The small-colony variant (SCV) phenotype represents a slow growing subpopulation of Staphylococcus aureus associated with chronic, persistent, and recurring infections, which are particularly difficult to diagnose and challenging in terms of treatment (Proctor et al., 2006; Vaudaux et al., 2006; Kahl et al., 2016). The colony morphology and physiological characteristics of SCVs differ widely from the WT, not only due to slower growth, pinpoint colonies, reduced or no pigmentation and hemolytic activity, respectively, but also in several biochemical traits such as altered expression of virulence factors, decreased respiration and coagulase activity as well as frequently by auxotrophism for hemin, menadione, thymidine, or fatty acids (Proctor et al., 2006; Seggewiß et al., 2006; Kriegeskorte et al., 2011, 2014; Proctor et al., 2014; Schleimer et al., 2018).

The identification of the genetic mechanisms leading to the phenotype switch are prerequisite to prevent SCV-related recurrence and chronic infection and to develop new antimicrobials not selecting for SCVs. Hitherto, however, only some phenotype switch-related genes and mechanisms were identified by in vitro generation of knockout mutants (including accC, accD, aroD, cspB, hemA, hemB, menA, menB, menD, plsX, sdhCAB, and thyA) and sequencing approaches (including accC, accD, aroB, aroC, aroD, ecfA, ecfT, fabF, fabI, hemA, hemB, hemC, hemD, hemE, hemG, hemH, menA, menB, menC, menE, menF, relA, stp, and thyA) (von Eiff et al., 1997; Bates et al., 2003; Schaaff et al., 2003; Chatterjee et al., 2008; Lannergård et al., 2008; Duval et al., 2010; Gao et al., 2010; Gaupp et al., 2010; Parsons et al., 2011, 2013, 2014; Köser et al., 2012; Wakeman et al., 2012; Hammer et al., 2013; Dean et al., 2014; Painter et al., 2015; Lin et al., 2016; Cao et al., 2017; Zhang et al., 2017; Bazaid et al., 2018; Giulieri et al., 2018; Schleimer et al., 2018; Vestergaard et al., 2018). Besides these SCVs triggered by mutational events restricted to one gene locus, SCVs being the consequence of combined mutations in two or more genes were also rarely described (Hammer et al., 2013; Bui and Kidd, 2015; James et al., 2019). Moreover, the differential expression of some key transcriptional regulators such as SigB and non-protein-coding RNAs (npcRNAs) were shown to contribute to the phenotype switch (Moisan et al., 2006; Abu-Qatouseh et al., 2010; Mitchell et al., 2013; Bui and Kidd, 2015; Tuchscherr et al., 2015). Nevertheless, the (potential) genetic mechanisms of several clinical SCVs remain unidentified and knockout mutant generation and sequencing approaches including whole genome sequencing (WGS) are time-consuming. Moreover, selective disruption of genes essential for growth is not feasible. Aggravating the analysis of clinical SCVs so far, indubitable identification of genes causative for the phenotype switch is only possible for stable SCVs that already underwent mutational adaption and by comparison with a revertant phenotype that spontaneously emerged from the SCV phenotype (Becker et al., 2006; Schleimer et al., 2018). Besides the rarely isolated stable clinical SCVs, in vitro selection for stable SCVs mandatory for WGS as well as for transcriptomics and proteomics can be performed by cultivation in the presence of sublethal concentrations of certain antibiotics (e.g., aminoglycosides) (Balwit et al., 1994) or by serially passaging bacteria in the presence of HeLa cells using modified gentamicin protection assays (McLean et al., 2019). By using these methods, however, the SCV-selective mutations only occur in the corresponding drug target genes or genes essential for antibiotic uptake (Schaaff et al., 2003; McLean et al., 2019). Further approaches selecting for a stable SCV phenotype include a chemostat to generate steady-state growth conditions with low nutrients and low growth rate for a prolonged time and with specific chemical stress (Bui and Kidd, 2015; Bui et al., 2015). Although this is an adequate method, it is also time-consuming and not applicable in every laboratory. Thus, there is an urgent need for simpler, faster, and more cost-effective methods to identify genes that trigger the phenotype switch upon mutational or transcriptional inactivation or downregulation.

Inducible gene expression systems comprising repressor and operator regulatory elements have been reported for use in the genus Staphylococcus for selectively controlling the (over)expression of genes (Wieland et al., 1995; Ji et al., 1999; Jana et al., 2000; Helle et al., 2011). Amongst these, tetracycline- (Tc) or anhydrotetracycline (ATc)-inducible expression plasmids comprising improved elements originating from the transposable element Tn10 were used not only for overexpression, but also for regulated antisense silencing of distinct genes (Ji et al., 1999, 2004; Stary et al., 2010; Helle et al., 2011). This study focused on the adaption of the expression plasmid pRAB11 (Helle et al., 2011) for the identification of genes potentially triggering the phenotype switch from WT to SCV. Therefore, we developed an ATc-inducible antisense approach outlined in Figure 1. This tet-regulatory system is based on autoregulated expression of the repressor TetR, which is responsible for the adequate regulation of a target gene. TetR attaches to the cognate operator sequences within both its own promoter and in the divergently located promoter of the target gene, thus, inhibiting the transcription of the target gene (Meier et al., 1988). Upon binding of the effector, TetR undergoes a conformational change and the detachment from the operator proceeds (see Figure 1).

