Genome-scale transcriptional profiling in Staphylococcus aureus: Bringing order out of chaos

Abstract

We used the Staphylococcus aureus microarray meta-database (SAMMD) to compare the transcriptional profiles defined by different experiments targeting the same phenomenon in S. aureus. We specifically examined differences associated with the accessory gene regulator (agr), the staphylococcal accessory regulator (sarA), and growth within a biofilm. We found that in all three cases, there was a striking lack of overlap between the transcriptional profiles. For instance, while all experiments focusing on biofilm formation identified hundreds of differentially-expressed genes, only one of these was common to all transcriptomes. Several factors could potentially contribute to this variability including the use of different biofilm models, different growth media, different microarray platforms and, perhaps most importantly, different strains of S. aureus. The last appeared to be particularly important in the case of the agr and sarA transcriptomes. While these results emphasize the need to introduce some degree of standardization into genome-scale, microarray-based transcriptional profiling experiments, they also demonstrate the need to consider multiple strains of S. aureus in order to avoid any strain-specific bias in the interpretation of results. Our comparisons also illustrate how identification of strain-dependent differences using SAMMD can lead to the development of specific hypotheses that can then be experimentally addressed. Based on this, we have added new features to SAMMD that allow for direct comparisons between transcriptional profiling experiments.

Introduction

Staphylococcus aureus is an important human pathogen capable of causing a diverse array of infections. The pathogenic diversity of S. aureus is due to its capacity to produce a wide range of virulence factors, all of which are under the control of complex regulatory circuits. Modulating the production of these virulence factors allows S. aureus to adapt to diverse and changing conditions within the host, and this no doubt plays a major role in its ability to cause such diverse forms of infection. Genome-scale, microarray-based transcriptional profiling is an important tool that allows the comprehensive analysis of changes that occur in response to diverse stimuli [1], and a growing number of investigators have used this tool to help dissect global regulatory pathways and the nature of the adaptive response of S. aureus.

Studies done to date have examined the differential pattern of gene expression in response to various stress conditions including acid tolerance, the stringent response, exposure to various antibiotics, and growth within a biofilm. Studies have also been done examining the impact of mutations in specific regulatory genes including agrA, sarA, saeRS, vraSR and mgrA. All together, genome-scale transcriptional profiling has been used in at least 92 independent experiments. These studies have generated a vast amount of data that is difficult to analyze in detail, particularly with respect to making comparisons between different stimuli and/or strains of S. aureus. To help address this issue, we developed the S. aureus microarray meta-database (SAMMD; http://www.bioinformatics.org/sammd/) [2]. Specifically, the SAMMD database contains the complete lists of all differentially expressed genes from all published genome-scale transcriptional profiling experiments. Where possible, all ORF designations in SAMMD are mapped to the N315 genome irrespective of the original strain used in each study. To illustrate the utility of SAMMD, and to emphasize the importance of strain-dependent differences, we used SAMMD to compare gene expression profiles from different experiments with a specific emphasis on strain-dependent differences with respect to the accessory gene regulator (agr), the staphylococcal accessory regulator (sarA), and growth in biofilm.

Materials and methods

We selected agr, sarA, and biofilm for inclusion in this analysis because several microarray experiments from different strains were done on them. We used the SAMMD database “compare” function to compare and contrast transcriptomes from different strains. We tested the contribution of the strains towards gene regulation by statistical analysis using the software JMP. In all three cases, the null hypothesis that the strains did not have any association with the gene regulation was rejected based on significant p-values (see below).

The regulatory impact of agr

Although a growing number of regulatory loci have been described, it is clear that the accessory gene regulator (agr) plays a central role in the pathogenesis of S. aureus infection. In fact, recent data suggests that a high level of agr expression accounts in large part for the hypervirulence of community-acquired, methicillin-resistant S. aureus (CA-MRSA) [3]. The agr operon itself contains 4 genes that collectively make up a classic quorum-sensing signal transduction system [4–5]. These 4 genes are transcribed as part of a polycistronic message designated RNAII. Activation of this system results in increased production of the divergently-transcribed product RNAIII. Although it includes the gene for delta toxin (hld), it is clear that RNAIII functions as a regulatory RNA [6]. Until recently, the response regulator of the agr system (AgrA) had no known function other than activation of transcription from the RNAII (P2) and RNAIII (P3) promoters [4–7], but a recent report demonstrated that AgrA also regulates production of the cytolytic phenol-soluble modulins (PSMs) in an RNAIII-independent manner [8]. Production of RNAIII modulates the production of a wide variety of genes at both the transcriptional and post-transcriptional levels [4–5]. Although there are exceptions to the rule, increased production of RNAIII generally results in increased production of extracellular virulence factors and decreased production of surface-associated virulence factors [4–9].

