5 were considered to be highly stable as the RNA degradation value determined for these transcripts represents no less than 50% (2-fold) of that observed for the hup transcript over the course of the experiment ( Fig. 2A). Conversely, sRNAs with t 10 /t 0 values of Ͻ0.01 (RNA decay, Ͼ99%)  (A) Highly stable sRNAs. Thirty four sRNAs were deemed highly stable based on an overall stability profile similar to that of hup. The highly stable sRNAs were either more stable than hup or their stability was 2-fold less than that of hup. FMN, flavin mononucleotide; SAM, S-adenosylmethionine. (B) Highly unstable sRNAs. A total of 23 sRNAs were deemed highly unstable, with stability 100-fold less than that of hup.
were considered to represent highly unstable RNA and were rapidly degraded (Fig. 2B). We acknowledge that the designations "highly stable" (Fig. 2A) and "highly unstable" (Fig. 2B) are somewhat arbitrary and reflective only of the conditions under which this experiment was performed (strain SA564, mid-exponential phase, Mueller-Hinton media) (11). Nonetheless, this analysis provides a valuable snapshot of the level of sRNA stability seen under the conditions tested. A total of 34 sRNAs were found to be highly stable after employing our cutoffs ( Fig. 2A). Interestingly, many of the stable sRNAs are well characterized and have known functions, such as ssrA (transfer-messenger RNA [tmRNA]), which functions in protein production; the signal recognition particle component ffs (4.5S RNA); and ssrS (6S RNA), which is a global transcriptional regulator. These results agree with work previously reported by Roberts et al. and performed in the UAMS-1 background, where tmRNA was shown to be highly stable (9). The results of the study reported by Roberts et al. also demonstrated that rnbP, the RNA component of RNase P, is highly stable, which was also observed in our analysis ( Fig. 2A). Additionally, our analysis identified a number of sRNAs involved in virulence regulation that were also highly stable such as RNAIII, Teg41, and ssr42. To experimentally validate our results for stable sRNAs, we performed Northern blotting over time on the RNAIII and Teg41 transcripts, after treatment with rifampin. Although our experimental analysis was performed in a different S. aureus background (i.e., USA300 AH1263), the results confirm that each of these transcripts was stable. The majority of S. aureus RNAs have a half-life of Ͻ2.5 min (10); however, a band was detected for RNAIII even after 10 min whereas Teg41 was found to be present through the 5-min time period and was barely detected at 10 min (Fig. 3). Twenty-three sRNAs were deemed to be "highly unstable" by our analysis (Fig. 2B). Most (if not all) of these unstable sRNAs have no known role in S. aureus.
This analysis, while limited in terms of strains and growth conditions, provides valuable insight into the stability of sRNAs in S. aureus. The results confirm the stability of RNAs that had previously been shown to be stable, validating the analytical approach. In addition, a large number of sRNAs with unknown function were shown to be highly stable. Additional investigation will be required to determine their biological activity, as their degree of stability suggests that they may play important roles.
Identification of sRNAs that potentially encode proteins/peptides. It is not uncommon for sRNAs to have dual functions in the bacterial cell. RNA can exert a regulatory function; in addition, a small open reading frame(s) (ORF[s]) found within the sRNA can encode functional peptides. Perhaps the best example of such a dualfunction sRNA in S. aureus is RNAIII (which also encodes the delta toxin). We hypothesize that additional sRNAs, included in our original sRNA annotation study (2), could represent dual-function sRNAs that also encode peptides. In addition, it is possible that Global Analysis of sRNAs by RNA-Seq some of the sRNA annotations added in our original study represent protein coding genes that were omitted from the original genome annotation but added in our study as sRNAs. To investigate this further, and due to the availability of the data, we reanalyzed the results of a ribosome profiling (Ribo-seq) study performed previously by Basu and Yap (12). Our goal was to determine the degree of ribosome binding to the 303 sRNAs in the USA300 genome. Ribosome profiling identifies mRNA transcripts that are actively being translated (and therefore are protected by a ribosome) under the conditions tested. We reasoned that any sRNAs identified at a high frequency in the Ribo-seq data set might have the potential to code for small peptides or proteins.