Using the further improved shuttle plasmid pSN1, silencing of the SCV phenotype-related gene fabI (encodes for enoyl-acyl-carrier-protein reductase) via production of antisense RNA was performed in a one-plasmid setup. Application of this type of episomal silencing system may provide an easy-to-use and fast tool for the identification of phenotype switch-related genes. The identified genes producing the SCV phenotype will not only allow the screening of clinical SCV isolates, in particular those with undefined auxotrophy, for mutations in these genes by simple Sanger sequencing. Furthermore, knockout mutant generation of the identified genes may also contribute to the identification of possible new auxotrophies and thus to the characterization of undefined SCV types, further simplifying the identification of SCVs. Another advantage of the pSN1 tool is that its inducer ATc is able to penetrate cell membranes. Thus, it may not only help to develop future alternative treatment options of therapy-refractory SCV-related infections, e.g., by iatrogenically induced phenotypic switch, but will also offer the opportunity to monitor the effects and consequences of the switching process on the host cell, organs, and the host organism. Finally, this tool may enable the fast suitability analysis of potential new drug target genes regarding their influence on an adverse phenotype switch to the SCV.

Since SCV-caused infections are difficult to treat, it is of major interest to identify and characterize those genes, which are involved in the phenotypic switch from the WT to the SCV phenotype. However, the approaches used so far for this purpose are expensive, time-consuming, and/or difficult in handling. Moreover, clinically derived strain pairs consisting of both stable SCVs and revertant WTs are prerequisites. Therefore, this study focused on the generation of an easy-to-handle genetic tool for the rapid identification of SCV phenotype generation-related genes.

In our study, we adapted the Tn10-encoded Tc- or ATc-inducible tet-regulated repressor-operator expression system, which was first described in E. coli and then modified in several ways to meet the different requirements for Gram-positive bacteria (reviewed in Bertram and Hillen, 2008). For episomal antisense silencing of phenotype-related genes in S. aureus (see Figure 1), we employed plasmid pRAB11 (Helle et al., 2011) containing the tet-regulatory elements tetR (encodes the repressor), its enhanced autoregulated promoter P_tetR (Geissendörfer and Hillen, 1990; Corrigan and Foster, 2009), and the divergently located promoter-operator fusion P_xyl/tet with two tetO sequences (Geissendörfer and Hillen, 1990; see Figure 1). Since for fatty acid metabolism-related gene fabI, mutations were already shown to be responsible for the phenotype switch to the SCV (Bazaid et al., 2018), we exemplarily used this gene in our experiments. Initial experiments to determine the suitability of pRAB11 unexpectedly revealed pronounced growth retardation for strain SAS32/1 carrying the empty pRAB11 upon induction with the effector ATc (Figure 2). ATc represents a derivative of Tc, but since growth of SA113 in ATc-containing TSB was not affected, its low antibiotic activity (Degenkolb et al., 1991) was not responsible for the slow growth phenomenon. During generation of plasmid pRAB11 for the use in Gram-positive bacteria, several modifications were performed (see Table 1; Geissendörfer and Hillen, 1990; Bateman et al., 2001; Corrigan and Foster, 2009; Helle et al., 2011). Amongst these, adaptation of the −10 sequence within the tetR-driving promoter P_R^∗ formally designed by Geissendörfer and Hillen (1990) resulted in the consensus sequence (5′-tag ag t-3′→ 5′-tat aa t-3′) (Corrigan and Foster, 2009), but unintentionally led to a disruption of its palindromic tetO₁ sequence (see Figure 1C) and therefore, presumptively resulted in increased amounts of TetR due to loss of its negative regulation (Stary et al., 2010). Whilst increased amounts of TetR provide improved repression for P_xyl/tet promotor constructs harboring only one tetO (Corrigan and Foster, 2009) in the uninduced state, these high TetR concentrations may be toxic to the cells (Lutz and Bujard, 1997) as the unbound TetR homodimers are freely available in the induced state. Furthermore, since induction of the P_xyl/tet promoter may be negatively affected by overexpression of TetR (Ehrt et al., 2005), we adapted the system by restoring the tetO₁ sequence within the TetR promoter generating plasmid pSN1. S. aureus cells harboring pSN1 (SAS99/1) showed no signs of growth retardation in presence of ATc (Figure 3) presumptively linked to the fact that, due to the restored tetO₁, the negatively controlled transcription of tetR by TetR as in its original state (Yang et al., 1976; Beck et al., 1982; Wissmann et al., 1988) is reconstituted.

