Proteomic and Transcriptomic Profiling of Staphylococcus aureus Surface LPXTG-proteins: Correlation with agr Genotypes and Adherence Phenotypes*

Staphylococcus aureus infections involve numerous adhesins and toxins, which expression depends on complex regulatory networks. Adhesins include a family of surface proteins covalently attached to the peptidoglycan via a conserved LPXTG motif. Here we determined the protein and mRNA expression of LPXTG-proteins of S. aureus Newman in time-course experiments, and their relation to fibrinogen adherence in vitro. Experiments were performed with mutants in the global accessory-gene regulator (agr), surface protein A (Spa), and fibrinogen-binding protein A (ClfA), as well as during growth in iron-rich or iron-poor media. Surface proteins were recovered by trypsin-shaving of live bacteria. Released peptides were analyzed by liquid chromatography coupled to tandem mass-spectrometry. To unambiguously identify peptides unique to LPXTG-proteins, the analytical conditions were refined using a reference library of S. aureus LPXTG-proteins heterogeneously expressed in surrogate Lactococcus lactis. Transcriptomes were determined by microarrays. Sixteen of the 18 LPXTG-proteins present in S. aureus Newman were detected by proteomics. Nine LPXTG-proteins showed a bell-shape agr-like expression that was abrogated in agr-negative mutants including Spa, fibronectin-binding protein A (FnBPA), ClfA, iron-binding IsdA, and IsdB, immunomodulator SasH, functionally uncharacterized SasD, biofilm-related SasG and methicillin resistance-related FmtB. However, only Spa and SasH modified their proteomic and mRNA profiles in parallel in the parent and its agr- mutant, whereas all other LPXTG-proteins modified their proteomic profiles independently of their mRNA. Moreover, ClfA became highly transcribed and active in fibrinogen-adherence tests during late growth (24 h), whereas it remained poorly detected by proteomics. On the other hand, iron-regulated IsdA-B-C increased their protein expression by >10-times in iron-poor conditions. Thus, proteomic, transcriptomic, and adherence-phenotype demonstrated differential profiles in S. aureus. Moreover, trypsin peptide signatures suggested differential protein domain exposures in various environments, which might be relevant for anti-adhesin vaccines. A comprehensive understanding of the S. aureus physiology should integrate all three approaches.

Staphylococcus aureus is a highly successful opportunistic pathogen that can produce a wide variety of diseases (1). Moreover, it has a unique ability to develop antibiotic resistance, which reflects its extraordinary capacity to adapt and survive in a great variety of environments. Over the last decades molecular and genetic dissections of S. aureus have revealed a great number of surface adhesins, secreted enzymes, and toxins that might be implicated in pathogenesis (2)(3)(4)(5). In particular, cell-wall-associated surface adhesinsreferred to as microbial surface components recognizing adherence matrix molecules or MSCRAMMs 1 (5)-mediate binding to extracellular matrix and plasma components, enabling staphylococci to colonize and invade host tissues and cells, as well as to escape immune defenses (6 -8). Surface proteins are also implicated in biofilm formation (9), which promotes chronic infections and helps bacteria to escape antibiotic-induced killing.
MSCRAMMs encompass several surface components including proteins, teichoic acids, lipoteichoic acids, and maybe polysaccharidic capsules, which have been implicated in tissue colonization and disease to various extents (5,10,11). Important surface proteins include polypeptides covalently attached to the peptidoglycan via a conserved anchoring mechanism. After membrane translocation, a transpeptidase called "sortase" cleaves the exported protein at a specific LPXTG motif present at its C terminus, and attaches its penultimate threonine to a side-chain of the peptidoglycan stem peptides, i.e. a pentaglycine in the case of S. aureus (12). Twenty-one genes encoding LPXTG-proteins have been identified by in silico analysis of S. aureus genomes (13). Experimental deletion or heterologous expression of these proteins helped identify their physiological functions and infer their roles in diseases (14,15). However, although highlighting the multiple facets of S. aureus pathogenesis, none of these experiments provided a comprehensive view of the infection process. For instance, none of the gene inactivation experiments could entirely abrogate the S. aureus disease capacity, suggesting that infection is a multifactorial process. Moreover, experimental results may be difficult to interpret, because of the complex regulatory gene network operating in S. aureus (e.g. agr, sae, srrAB, arlS, sarA sarR, sarS, sarT, sarV, sarU, sarY, and rot) (3, 16 -22). As an example, the TSST-1 toxin is positively regulated by the "global accessory gene regulator" agr when bacteria are grown in vitro. However, the agr-regulation pathway may become over-ruled by other regulators in vivo, as TSST-1 can be expressed in animals even in the absence of agr (23).
Several approaches have been used to understand the pathogenic behavior of S. aureus. These include genomics, transcriptomics, and more recently proteomics (15, 24 -29). In particular, proteomics might provide the most realistic picture of the infective process, because it detects the very endproducts of gene biosynthetic pathways, which may eventually determine a biological phenotype. Moreover, post-translational protein regulation (or modification) may affect the stability and function of proteins independently of their upstream transcriptional or translational regulation, e.g. proteins may persist longer than their encoding mRNAs. Therefore, in addition to transcriptional and translational regulation, understanding the behavior of an organism requires both qualitative and quantitative assessment of its protein equipment over time.