The first study that examined the global impact of mutating agr on gene expression focused on a strain designated RN27, which is derived from the commonly-studied NCTC8325 lineage [10]. This study found that agr modulates the expression of over 100 genes and, in general, confirmed the regulatory paradigm of agr-mediated activation of extracellular virulence factors and repression of surface-associated virulence factors. However, when the same experiment was repeated with the clinical isolate UAMS-1, only 16 genes were found to overlap in the two agr regulons [11]. Genes that were down-regulated by agr in both strains included an amino acid permease and a putative drug transporter (Table 1). Among the genes that were up-regulated by agr in both strains were the agr operon itself, including RNAIII, the gene encoding alpha-toxin (hla), and the genes involved in arginine metabolism. Importantly, several genes that are classically associated with the agr regulon were not found in common between the two strains. This includes spa, hlgB, hlgC, geh and the genes encoding a variety of other toxins.

Table 1.

Comparison of agr transcriptomes from strains RN27 and UAMS-1

Genes down-regulated by agr in both strains
SA1269	-	putative drug transporter,
SA1270	-	amino acid permease
Genes up-regulated by agr in both strains
SA0184	-	conserved hypothetical protein
SA0185	murQ	SIS domain protein
SA0187	-	transcriptional regulator, rpiR family domain protein
SA1007	-	alpha-hemolysin precursor
SA1842	agrB	accessory gene regulator B
SA1843	agrC	accessory gene regulator C
SA1844	agrA	accessory gene regulator protein A
SA2006	-	putative map protein
SA2424	-	Crp/Fnr family protein
SA2425	arcC	carbamate kinase
SA2426	arcD	arginine/ornithine antiporter
SA2427	arcB	ornithine carbamoyltransferase
SA2428	arcA	arginine deiminase
SA2463	lip	triacylglycerol lipase precursor

The contribution of the strains towards gene regulation by agr was statistically tested using LR Chi-Squared tests and the null hypothesis that the strains did not have any association with the gene regulation was rejected based on a significant p-value of less than 10⁻³⁴.

Although the regulatory events controlling expression of many of these genes has not been examined in detail, spa is an exception in that it is often taken as a “classic” surface protein, production of which is dramatically increased in an agr mutant [4]. We cite this specific example to emphasize how SAMMD can be used to identify such discrepancies and thereby point investigators toward relevant experiments that may be useful in resolving them in a coherent and meaningful way. For example, in the currently-accepted S. aureus regulatory paradigm, the impact of agr on expression of spa is at least partially an indirect effect in that agr represses expression of sarT, which induces expression of sarS. Increased production of SarS then induces increased expression of spa [12]. In this context, mutation of agr would result in increased expression of sarT, which would in turn result in increased expression of sarS and consequently spa. However, sarT is absent in many clinical isolates including UAMS-1 [13]. Whether this accounts for or even contributes to the observed discrepancy with respect to agr and spa expression remains unknown. Indeed, by comparison to the 8325-4 strain RN6390, sarS is expressed at greatly increased levels in UAMS-1 despite the presence of sarT in the former and its absence in the latter [11–13]. This clearly implies that an additional regulatory factor is involved, one possibility being tcaR, which is known to modulate the expression of sarS and, while conserved in both strains, is nonfunctional in all 8325-derived strains including RN27 [14]. In fact, recent evidence suggests that sarS may play an important role with respect to other strain-dependent regulatory differences associated with the staphylococcal accessory regulator (sarA) [15–16].

The regulatory impact of sarA

The staphylococcal accessory regulator (sarA) is a second global regulator that modulates the production of multiple virulence factors. The effector molecule of the sarA regulatory system is a 15 kDa DNA-binding protein (SarA). Because mutation of sarA results in increased production of some virulence factors and decreased production of others, it is widely presumed that SarA can act as both a repressor and an activator [17–18]. However, other reports have suggested that SarA functions primarily as a repressor, with the increased production of certain virulence factors observed in sarA mutants being an indirect effect [19–20]. Similarly, it is widely assumed that SarA functions primarily at the transcriptional level [21], but recent data suggests that sarA may also have post-transcriptional regulatory effects associated with stabilization of mRNA transcripts [22].