The Ribo-seq data were reanalyzed by mapping the ribosome protected fragments (RPFs) to the S. aureus genome (containing sRNA annotations) and calculating RPKM (reads per kilobase per million) values for each sRNA. Next, RNA-Seq transcriptomic data sets, corresponding to the Ribo-seq data sets, were analyzed and expression values were calculated for each sRNA. RPKM values from the Ribo-seq experiment were divided by expression values from the RNA-Seq analysis, resulting in an RPF/expression ratio (Data Set S4). Normalizing the Ribo-seq data in this way accounts for variations in the expression level of each sRNA (12). The majority of sRNAs analyzed (199 of the 303 or 65.68%) had an RPF/expression ratio value between 0 and 3 (Fig. 4A), indicating that they likely do not encode peptides and are true sRNAs. A total of 43 sRNAs (14.19%) had a ratio between 3 and 6, 19 (6.27%) had a ratio between 6 and 9, 11 (3.63%) had a ratio between 9 and 12, and 31 (10.23%) had a ratio greater than 12 (Fig. 4A). To investigate if this distribution is indicative of noncoding RNAs, we repeated the analysis and examined the RPF/expression ratio of all coding DNA sequences (CDS) on the S. aureus genome. Results show a much lower percentage of CDS genes with an RPF/expression ratio between 0 and 3 (26.76% compared to 65.68% for sRNA genes; Fig. 4B). Conversely, the percentage of CDS genes with an RPF/expression ratio greater than 12 was 47.45%, compared with 10.23% for sRNA genes. On the basis of these results, we consider it probable that RNAs with a RPF/expression ratio between 0 and 3 have few to no ribosomes bound and likely represent noncoding sRNAs. sRNAs with a higher RPF/expression ratio may potentially encode proteins or peptides. To further investigate the validity of the approach outlined above, we examined the RPF/expression data for RNAIII (which is known to encode a peptide) and also examined RPF/RNA-Seq read alignment data for the RNAIII locus in S. aureus. The RPF/expression ratio for RNAIII was 4.33, which was slightly lower than expected. An examination of the RPF/RNA-Seq read alignment data for the RNAIII locus shows a clear increase in the number of reads mapping to the hld locus in the RPF data set compared to the RNA-Seq data set (Fig. 5). The lower-than-expected RPF/expression ratio observed for RNAIII might have resulted from the fact that the hld coding sequence comprises a small portion of the entire RNA molecule (135 of 551 nucleotides, 24.5%). It also implies that any RNA with an RPF/expression ratio of Ͼ3 may potentially encode a protein/peptide. Taken together, the results from our reanalysis of the ribosome profiling data set indicate that the absolute RPF/expression ratio does not automatically pinpoint translated and untranslated RNAs; however, they do indicate that a RPF/expression ratio value below 3 is suggestive of noncoding transcripts whereas values above 3 may indicate the potential to encode proteins.
Confirmation that specific tsr genes encode peptides. Previous work from our group identified 39 novel sRNAs (named tsr1 to tsr39) in S. aureus strain USA300 FPR (2). In annotating these sRNAs on the S. aureus genome, we noted that the chromosomal loci corresponding to tsr9, tsr17, tsr18, tsr21, tsr22, tsr35, and tsr37 were annotated as protein coding genes in S. aureus strains MRSA252 and NCTC 8325 (2). Consequently, it is possible that these genes encode proteins in addition to/instead of being sRNAs. In examining the ribosome profiling data, we noticed that a large number of tsr sRNAs had an RPF/expression ratio above 3, including six of the seven tsrs listed above. On the basis of this result, we hypothesized that tsr9, tsr17, tsr18, tsr21, tsr22, tsr35, and tsr37 encode proteins/peptides.
To experimentally investigate this hypothesis, we first examined the sequence of each tsr for potential open reading frames (ORFs). ORFs identical to the genes annotated at the corresponding loci in MRSA252 and NCTC 8325 were identified in each of the trs genes. We next cloned each tsr gene along with its native promoter into the high-copy-number pMK4 plasmid and tagged the predicted ORF on the C terminus with a 6ϫ histidine tag. The tsr21 locus contained two unannotated potential ORFs (in addition to one annotated ORF); therefore, we individually tagged the previously unannotated ORFs and named them tsr21B and tsr21C, bringing the total number of tagged potential tsr peptides to eight. The overexpression plasmids were transformed into wild-type S. aureus (USA300 AH1263) and the resulting strains grown for 4 h. Cultures were centrifuged to collect bacterial cells, and protein samples were prepared from the cytoplasmic, cell envelope, and secreted fractions. Western blot analysis was performed using an antihistidine antibody to identify putative small proteins. Bands corresponding in size to the predicted peptides encoded by tsr21C, tsr22, and tsr37 were identified in the cell envelope fraction (Fig. 6A and B). No peptides were detected for tsr9, tsr17, tsr18, tsr21B, or tsr35 in any of the protein fractions (for tsr17, we acknowledge that it is unlikely that a peptide of 0.5 kDa would be detected on this gel). An analysis of the predicted amino acid sequence of the tsr21C, tsr22, and tsr37 peptides shows that all three peptides have predicted transmembrane helices (Fig. 6C). Collectively, these data suggest that the genes coding for tsr21C, tsr22, and tsr37 may actually code for small peptides that are located in the bacterial cell membrane. The Western blot analysis was performed under only one set of growth conditions (i.e., in the exponential phase in tryptic soy broth [TSB]), and so the failure to detect tsr9, tsr17, tsr18, tsr21B, or tsr35 in any of the protein fractions may simply indicate that these proteins were not expressed under the conditions tested.