Here we describe a semiquantitative proteomic approach based on trypsin digestion (i.e. trypsin shaving) of surfaceexposed proteins and spectral counting of resulting peptides. This technique was applied to time course analysis in order to determine the level of LPXTG-proteins expressed in a variety of conditions known to affect the expression of their corresponding mRNAs. Transcriptomic controls were performed in similar conditions using microarrays. Experimental conditions included mutants inactivated in the global regulator agr, which promotes expression of adhesin mRNAs in post-exponential growth phase, and shuts it off in the stationary phase (16,30,31), as well as growth of bacteria in iron-poor or iron-rich media, promoting or repressing the expression of mRNAs of LPXTG siderophore proteins, respectively (32,33). Eventually, the adherence phenotype to host matrix proteins was determined. The results show that the time course profiles of LPXTG-proteins detected on the bacterial surface do not systematically follow the time course profile of their encoding mRNAs (16,30,31), and that some of these proteins can be functionally very active, e.g. in case of adherence to fibrinogen, despite the fact that they are poorly detected in vitro. The results also suggest that surface proteins may adopt different conformations and expose different portions on the surface of different bacteria. Indeed, trypsin digestion released different sets of peptides when LPXTG-proteins were expressed in parent S. aureus or surrogate L. lactis, as exemplified by protein A (Spa), clumping factor A (ClfA), clumping factor B (ClfB) and fibronectin-binding protein A (FnBPA). Our work extends previous proteomic studies in S. aureus (27,34,35) and adds a level of subtlety in the continuum of transcriptional to post-translational protein regulation. Notably, the differences in domain exposure in various bacterial backgrounds might have unforeseen implications in vaccine development.

EXPERIMENTAL PROCEDURES
Bacterial Strains and Growth Conditions-Bacterial strains used in this study are listed in Table I. Staphylococcal strains included the well-described S. aureus Newman, one isogenic mutant (S. aureus ALC355) deleted in the global regulator agr (36), one isogenic mutant (S. aureus DU5873) deleted in the protein A gene (spa) (37), and one isogenic mutant (S. aureus DU5852) deleted in the clumping factor A gene (clfA) (38). Lactococcal recombinants were used for method validation and included the18 previously described L. lactis constructs, each expressing a different staphylococcal LPXTG-protein (39 -41) (see below). Staphylococci were grown at 37°C either in tryptic soy broth (TSB, Becton Dickinson, NJ) or agar, or in Roswell Park Memorial Institute culture medium 1640 (RPMI, Life Technology, Carlsbad, CA), without agitation. Lactococci were grown at 30°C in GM17 broth (M17 medium containing 0,5% glucose, Becton Dickinson) supplemented with 5 g/ml of erythromycin (Sigma-Aldrich) without agitation, or on GM17 agar. Growth was followed by colony counting on plates and OD 600 nm measurements of the different cultures using a spectrophotometer (Ultrospec 500 pro, GE Healthcare). Bacterial stocks were kept frozen at Ϫ80°C in 20% (v/v) glycerol.
Bacterial Cell Preparation for Proteolysis of Surface Proteins-Three different protocols were tested including (1) bacterial cell wall purification (35) prior to proteolysis, (2) bacterial cell wall purification followed by teichoic acid removal with hydrofluoric acid for 48 h (42) prior to proteolysis, and (3) trypsin surface shaving of live bacteria according to a slightly modified described methods (34). Protocols (1) and (2) (described in supplemental experimental procedures) appeared too stringent and resulted in the loss of up to two-thirds of the released peptides. Therefore, trypsin-shaving was used and is described here. In brief, bacteria were grown in 300 ml liquid cultures in the different media described above. At various times of the logarithmic or stationary growth phases, samples (between 10 and 100 ml depending on the cell density) were removed, immediately chilled at 4°C, and harvested by centrifugation. Pellets were washed three times with ice-cold phosphate-buffered saline (PBS) and finally resuspended in 1 ml of the same buffer. To allow semiquantitative comparisons between the proteomes of different samples, cell concentrations were adjusted to 1 ϫ 10 9 bacteria/ml in all samples. Cell counts were validated by optical microscopy (Neubauer cell) and viable colony counts on nutrient agar. There were Ͻ0.5 log 10 differences between the Neubauer cell and viable counts, indicating that the large majority of cells were alive. Samples were then shaved for 1 h with 1 g/ml (final concentration) of trypsin (Promega, Madison, WI) at 37°C, after which they were chilled at 4°C and bacterial cells removed by centrifugation for 10 min at 4000 rpm and 4°C. Supernatants containing trypsin-shaved peptides were filtered (0,22 m) and freeze-dried until further use.
Peptide Preparation for LC-MS/MS Analysis-The freeze-dried shaved peptides were rediluted in 100 l of 100 mM ammonium bicarbonate, reduced with 10 l of 45 mM dithiothreitol (Sigma-Aldrich) for 30 min at 60°C, and alkylated with 10 l of 100 mM iodoacetamide (Sigma-Aldrich) for 30 min at room temperature in the dark. The resultant mixtures were digested a second time at 37°C with 1 g of trypsin (Promega) for 4h. The digested peptides were desalted through Sep-Pak tC18 cartridges (Waters, Milford, MA) following the manufacturer's recommendations and eluted with 1 ml of 60% (v/v) and 1 ml of 30% (v/V) acetonitrile (Merck). Solutions of purified peptides were pooled, dried under vacuum, and kept at Ϫ20°C.