To date, transcriptional profiling experiments have been done with sarA mutants generated in the S. aureus strains RN27, COL, and UAMS-1. Those done with RN27 and UAMS-1 have been published [10–11], while the COL experiments are currently limited to inclusion in SAMMD. The growth medium used in these experiments was similar but not identical, with brain heart infusion (BHI) broth being employed with RN27 and tryptic soy broth (TSB) being employed with COL and UAMS-1. Results from all three experiments were validated by real-time quantitative PCR (RTqPCR). Based on analysis of RNA samples prepared from the post-exponential growth phase alone, more than 100 genes were identified in each experiment. However, only four of these were common to all three sarA regulons. Expression of all four of these genes was increased in a sarA mutant. Three of these genes encode extracellular proteases (SspC, SspB, and Aur), while the 4^th encodes a threonine dehydratase (Table 2). No common genes were expressed at decreased levels in a sarA mutant. Overall, this is consistent with the hypothesis that SarA functions primarily as a repressor. Pairwise comparison of the sarA transcriptomes shows that the RN27 and COL share the highest overlap (18 genes), followed by COL and UAMS-1 (13 genes) and finally UAMS-1 and RN27 (7 genes). The pairwise comparison reflects the close relatedness of strains RN27 and COL by comparison to UAMS-1 [13]. More importantly, several genes that are classically associated with the sarA regulon are conspicuously absent. For instance, AgrA is found in the RN27 and COL transcriptomes but not in UAMS-1. This is consistent with targeted mutagenesis experiments demonstrating that sarA activates agr transcription in the 8325-4 strain RN6390 [23] but plays a limited role in that regard in UAMS-1 [11]. Other genes that did not appear in the transcriptomes of all three strains were hla, spa, fnbA, sspA, sarT, sarS, sarU, sarV, and the rest of the genes in the agr operon (agrB, agrC and agrD).

Table 2.

Comparison of sarA transcriptomes in strains UAMS-1, RN27, and COL

Genes down-regulated by sarA in all three strains
SA0899	sspC	cysteine protease
SA0900	sspB	cysteine protease precursor
SA2430	aur	zinc metalloproteinase aureolysin
SACOL1477	ilvA1	threonine dehydratase, catabolic
Genes down-regulated by sarA only in strains UAMS-1 and RN27
SA0746		staphylococcal nuclease
Genes down-regulated by sarA only in strains RN27 and COL
SA0107	spa	Immunoglobulin G binding protein A precursor
SA0271	-	Unknown
SA0923	purM	phosphoribosylformylglycinamidine cyclo-ligase
SA2424	-	Crp/Fnr family protein
Genes down-regualted by sarA only in strains UAMS and COL
SA0091	plc	1-phosphatidylinositol phosphodiesterase precursor
SA0136	-	phosphonates ABC transporter, permease protein CC0363
SA0901	sspA	glutamyl endopeptidase precursor
SA0904	-	transcriptional regulator, MarR family family
SA1269	-	drug transporter, putative
SA1270	-	amino acid permease
SA1272	-	alanine dehydrogenase
SA1725	-	Staphopain, Cysteine Proteinase
SAR1022	sspA	V8 Protease
Gene upregulated only in RN27 and COL
SA0174	-	4-phosphopantetheinyl transferase superfamily family
SA1844	agrA	accessory gene regulator protein A
SA2319	-	L-serine dehydratase, iron-sulfur-dependent, beta subunit
SA2447	-	LPXTG-motif cell wall anchor domain protein
SACOL1187	-	antibacterial protein (phenol soluble modulin)
SAS016	-	conserved hypothetical protein

The contribution of the strains towards gene regulation by sarA was statistically tested using LR Chi-Squared tests (Whole Model Tests) and the null hypothesis that the strains did not have any association with the gene regulation was rejected based on the significant p-value 0.0004.

The preceding discussion with respect to the agr regulon in different strains makes it obvious why certain of these genes were not identified in all three sarA regulons (e.g. the absence of sarT and sarU in the UAMS-1 genome). In other cases, the reasons are harder to define. For instance, it is clear that mutation of sarA results in a reduced capacity to bind fibronectin in all strains studied to date [20], and because in vitro assays indicate that SarA can bind cis elements associated with fnbA, it has been proposed that this is a direct regulatory effect associated with SarA acting as an activator of fnbA transcription [24]. However, other studies have concluded that the impact of SarA on the ability to bind fibronectin is an indirect effect associated with the increased production of proteases in sarA mutants [20–25]. In the case of hla, mutation of sarA results in decreased transcription in RN6390 but increased transcription in UAMS-1 [10–11-20–26], and it was recently proposed that this disparate effect may also revolve around the tcaR mutation in 8325-4 strains like RN6390 and its impact on expression of sarS [27].

Gene expression patterns in S. aureus biofilms

When assessing the impact of specific regulatory genes on transcriptional profile, the primary variable is strain choice. However, it is certainly not the only consideration in gene profiling experiments. For example, three independent experiments have examined the differential pattern of gene expression in S. aureus biofilms [28–30], but a comparison of these experiments using SAMMD reveals an almost complete lack of overlap between the regulons defined by these studies. In addition to employing different strains, all three experiments also employed different growth media and different growth stages at which RNA was harvested for analysis. Additional differences existed between the microarray platforms themselves (Agilent vs. Affymetrix vs. TIGR-PFGRC spotted arrays) and, to a limited extent, the degree to which expression of a gene was altered before being identified as “differentially expressed” (1.5 to 2.5-fold). None of the various growth conditions or experimental approaches employed in these experiments can be taken as more definitive than another, and in fact the discrepancies may only reflect differences between the gene expression pattern in newly formed vs. mature biofilms, but such comparisons nevertheless make it obvious that determining whether or not that is the case would be facilitated by some degree of standardization.