Interestingly, in growing the strains outlined above for the Western blot analysis, we observed that S. aureus cultures overexpressing histidine-tagged tsr37 rapidly autoaggregated and settled to the bottom of the culture tube. For simplicity, we refer to this process as "clumping," but note that no agglutinating agent (such as fibrinogen or plasma) had been added to these cultures. To quantify the results, an aggregation assay (described previously [13]) was performed to compare the degree of clumping seen with the tsr37 overexpressing strain to that seen with a pMK4 empty vector control. Exponentially growing cells were washed and resuspended in 1 ml of phosphatebuffered saline (PBS), and an initial measurement of the optical density at 600 nm (OD 600 ) of the sample was performed. After 2 h, 100 l was withdrawn from the top of each sample and the OD 600 measurement performed again. Percent clumping of each strain was then determined. A 10-fold increase in clumping was observed in the strain overexpressing the His-tagged tsr37 peptide (79.2% clumping at 2 h) compared to empty vector controls (7.9% clumping at 2 h) (Fig. 7). These data suggest (i) that tsr37 encodes a novel small peptide and (ii) that tsr37 has a biologically meaningful role in the cell. While it is tempting to speculate that the clumping phenotype observed upon tsr37 overexpression was due to the activity of the tsr37-encoded peptide, it remains possible that tsr37 functions exclusively as an RNA molecule and that the clumping phenotype is mediated via the tsr37 sRNA. Work is ongoing in our lab to investigate the nature and biological activity of tsr37.
Analysis of in vivo RNA-Seq data sets. Recent advances in RNA-Seq technology, including reductions in cost, the ability to generate more reads, and the ability to generate data from small amounts of input RNA, have led to the application of this method in vivo to perform transcriptomic analysis of bacteria during the infectious process. Recently, two studies have utilized RNA-Seq to examine the transcriptional response of S. aureus in vivo. Work by Ibberson and Whiteley examined the S. aureus transcriptome in the lung of patients with cystic fibrosis (CF) (14), while Deng et al. examined the S. aureus transcriptome during vaginal colonization (15). The RNA-Seq data sets for these two studies were not yet available at the time that we performed the original search of the GEO database (for our global sRNA expression analysis described above). We hypothesized that a reanalysis of these data sets could potentially identify sRNAs with altered expression in vivo (and, consequently, sRNAs that may play a role in vivo); therefore, we analyzed both data sets with our updated reference genome file. For analysis of the CF data set, we compared the transcriptomes of the nine samples isolated from CF patients with transcriptomes from nine S. aureus cultures grown in chemically defined media (CDM) and from nine S. aureus cultures grown in synthetic cystic fibrosis media (SCFM). We examined the data for sRNAs that were upregulated a FIG 7 Quantitative analysis of clumping phenotype observed in tsr37 overexpressor strain. A clumping assay was performed using S. aureus strain AH1263 overexpressing tsr37 (pMK4 tsr37-his), and results were compared to those seen with an empty vector control (pMK4 empty vector). The proportion of clumping of the empty vector control after 2 h was 7.9% compared to 79.2% for the tsr37-overexpressing strain.