For spraying, a 400 nozzle ESI Chip (Advion Biosciences) was used with a voltage of 1.65 kV, and the mass spectrometer capillary transfer temperature was set at 200°C. In data-dependent acquisition controlled by Xcalibur 2.0 software (Thermo Scientific), the four most intense precursor ions detected in the full MS survey performed in the Orbitrap (range 350 -1500 m/z, resolution 60000 at m/z 400) were selected and fragmented. MS/MS was triggered by a minimum signal threshold of 10Ј000 counts, carried out at relative collision energy of 35%, and with isolation width of 4.0 amu. Only precursors with a charge Ͼ1 were selected for CID fragmentation and fragment ions were analyzed in the LTQ linear trap. The m/z of fragmented precursors was then dynamically excluded, with a tolerance of 0.01 amufrom any selection during 120 s. From raw files, MS/MS spectra were exported as mgf files (Mascot Generic File, text format) using the extract_msn.exe script from Thermo Scientific.
MS/MS spectra were analyzed using Mascot 2.2 (Matrix Science, London, UK). Mascot was set up to search the UNIPROT database (SWISSPROT ϩ TrEMBL, www.expasy.org) restricted to Other Firmicutes taxonomy (database release used was 13.2 of April 8th 2008, 527Ј426 sequences after taxonomy filter). For time course experi-ments, a subset database of UniProt was used (2Ј594 sequences), which contained only proteins of S. aureus strain Newman, as well as sequences of S. aureus LPXTG-proteins expressed in L. lactis clones. Trypsin (cleavage at K, R, not before P) was used as the enzyme definition. Mascot searches were done with a fragment ion mass tolerance of 0.50 Da and a parent ion tolerance of 10 ppm. Iodoacetamide derivative of cysteine was specified in Mascot as a fixed modification. Deamidation of asparagine and glutamine, and oxidation of methionine were specified as variable modifications.
Scaffold (version Scaffold_2_06_02, Proteome Software Inc., Portland, OR) was used to validate MS/MS based peptide and protein identifications, and to perform data set alignment. Peptide identifications were accepted if they could be established at a probability Ͼ90.0% as specified by the Peptide Prophet algorithm (43). Protein identifications were accepted if they could be established at a probability Ͼ95.0% and contained at least one identified peptide. Protein probabilities were assigned by the Protein Prophet algorithm (44). Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony. Relative quantification of proteins between samples was based on spectral counting (45). Spectral counts were normalized by Scaffold (semiquantitative values) to take into account variations of protein amounts between samples.
Microarrays-Total RNAs from two independent triplicates of 100 ml bacterial cultures of S. aureus Newman and its isogenic agrmutant were harvested at OD 600 nm of 0.2, 0.6, 1.8, and 2.2 by centrifugation at 4000 rpm at 4°C for 10 min and processed as follows. Resuspended bacterial cells were first lysed in 100 l TE containing 800 g/ml lysostaphin (Sigma-Aldrich, Saint Louis, USA) for 1 h at room temperature. Total RNA were further purified and stabilized using the RNeasy Protect Bacteria mini kit (Qiagen) following the manufacturer's recommendations. All RNA quantities were assessed by NanoDrop®ND-1000 spectrophotometer and the RNA quality was assessed using RNA 6000 NanoChips with the Agilent 2100 Bioanalyzer (Agilent, Palo Alto, USA). Triplicates were equitably pooled to obtain at least 10 g of RNA. For each sample, 10 g of total RNA were reverse transcribed using dUTP for enzymatic fragmentation; 2 g of the resulting sense cDNA was fragmented by UDG (uracil DNA glycosylase) and APE 1 (apurinic/apyrimidic endonuclease 1) and biotin-labeled with TdT (terminal deoxynucleotidyl transferase) using the GeneChip ® WT Terminal labeling kit (Affymetrix Cat. no. 900671, Santa Clara, CA). Affymetrix GeneChip S. aureus Genome Array (Affymetrix, Cat. no. 900514) were hybridized with 1.8 g of biotinylated target, at 45°C for 16 h washed and stained according to the protocol described in Affymetrix GeneChip® Expression Analysis Manual (Fluidics protocol FS450_0007).
The arrays were scanned using the GeneChip ® Scanner 3000 7G (Affymetrix) and raw data was extracted from the scanned images and analyzed with the Affymetrix Power Tools software package (Affymetrix).
All statistical analyses were performed using the free high-level interpreted statistical language R and various Bioconductor packages (http://www.Bioconductor.org). Hybridization quality was assessed using the Expression Console software (Affymetrix). Normalized expression signals were calculated from Affymetrix CEL files using RMA normalization methods. Differential hybridized features were identified using Bioconductor package "limma" that implements linear models for microarray data (46). The p values were adjusted for multiple testing with Benjamini and Hochberg's method to control the false discovery rate (FDR). Probe sets showing a FDR Ͻ 0.05 were considered significant.