At the same time, even absent that standardization, useful information can be obtained using comparisons made with SAMMD. For instance, a comparison of the single stage from each experiment that would arguably represent the most mature form of a biofilm analyzed in that experiment ([28], 7 day biofilm vs. stationary phase planktonic culture; [30], 48 hr biofilms vs. 48 hr planktonic growth; [29], “mature” biofilms vs. planktonic growth) reveals that only one gene is present in all three regulons (Table 3). This gene is arcB (SA2427), which is part of the arginine deiminase operon. However, eight genes were present in two of three regulons, and included among these eight genes were SA2424, SA2425 (arcC), SA2426 (arcD) and SA2428 (arcA). Taken together, this provides strong support for the hypothesis that the arginine deiminase operon warrants further investigation in the specific context of an S. aureus biofilm. While some studies have subsequently concluded that the arc operon plays a limited role in biofilm formation [31], others have confirmed that one of the characteristics that define CA-MRSA is the presence of a second copy of the arc operon within the arginine catabolite mobile element (ACME). In fact, it has been proposed that the presence of ACME enhances the growth and overall “fitness” of CA-MRSA isolates including clones of the USA300 lineage [32]. How this will ultimately play out remains to be determined, but the more important point in the context of this report is to illustrate how comparisons made using SAMMD can be used to formulate questions that need to be asked and define the best context in which to pursue the answer.

Table 3.

Comparison of biofilm transcriptomes in strains MRSA-M2, SA113, and UAMS-1

Genes up-regulated in biofilm formed by strains MRSA-M2 and SA113 while down-regulated in UAMS-1
SA2204	-	phosphoglycerate mutase
Genes up-regulated in biofilm formed by all three strains
SA2427	arcB	ornithine carbamoyltransferase
Genes up-regulated in biofilm formed by strains SA113 and UAMS-1
SA2007	-	alpha-acetolactate decarboxylase
SA2008	alsS	acetolactate synthase
SA2086	ureF	urease accessory protein UreF
SA2088	ureD	urease accessory protein UreD
SA2424	-	Crp/Fnr family protein
SA2425	arcC	carbamate kinase
SA2426	arcD	arginine/ornithine antiporter
SA2428	arcA	arginine deiminase
Gene down-regulated in biofilm formed by strains MRSA-M2 and UAMS-1
SA1382	sodA	superoxide dismutase, Mn
Gene down-regulated in biofilm formed by strains SA113 and UAMS-1
SA0916	-	phosphoribosylaminoimidazole carboxylase, catalytic subunit
Gene up-regulated in MRSA-M2 biofilm while down-regulated in UAMS-1
SA1150	glnA	glutamine synthetase, type I
Genes up-regulated in SA113 biofilm while down-regulated in UAMS-1
SA0572	-	hydrolase, alpha/beta fold family
SA1531	ald	alanine dehydrogenase
SA2203	-	drug resistance transporter, EmrB/QacA subfamily

The contribution of the strains towards gene regulation in biofilm was statistically tested using LR Chi-Squared tests (Whole Model Tests) and the null hypothesis that the strains did not have any association with the gene regulation was rejected based on a significant p-value of less than 10⁻⁴¹.

Conclusion

This report highlights the fact that S. aureus strains are profoundly different from each other and serves as a cautionary tale for drawing sweeping conclusions about regulatory pathways in S. aureus. This is a particularly important consideration in light of the fact that current gene regulation paradigms in S. aureus are based almost exclusively on studies done with derivatives of NCTC8325, all of which are known to carry mutations in at least two important regulatory loci [14-33-37]. However, we also illustrate how SAMMD can be used to sort through the growing body of transcriptome data to identify key questions that warrant further investigation and perhaps even point the way toward specific approaches that can be used to address those questions. To facilitate this approach, we have added a “comparison” feature that allows SAMMD users to make direct comparisons between up to three different transcriptional profiling experiments. We have also added a feature that would allow researchers to examine transcriptional data collectively or by specific strains. It is hoped that this will not only allow investigators to identify relevant discrepancies in a manner that will facilitate their resolution but also to identify those features that are consistent across all strains and perhaps even all growth conditions. The latter is important in that the utility of any therapeutic agent would be greatest if it were effective against as many strains and in as many clinical situations as possible.