For episomal antisense silencing, a 382-bp fragment of the gene fabI mostly congruent with fragments used before (Ji et al., 2004; Stary et al., 2010) was cloned downstream of promoter P_xyl/tet in pSN1. Induction with ATc resulted in distinct retardation of growth (Figure 3) as shown before with chromosomally encoded tetR (Stary et al., 2010) as well as for episomal silencing with a plasmid harboring a leaky (Zhang et al., 2000) one-tetO-version that was induced with Tc (Ji et al., 2004). Phenotypic characterization of fabI antisense silencing on solid agar revealed colonies exhibiting the SCV phenotype only upon induction with ATc (Figure 4). The SCV phenotype was more pronounced on agar containing 0.4 μM ATc than on agar with 0.2 μM ATc. However, on agar with 0.4 μM ATc, SCVs were heterogeneous in size with 74.83% micro-SCVs, 19.14% mini-SCVs, and 3.73% SCVs, respectively. This differentiation into different SCV phenotypes despite the same treatment and environmental conditions is a phenomenon known for SCVs and implicates an adaption strategy during the infection and persistence process (Tuchscherr et al., 2011). It was therefore expected that the bacterial cells have the ability to react slightly different to the silencing of a particular gene and the environmental conditions. Unaffected WT cells remained for both ATc concentrations at a ratio of 1.17% for 0.2 μM and 2.30% for 0.4 μM ATc. This may indicate that some of the cells were able to escape induction with ATc, which needs to be further investigated. Since qPCR showed a considerably elevated expression of antisense ‘fabI’ only for strain SAS118/2 in the induced state (∼ 200-fold increase compared to induced control strain SAS99/1, both measured after 2 h of growth) (Figure 5), this indicates functional silencing due to pSN1-derived antisense expression.

Since not all antisense RNA fragments are effective in inhibiting gene expression in S. aureus (Ji et al., 2004), we analyzed several fabI fragments from different regions of fabI with and without Shine-Dalgarno sequence that exhibited 75- to 300-bp in size, a GC content of 29.3 to 36.7% and ΔG values of −45.6 to −11.8 kcal/mol (for comparison, the 382-bp fragment exhibited a GC content of 34.8% and a ΔG of −70.3 kcal/mol; see Supplementary Material). However, a silencing effect could only be observed for the 382-bp fabI fragment, not for any of the other fragments tested (data not shown) as it could be also not shown before for a fragment bigger than 500 bp (Ji et al., 2004). Therefore, contrary to expectations, the presence of the Shine-Dalgarno sequence does not seem to be relevant as it was described for gene silencing in E. coli (Stefan et al., 2008). Secondary sequence structure prediction for each fragment revealed different loop regions with unpaired bases that are complementary and with accessible conformation to hybridize with fabI mRNA (data not shown). Even though some of these fragments showed loop regions of a bigger size compared to the 382-bp fabI fragment, this seemed to have no impact on silencing. Instead, the quantity of loop regions may contribute to silencing, as the 382-bp fabI fragment exhibited 20 loops compared to 4 to 16 loops of the other fragments (see Supplementary Material). Therefore, antisense silencing seems to depend significantly on fragment size and target region. Thus, the final effectiveness of the fragments is difficult to predict necessitating further examinations.

Our results demonstrate the crucial role of TetR and its major importance in negative regulation of repressor expression for episomal antisense silencing. However, to perform efficient antisense silencing, size and region of the antisense fragment complementary to the region of the target gene remain critical factors that need further detailed investigations. Plasmid pSN1 allowed antisense silencing of the fabI gene in an easy-to-handle one-plasmid setup. Due to the MCS directly downstream of the inducible P_xyl/tet promoter, genes of interest can easily be cloned into the plasmid. Moreover, the two-tetO P_xyl/tet version exerted a tight repression of the gene of interest in the uninduced state, thus there is no necessity of chromosomal tetR integration prior to use. Also other SCV characteristic (e.g., hemolysis behavior, coagulase and protease production, as well as antibiotic resistance) can be analyzed with this system as already shown for antisense silencing of hla (encodes for α-toxin) (Ji et al., 1999). Owing to induction with ATc, pSN1 mediated antisense silencing may not only be applicable to different S. aureus backgrounds, but also in cell culture studies (Ji et al., 1999; Bateman et al., 2001). Therefore, pSN1 mediated antisense silencing may not only help to identify SCV phenotype-related genes but also to analyse the consequences of the switching process to the host cells and organs.

All relevant data for this study are included in the manuscript and/or the Supplementary Files.

KB and AK designed the study concept. NS, AK, and BB designed the experiments. NS performed the laboratory work, evaluated the data, and drafted and wrote the manuscript. UK contributed to the data evaluation and writing of the manuscript. SF and JS contributed to the laboratory work. RP provided the scientific support regarding SCVs and interpreted the data. All authors have read and approved the final draft of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.