Bacterial Adherence to Solid-phase Extracellular Matrix Compounds-We used a previously described in vitro adherence assay to measure the ability of S. aureus to adhere to increasing concentra-tions of surface-adsorbed fibrinogen, fibronectin, and collagen (47). Briefly, 96-well plates (Nunc-Immuno plates; MaxiSorp surface; Thermo Fisher Scientific) were filled with 100 l of twofold serial dilutions of fibrinogen (1 mg/ml initial concentration; Sigma-Aldrich), fibronectin (250 g/ml initial concentration; Sigma-Aldrich) and collagen I and VI (20 g/ml initial concentrations; Sigma-Aldrich). The last well served as a negative control and was filled with 100 l of PBS without ligand. After washing, bovine serum albumin (Sigma-Aldrich) was added to each well to block nonspecific binding sites. Bacterial cultures were harvested at different times during growth by centrifugation (4000 ϫ g at 4°C for 20 min). Cells were re-suspended in PBS and bacterial cell concentrations were adjusted to 5.10 9 CFU/ml. Fifty microliters (i.e. 2,5.10 8 cells) were added to each well. Plates were incubated for 1.5 h at 37°C, after which wells were washed with PBS and fixed at 55°C. Adherent bacteria were detected by staining with crystal violet, and the OD 570 nm was determined with an enzymelinked immunosorbent assay plate reader (47).

RESULTS
Trypsin-shaving of Live Cells-As mentioned under "Experimental Procedures," initial attempts to recover LPXTG-proteins by trypsin digestion of purified staphylococcal cell walls resulted in too much contamination with nonwall proteins, and/or poor recovery of LPXTG-proteins (described in supplemental Experimental procedures). In contrast, trypsinshaving decreased contamination with nonwall proteins by Ն5 times and reproducibly released similar sets of peptides from individual LPXTG-proteins. Fig. 1 indicates that the recovery of peptides during trypsin treatment of live staphylococci was time-dependent, and that 1 h of treatment appeared optimal. This duration was experimentally amenable for serial extractions during time course experiments and thus was used in all subsequent experiments. The decrease in peptide recovery after longer incubation periods is not ex-plained, but could be because of concomitant proteolysis by intrinsic S. aureus proteases (see Discussion section).
Construction of a Reference Peptide Library of S. aureus LPXTG-proteins Expressed in Lactococci-An important prerequisite to this study was to unambiguously identify the trypsin peptide signatures of S. aureus LPXTG-proteins using the LC-MS/MS system described herein. This was achieved thank to a preliminary analysis of each of these proteins expressed in L. lactis, which does not carry S. aureus proteins (35). This permitted alleviation of certain ambiguities regarding to different adhesin denominations in UniProt, because of redundancies or isoforms. It also allowed verifying if an observed set of peptides could be attributed to a unique parent protein, or whether peptides were found in different proteins showing sequences similarities. To address these questions, we reinvestigated the 18 lactococcal clones successfully expressing unique S. aureus MSCRAMMs (Table I) (35) using the shaving technique, and assigned the obtained peptides to the corresponding proteins. Table II shows that the shaving procedure generated sets of peptides (between 3 and 59 peptides) for 16 out of the 18 LPXTG-proteins studied. The great majority of the detected peptides was specifically assignable to a unique parent LPXTG-protein (Table II). Only few peptides were redundant between more than one protein species, for instance between IsdB and IsdH, Spa and Pls, and between SdrD, SdrE, and SdrC. No peptides were detected for IsdB and SasH when expressed in lactococci (Table II). Possible explanations could be either poor expression in this particular organism, or poor detectability of these peptides by LC-MS/MS. This allowed constructing a concordance table between UniProt protein nomenclatures (shown in Table III) as well as a dedicated sequence database specific for S. aureus Newman (supplemental Table S1). Specifically, Table III also presents the number of unique peptides and the percentage of peptide coverage of each of the LPXTG-proteins detected in recombinant L. lactis. Coverage varied from 6% to 60% (median 31%).
Profiling of LPXTG-proteins in S. aureus Newman and Its agr-Mutant in Various Growth Conditions-Time course profiling of the surface proteome of S. aureus was performed during growth from early logarithmic (OD 600 nm ϭ 0.2) to late stationary (OD 600 nm ϭ 2.2) phases ( Fig. 2). At each time point, the proteomic analysis assessed the relative quantity of proteins in 1 ϫ 10 9 bacterial cells. This semiquantification was based on spectral counting (45) normalized to take into account the variations of protein amounts between samples (n ϭ 3 to 4). Spectral counting measures the number of times that a peptide is selected for fragmentation during a LC-MS/MS analysis, and is correlated to abundance of specific peptides and proteins (45). As dynamic exclusion has a direct impact on spectral counting, a value of 120 s was chosen as a compromise between redundant peptide fragmentations requested for better quantitation accuracy and the need of selecting low abundant peptides for a higher proteome coverage.
Overall Protein Profiling- Fig. 3 depicts the results obtained for the wild-type S. aureus Newman and its agr-mutant grown either in iron-rich TSB ( Fig. 3A and 3C, respectively) or ironpoor RPMI ( Fig. 3B and 3D, respectively). Overall, 16 of the 21 putative LPXTG-proteins described in S. aureus (13) were successfully identified in our tests. Three (i.e. Cna, Pls and SasK) of the 5 undetected species had no gene counterparts in the genome of S. aureus Newman (GenBank Accession Number AP009351) (48) and thus were not expected to be found, and two (Srap and SasC) remained undetected, maybe because of poor detectability of their corresponding peptides by LC-MS/MS. As a negative control, no Spa was detected at the surface of the spa-negative mutant DU5873, whereas the profile of the other LPXTG-proteins remained unaffected in this mutant (data not shown).