minimum of 2-fold in the CF lung compared to both CDM and SCFM and found that a total of 122 sRNAs met our cutoffs ( Fig. 8; see also Data Set S5). Similarly, for the vaginal persistence data set we compared S. aureus growing in laboratory media to samples isolated from the vagina of mice at 5 h, 1 day, and 3 days postinoculation. We utilized multiple comparisons to determine which sRNAs were upregulated a minimum of 2-fold at each of the three time points in the vagina. These cutoffs led to the identification of 60 sRNAs that were consistently upregulated during vaginal persistence ( Fig. 8; see also Data Set S6). Finally, we cross-referenced these analyses to identify sRNAs that were upregulated in both studies (and that therefore may be broadly important for S. aureus in vivo). A total of 29 sRNAs were identified as upregulated in both the CF lung and the vagina compared to the corresponding in vitro conditions (Fig. 8B). These sRNAs are excellent candidates for future studies as they may play important roles during in vivo colonization and/or virulence. Interestingly, very few (if any) of the 29 sRNAs identified have been studied in detail or have any known role in S. aureus. With that in mind, we compiled all of the data generated in this study for these 29 in vivo-expressed sRNAs, including their relative stabilities, RPF/expression ratios, and numbers of pairwise analyses where they demonstrated Ͼ3-fold variation in expression (Fig. 8B). Of particular interest, we observed that both ssr128 and RsaG were highly stable and had high RPF/expression ratios (7.9 and 8.7, respectively), suggesting the potential for these sRNAs to encode proteins. Indeed, when we examined the DNA sequence of ssr128 and RsaG, we observed ORFs that could potentially encode proteins of 56 and 25 amino acids, respectively. While not conclusive, these data suggest that ssr128 and RsaG warrant further investigation and that each could function as an sRNA and/or protein coding transcript.
Conclusions and limitations. The variety of ways in which RNA-Seq can be used is rapidly expanding, and the versatility of the technique is reflected in the variety of analyses presented here. We examined variation in expression of sRNAs under different environmental/growth conditions and in different genetic backgrounds and then went on to analyze sRNA stability/degradation and binding of ribosomes. Finally, we reexamined two recently published studies that used RNA-Seq to examine bacterial gene expression in vivo and identified sRNAs that are upregulated in each niche. The results generated represent a useful guide for future studies, but we also acknowledge the limitations regarding their interpretation. In certain cases, the experimental data analyzed arose from a single set of growth conditions (e.g., the RNA stability and Ribo-seq data sets). The stabilities of sRNAs and their potentials to bind ribosomes and to be translated might, and most likely do, differ under differing growth conditions (9)(10)(11). In addition, while the USA300 FPR3757 genome file was used as a reference throughout, the RNA-Seq data sets used in this study were not all generated in the same strain of S. aureus. We acknowledge the limitation that this represents; however, comparative analyses were performed only on data sets from the same study (and thus the same strain); therefore, any error introduced (due to differences in the experimental strain and reference genome) should be consistent throughout.

MATERIALS AND METHODS
Bacterial strains and plasmids. The bacterial strains and plasmids used in this study are listed in Table 5. S. aureus and E. coli strains were grown routinely at 37°C with shaking in tryptic soy broth (TSB) and lysogeny broth (LB), respectively. When necessary, antibiotics were added at the following concentrations: chloramphenicol at 10 g/ml and ampicillin at 100 g/ml.
Bioinformatics and sRNA expression analysis. All data sets were downloaded directly from GEO. RNA-Seq data analysis was performed using the CLC Genomics Workbench software platform (Qiagen). The genome file containing sRNA annotations for S. aureus USA300 was previously published by our group and is available online (2). RNA-Seq data analysis was carried out using a protocol previously published by our research group (16). The resulting data were exported, and only the data for sRNAs were further analyzed.
RNA stability analysis and normalization. To evaluate RNA stability, data were downloaded from GEO (accession number SRP058249) (11). Results from the RNA-Seq analysis were used for a multigroup, unpaired comparison to generate expression values in RPKM. Mean expression values were determined for each sRNA at times 0 min, 2.5 min, 5 min, and 10 min after rifampin treatment. The mean expression value for each sRNA was then normalized to the expression value of hup at the same time point in order 4°C. The membrane was washed three times with PBST for 10 min each time. The membrane was incubated with 1:10,000 of a goat anti-rabbit fluorescent antibody for 1 h, with rocking, at room temperature. The membrane was washed three times with PBST and imaged using an Odyssey CLx imager (Li-Cor).
Analysis of in vivo data sets. In vivo RNA-Seq data sets were acquired from GEO under accession number SRP222773 for CF data and accession number SRP229518 for vaginal data (14,15). For the CF lung data, sRNA expression values (in RPKM) from nine clinical isolate samples were averaged and compared to the averaged sRNA expression values from nine S. aureus cultures grown in chemically defined media (CDM) as well as to those from nine samples from synthetic cystic fibrosis media (SCFM). A 2-fold cutoff value was employed to identify sRNAs with increased expression in the CF lung compared to both media. For the vaginal data, results from S. aureus growing in laboratory media were compared to results from samples isolated from the vagina of mice at 5 h, 1 day, and 3 days postinoculation. A 2-fold cutoff value was used to identify sRNAs with consistently increased expression in the vagina.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. TEXT S1, DOCX file, 0.