Effect of agr- Fig. 3 indicates that the amounts of several LPXTG-proteins depended on agr integrity and growth conditions. When wild-type S. aureus Newman was grown in TSB, nine of the 16 LPXTG-proteins (i.e. Spa, FnBPA, ClfA IsdA, IsdB, SasD, SasG, SasH, and FmtB) showed a time-dependent agr-like bell-shape expression, with an increase in abundance during late logarithmic growth followed by a decrease up to the late stationary phase (Fig. 3A). Although there were some variations between individual adhesins, as well as few relatively unexpected findings (e.g. poor detection of ClfA, see below), such a time-dependent expression pattern is in general agreement with proteins regulated by agr (16,30,31). On the other hand, ClfB was present quite early during growth, and was rather stable until it disappeared in late stationary phase, and SdrD and SdrE reproducibly presented a biphasic expression pattern. Iron-regulated IsdH was not detected in TSB grown staphylococci, most probably because this rich medium provides ample iron for growth (see below).
Strikingly, this agr-like pattern was abrogated when the isogenic agr-negative mutant of S. aureus Newman was tested in similar conditions (Fig. 3C). In this case, the bellshape pattern was replaced by a continuing increase of protein quantities into the late stationary phase for all nine proteins mentioned above. Moreover, several proteins became either detectable or became more expressed in the agr mutant, including FnBPB, ClfA, SdrC, SdrE, SasF, and SasG. This is compatible with the loss of agr-mediated down-regulation of surface protein synthesis during stationary growth phase (16). Besides, some protein decreased (e.g. FmtB), whereas the atypical patterns of ClfB, SdrD, and SdrE persisted.
Effect of Iron-The experiments were repeated in the ironpoor medium RPMI. Of note, the growth rate of strains was substantially slower than in TSB (Fig. 2). Nevertheless, when wild type S. aureus Newman was tested in this condition (Fig.  3B), the global expression profile was very similar to that in S. aureus Newman defective in ClfA (38) TSB, except for the sharp increase in iron-regulated surface determinants IsdA, IsdB, and IsdH (32,33). The IsdC determinant of the iron-capturing system was not analyzed herein. IsdC is processed by Sortase B and associated to the peptidoglycan via a NPQTN module, and its expression should increase as well (32,33). Aside from these major changes, minor differences were also observed, notably increase in the presence of iron of SasD and SasH, and decrease of SdrE (Figs. 3A and 3B). When the isogenic agr-negative mutant was examined in RPMI (Fig. 3D), the loss of the agr bell-shape pattern was much less striking than in TSB. Nevertheless, some obvious modifications occurred such as a significant (p Ͻ 0.05) increase of the detection of SasF and a decrease for SasH.
Transcriptome Analysis-To assess the relationships between the profiles of time course expression of LPXTG-proteins and their mRNAs, we determined the parallel time course transcriptomes of the parent S. aureus Newman and its agr-mutant grown in TSB. Transcriptomic results indicated that the two organisms segregated very well at the level of their global transcriptomes. In addition, all duplicated microarray experiments clustered together, using all 7668 Probe sets, indicating high reproducibility and consistency of the data (supplemental Fig. S1). Fig. 4 presents the dynamics of the relative changes in mRNA amounts for specific transcripts in the parent S. aureus Newman and its agr-mutant. Note that these are relative changes-not absolute mRNA quantitieswith regard to a basal value arbitrarily fixed at 1 for the first time point of the growth curve, i.e. at OD 600 nm ϭ 0.2. Therefore, the relative dynamics of proteomic and transcriptomic profiles ( Fig. 3 and 4, respectively) can be compared.
Considering agr-related genes, the agrϩ parent demonstrated a linear increase (by 1.6-fold) of the RNAIII transcript over the whole growth duration. Conversely, the agr-mutant did not show any hybridization to the structural genes of the arg locus (i.e. argA, agrB, argC, and argD), as well as a Ͼ300-fold decrease in hybridization to RNAIII and hemolysin ␦ as compared with the parent strain (supplemental Table S2). Thus, the transcription of agr was genuinely silenced in the mutant (supplemental Table S2). As an additional control, the mRNA of the gene of protein A (spa), which is typically regulated by agr, followed an agr-like bell-shape profile in the parent strains whereas this profile was flattened in the mutant (Fig. 4), as previously described (30). One additional LPXTGprotein genes sasH adopted an agr-like bell-shape pattern in the parent, which was modified in the arg-mutant. Moreover, the transcription of sasD and fmtb showed statistically significant modifications (p Ͻ 0.001) in the late growth phase, i.e. at OD 600 nm of 2.2, in the agr-mutant. Of note, the transcript of clfA showed a sharp increase in the late growth phase, i.e. at OD 600 nm of 1.8 (p Ͻ 0.001) and 2.2 (p Ͻ 0.05), in both the parent and the agr-mutant (Fig. 4), an observation that comes in support to recent observations of clfA regulation using the RNAseq technology (49). Thus, at least four of the LPXTG-

Number of unique peptides released after trypsin-shaving of the surface of L. lactis strains expressing S. aureus recombinant LPXTG-proteins
The presence of two numbers for certain lactococci expressing a single LPXTG-species (i.e. SdrC, SdrD, and SdrE) indicates the existence of redundant peptides that are shared with other LPXTG-species.   protein mRNAs (spA, sasD, sasH and fmtB) showed clear modifications between the agrϩ and agr-strains, and two of them (spA and sasH) had a clear agr-like profile. On the other hand, most of the other LPXTG-protein genes adopted various mRNA profiles that were essentially not affected by inactivation of agr. When comparing mRNA and proteomic profiles, the expression of Spa clearly followed an arg-like profile at both the transcriptional and translational levels, which was abrogated in the agr-mutant. SasH followed a relatively similar pattern. On the other hand, FnBPA and to a lesser extend FnBPB, ClfA, ClfB, IdsB, SdrC SdrE, and SasF modified their protein expression patterns between the two mutants, but not their mRNA profiles. Only SasG did not modify its proteins and mRNA profiles in both parent strains. Therefore, although the transcriptome profile was remarkably predictive of the LPXTG-protein profile in some cases, it appears that additional factors were interfering with the physical presence-or access to trypsin-of several adhesins at the bacterial surface. Of note, Srap was detected at the transcriptional but not at the protein level.
Correlation Between Proteomic Expression Profiles and In Vitro Adherence Phenotypes- Fig. 3A shows that 9/16 LPXTG proteins detected in S. aureus Newman (i.e. Spa, FnBPA, ClfA, IsdA, IsdB, SasD, SasG, SasH, and FmtB) followed an agr-like expression pattern. Therefore, we tested whether the in vitro adherence profile of this organism followed a similar pattern when grown in the same conditions. Fig. 5 indicates that adherence to fibrinogen adopted differential profiles depending on both the growth medium and the presence or not of an intact agr.
In TSB (Fig. 5A), binding of the parent S. aureus Newman to fibrinogen was more pronounced during exponential growth and decreased in the early stationary phase, thus obeying an agr-like pattern. As a control, the clfA-mutant was virtually unable to bind to immobilized fibrinogen, indicating that ClfA was largely responsible of the phenotype. The early stationary phase drop was even more pronounced in the argmutant, despite the fact that adhesins are supposed to be more expressed during late growth in the agr-mutant (16,30,31). Finally, adherence increased again after 24 h in both strains, a phenotype that did not correlate with the proteomic detection of ClfA (Fig. 3), but correlated well with the clfA gene transcription profile (Fig. 4). In consequence, ClfAmediated binding to fibrinogen did not strictly follow an agr pattern in these experimental conditions, and senescent bacteria were still able to bind fibrinogen to a substantial extent.
Conversely, binding to fibrinogen adopted an agr-like bellcurve in RPMI (Fig. 5B) and this bell-curve was abrogated in the agr-inactivated mutant. Besides, binding to fibronectin and collagens was quasi null (data not presented), which is coherent with the fact that in S. aureus Newman, the genes encoding for fibronectin binding and collagen binding are either truncated (for fnA and fnB) or absent (for cna).
Trypsin Releases Different Sets of Peptides from LPXTGproteins Expressed on the Surface of S. aureus or L. lactis-S. aureus had a much lower adherence score to fibrinogen than ClfA-positive L. lactis in in vitro adherence tests (Fig. 5). This difference could result from a lower expression of ClfA on the surface of S. aureus than on the surface of L. lactis, or from differences in the accessibility to ClfA-binding domains when the protein is exposed on the surface of S. aureus versus L. lactis, or from both reasons. Fig. 6 compares the ClfA, ClfB, Spa, and FnBPA-specific sets of peptides released by trypsin shaving of the surface of recombinant lactococci or S. aureus Newman. In case of ClfA, nine peptides were released from lactococci expressing ClfA and eight from the surface of S. aureus. Thus, the peptides numbers were quite similar. However, among these, three peptides were specific of lactococci and two were specific of S. aureus. Therefore, although the majority of the released peptides were similar (i.e. 6/9 in lactococci and 6/8 in S. aureus), some were specific of the host bacteria, suggesting that different portions of the protein were accessible to trypsin digestion on the surface of the two microorganisms. Details on these peptides are presented in supplemental Table S1.
This small difference in peptide numbers was also true for ClfB and Spa (Fig. 6). Regarding to ClfB, among 20 and 21 released peptides 3 and 4 were specific for L. lactis and S. aureus, respectively. For Spa, among 23 and 29 released peptides, 0 and 5 were specific for L. lactis and S. aureus, respectively. Moreover, in these cases, some peptides, which were recovered in the same LC-MS runs, displayed redundancies between partial and complete hydrolysis (see overlapping black boxes in Fig. 6). Partial hydrolysis could result from a too short duration of trypsin digestion. However, extending the length of digestion to more than 1h did not yield more peptides (Fig. 1). Therefore, partial trypsin hydrolysis of LPXTG-proteins might depend on other factors, possibly including protein conformation and trypsin accessibility.
Finally, an unexpected observation was that twice as many peptides were released from FnBPA expressed in S. aureus than from FnBPA-positive L. lactis (18 versus nine peptides, respectively). This observation is interesting because the fnbpA gene of S. aureus Newman carries a stop codon, which leads to a premature arrest of the transcription and the translation of a protein devoid of the C-terminal LPXTG anchoring domain. Hence, this truncated protein could be free-floating in the cell envelope of S. aureus Newman and thus more accessible to trypsin digestion. Such a possibility would support the differential trypsin accessibility of other surface proteins expressed in either of the two tested bacteria, as suggested above. DISCUSSION S. aureus produces a plethora of virulence determinants (50), which are regulated by a complex network of two-component regulatory systems, DNA-binding proteins, and small RNAs (3,16,21,30,51,52). This explains why there is no simple approach to assess the presence or absence of each individual pathogenic feature along the successive steps of infection. Previous experiments in which specific genes were inactivated were sometimes difficult to interpret, particularly when bacteria were equipped with multiple genes encoding redundant or complementary functions (13,40,53). Moreover, FIG. 5. Adherence of parent S. aureus Newman and its agr-negative mutant to immobilized fibrinogen. The parent S. aureus Newman, its agr-negative mutant, and a clfA-negative (but agrϩ) mutant were grown in TSB (A) or RPMI (B). Samples were removed at various times during the exponential and stationary growth phases (OD 600 nm values are indicated on the graph; last sampling time was 24 h), titrated to identical concentrations of cell bodies, and tested for their ability to stick to immobilized fibrinogen. Negative and positive controls included L. lactis carrying the empty expression vector Pil253 and L. lactis carrying the same vector expressing ClfA, respectively (C). Columns and error bars indicate the mean Ϯ and S.D. of three independent determinations for each isolate. Statistical analysis was performed by pairwise comparisons with Student's t test: asterisks above columns indicate significant differences with the previous sample (*, p Ͻ 0.05; **, p Ͻ 0.001).
gene regulation may vary between in vitro and in vivo conditions (23).
To integrate this multilevel information, experimental systems should allow appraising quantitative snapshots of global macromolecule expression in both in vitro and in vivo conditions. Although this is possible at the level of mRNA (11,29,31), its equivalent at the protein level is as yet less developed (27)(28)(29). Individual proteins can be quantitatively evaluated by Western blotting or in situ hybridization. However, these methods are not amenable to evaluate multiple proteins simultaneously, because of the need of numerous different antibodies and the limited number of dyes that can be used together in a single experiment. Here, we attempted to bypass this limitation by using a proteomic approach. Because we previously contributed to the understanding of the role of S. aureus surface adhesins using heterologous gene expression (13, 39 -41, 53, 54), we in-tentionally concentrated our efforts on the analysis of the time course detection of the 21 known S. aureus LPXTG surface proteins.
Initial analyses indicated that the whole proteome of purified cell walls was much more complex than expected, revealing numerous proteins that were not anticipated to be found in the peptidoglycan and its appendages. The unexpected presence of these species was considered as contamination, at least in the setting of crude bacterial walls purified after mechanical cell breakage. However, this kind of contamination persisted both after harsher purification (e.g. removal of teichoic acids), and in the trypsin-shaving protocol, which was performed on Ͼ99,9% integral cells as assessed by microscopy and colony counts. Therefore, the question as to whether some cytoplasmic proteins might be constitutive parts of the normal wall environment, as also suggested by others (27,34), remains open.  Table II and Fig. 3. Major protein domains and gross amino acid numbering are indicated. Precise amino acid numbering of the peptides is presented in supplemental Table S1. Trypsin-released peptides are represented by the inserted black boxes. Note that some peptides displayed both completely and partially digested species simultaneously (indicated by overlapping boxes). LPXTG motifs are indicated by thin yellow bars and the peptide removed by sortase in gray. SP stands for signal peptides, which are indicated as green boxes. The ligand binding domains are highlighted in blue and S.D. repeat regions in purple.
We previously showed that LPXTG-proteins from S. aureus could be heterogeneously expressed in L. lactis and individually detected by LC-MS/MS in the recombinant lactococci (35,41). On this basis we constructed a peptide library specific to each of these LPXTG-proteins. This library was indispensable to further quantify LPXTG-proteins in the more complex S. aureus environment. With some exceptions, the amounts of LPXTG-proteins in S. aureus increased up to the early stationary growth phase, and decreased thereafter. This bell-shape behavior is reminiscent of agr-regulated surface proteins such as protein A, which is expressed during logarithmic growth and repressed in stationary phase (3,30). In the present experiments, comparisons between proteomic and transcriptomic profiles confirmed this parallelism for protein A, which is in accordance with previous studies (29,31,55). Moreover, we identified at least one additional LPXTGprotein, SasH, that demonstrated similar profile modifications between protein and mRNA detection in the parent strain and its agr-mutant, suggesting that it was also under tight control by agr. Conversely, however, several LPXTG-proteins modified their proteomic profiles between the parent and the mutant despite the fact that transcriptomic profiles remained unchanged. This suggests that, in addition to mRNA, protein expression was further affected by additional factors at the post-transcriptional level, e.g. via interference with mRNA, or post-translational level via protein modification (56) or protease degradation (29,31,55).
This was particularly relevant when comparing proteomic profiles with adherence phenotypes. Taking fibrinogen binding as a model, the present results show that adherence was indeed affected by both agr integrity and growth conditions, but did not follow an absolute agr paradigm. For instance, in TSB, adherence decreased in the early stationary phase of growth and re-increased later on (at 24 h) without a clear correlate with measured amounts of surface ClfA, but with a clear correlate with increasing clfA mRNA. Likewise, in RPMI adherence tended to decrease over time, without a good proteomic correlate either (mRNA was not measured in this condition). This seeming incoherence most likely reflects our lack of understanding of the subtlety of the wall environment, which implicates additional factors that may affect the phenotype. Indeed, the Gram-positive envelope is not a static peptidoglycan scaffold merely decorated with protein, polyols (teichoic and lipoteichoic acids) and polysaccharide appendages. It rather acts as a dynamic interface between the environment and the intracellular milieu. For instance, the S. aureus envelope is constantly traversed by secreted molecules including Ն10 different proteases (57), among which some were shown to regulate LPXTG-proteins by protein degradation (e.g. ClfB) (29,31,55). These could be responsible for the progressive decreases in recovery of unique peptides over time, as observed in Fig. 1.
Alternatively, mutual interactions between various wall polymers may influence the exposure of protein binding do-mains to the extracellular milieu. This was recently exemplified with recombinant ClfA, where artificial lengthening or shortening of the spacer region (R-repeats) between the proximal wall anchor and the outermost binding domains increased or decreased adherence to fibrinogen, respectively, because they modified the exposure of distal binding domains to their ligand fibrinogen (58). These authors reported similar variations in the presence of absence of an expo-polysaccharide capsule. Therefore, the bacterial surrounding may influence the access of exogenous ligands or proteases to LPXTGprotein domains, a phenomenon that observed with trypsin herein.
Apart from these differences, some other proteins also demonstrated differential regulation between TSB and RPMI, including genes of the iron-capturing isd locus, as well as sizable increases in SasD and SasH in iron starvation. The increase in isd genes is expected in low iron medium (32,33). On the other hand, the reason for the increase in SasD and SasH is more difficult to interpret. Although the physiological role of SasD is as yet unclear, SasH (recently renamed AdsA (59)) is a cell wall associated adenosine synthase that converts adenosine-monophosphate into adenosine, a strong immunomodulator helping staphylococci to escape phagocyteinduced killing. Hence, SasH (or AdsA) could well be coregulated with the siderophore locus isd, which expression is induced in experimental S. aureus nasal colonization (11). In this setting, expression of isd could be required for survival in the low-iron mucosal environment, while SasH could be required to damper host defenses and promote bacterial persistence. Eventually, the mRNA of Srap was detected but its encoded protein was not, suggesting the possible lack of access to trypsin shaving.
The present study yielded other interesting observations. First, FnBPA was detected in both purified walls and trypsinshaving experiments, despite the fact that it lacks the LPXTGanchoring module and the entire cell wall proximal D-W regions in S. aureus Newman, because of the presence of a stop codon (60). This is also true for its FnBPB counterpart, which in contrast was barely detected at all our experiments. One possibility for this difference is that the 741 (out of 1018) amino acids of truncated FnBPA is enough for nonspecific wall attachment, whereas the 678 (out of 940) residues of truncated FnBPB is too short. Alternatively, the two proteins could have been differentially expressed in the present experimental conditions, a possibility which has yet to be demonstrated. Another noteworthy observation was the fact that trypsin released different sets of peptides from LPXTG proteins expressed in S. aureus or in recombinant L. lactis. Thus, some kind of differential domain hindrance or exposure must have taken place in the two bacterial backgrounds, as suggested by others (27,34). In the same line, a few LPXTGproteins were not detected at all (i.e. Srap and SasC) in S. aureus Newman, although they were detected by LC-MS/MS in recombinant lactococci (35,41). This raises the question of their conditional expression, as observed in the present experiments and by others for isd genes (32,33), or of differential access of trypsin digestion.
Taken together, two sets of conclusions can be drawn from these results. First, from the biological point of view, they indicate that some LPXTG-proteins followed an agr-like regulation, which was abrogated in agr-negative mutants and correlated with the mRNA transcription profile in the parent strain and its agr-mutant. On the other and, several LPXTGproteins varied their expression without a good mRNA correlate, and remained functionally active for prolonged periods of time, such as, for instance, for fibrinogen-binding. This study revises somewhat the dogma that surface adhesins are essentially active during the exponential growth phase, to colonize new sites, and shut off after colonization, to facilitate bacterial detachment and colonization of other sites. Moreover, it also reveals that the bacterial wall environments are different in S. aureus and L. lactis, potentially resulting in different exposure of LPXTG-proteins at the bacterial surface, which in turns might lead to a differential accessibility for trypsin digestion. Whether this altered access has functional consequences for bacterial adherence remains to be determined. Likewise, whether this could influence the protective efficacy of blocking antibodies might be relevant for vaccine development.
Second, from the technical point of view, the results open the way to semiquantitative and time course proteome analysis of multiple S. aureus pathogenic polypeptides simultaneously. Hence, they add to other recently published proteomic analyses (27,29,34). One theoretical limitation of trypsinshaving is that it peptide-release is limited to the trypsinaccessible proteins. Thus, it may underestimate proteins buried deeper in the multipolymeric wall. However, preliminary purification of cell walls or removal of teichoic acids further decreased the recovery of LPXTG-protein peptides, suggesting that trypsin-shaving was a good compromise in this complex surrounding. Another limitation is strain-dependence, which may require recharacterization of each singular organism. However, the same remark is valid for any physiologic or phylogenic characterization of any isolates. Finally, the method could help determining the real-time behavior of numerous bacterial adhesins not only in vitro, but possibly also in vivo. For the latter case, targeted mass spectrometry techniques based on selected reaction monitoring (61, 62) could be easily developed to specifically detect and quantify bacterial surface molecules in complex matrices such as those obtained from animal models or clinical samples.