All strains used in this study are derivatives of the S. aureus JE2 (Table 1). The following transposon mutants were retrieved from the Nebraska Transposon Mutant Library (Fey et al., 2013): JE2 hemB::ΦNΣ (NE1845), JE2 menD::ΦNΣ (NE1345), JE2 rpoF::ΦNΣ (NE1109) and JE2 SAUSA300_1252::ΦNΣ (NE142). Strains were grown in TSB at 37°C with shaking at 200 rpm or on TSA agar plates with or without addition of gentamicin (Sigma). Minimal inhibitory concentrations (MICs) of gentamicin were determined using a twofold microbroth dilution assay according to Wikler and Clinical and Laboratory Standards Institute [CLSI] (2009) guidelines, except that cation-adjusted Mueller Hinton broth was substituted with TSB. Auxotrophy of selected gentamicin resistant SCVs was determined by visual growth compensation on TSA plates supplemented with menadione (1 μg/ml) (Sigma) or hemin (1 μg/ml) (Sigma). Construction of MV118 rpoF::ΦNΣ (MV216) and MV118 SAUSA300_1252::ΦNΣ (MV218) mutants was performed by transduction with bacteriophage φ11, selecting for transductants on erythromycin plates (5 μg/ml) (Fey et al., 2013).

Selection of spontaneous gentamicin resistant SCVs was performed on TSA plates supplemented with 4 μg/ml of gentamicin, by applying proper dilutions of JE2 over-night (ON) cultures on the plates. SCV colonies were picked after 48 h of incubation and re-streaked on gentamicin-free TSA plates to examine the stability of the SCV phenotype. If no revertants appeared in the re-streak, one colony from each independent selection was saved. The mutation frequency of gentamicin resistant SCVs was calculated as the ratio of the number of gentamicin resistant mutants divided by the total number of cells (determined by plating proper dilutions on TSA plates).

Genomic DNA (gDNA) was isolated from 1 ml ON cell culture using DNeasy Blood and Tissue Kit (Qiagen). Cell lysis prior to gDNA isolation was achieved by pre-treating ON cultures with lysostaphin (5 mg/ml) for 1–2 h. Genes were amplified with two primer pairs to secure proper quality along the entire genes of interest. Genes of interest were amplified by PCR using Taq DNA polymerase (Thermo Scientific). Primers used for amplification are available in Supplementary Table S1. The PCR program used for amplification was: 5 min denaturation at 95°C, followed by 30 cycles of (i) denaturation 95°C for 30 s, (ii) annealing at proper temperature for 30 s for each primer pair and (iii) elongation for 1 min/kb at 72°C. A final elongation step of 5 min was included. Amplification products were verified by gel electrophoresis and DNA concentration measured on a NanoDrop 1000 (Thermo Scientific). Sanger-sequencing was performed by Macrogen Europe Inc (Amsterdam, the Netherlands).

Whole genome sequencing was performed at the Department of Clinical Microbiology at Hvidovre Hospital on a MiSeq (Illumina, San Diego, CA, USA). DNA concentrations were normalized using a Qubit (Invitrogen, UK). Libraries were made with Nextera XT DNA sample preparation kit (Illumina, US), genomes multiplexed to 24 isolates per run and sequenced with 2 × 150 bp paired-end reads. Analysis of output reads from the MiSeq was performed in CLC Genomics Workbench version 8.0. Reads were aligned to the S. aureus USA300_FPR3757 reference genome (Genbank accession no. NC_07793). Variants were called at standard settings. To exclude false-positive variant calls due to alignment of reads to the S. aureus USA300_FPR3757 reference genome, we sequenced the JE2 strain that is the ancestral strain of the selected SCVs. Variants identified in our JE2 sequence were excluded in the selected SCVs and evolved strains by the function ‘Filter Variants against Control Reads.’

Compensatory evolution was performed in TSB and in TSB supplemented with gentamicin at aaa12 MIC of the respective SCV strain. Five independent lineages started from single colonies for each strain were passaged under both culturing conditions. The evolution experiment was performed via serial passaging in 10 ml falcon tubes containing 1 ml growth medium in each tube for proper aeration. Every 24 h 1 μl, corresponding to approximately 10⁶ cells, was transferred into fresh growth medium, allowing for growth of approximately 10 generations per transfer. After every 50 generations of evolution, samples from each lineage were plated on TSA plates and visually inspected for growth compensation based on colony size. Restored fitness in compensated lineages could also be recognized by visual inspection of the culture cell density. If cells in a lineage reverted to normal colony phenotype, one colony was saved from that lineage. The strains not showing changes in colony size were passaged for 50 days (∼500 generations), at which point one colony from each lineage was saved.

Fitness was estimated by two parameters. (i) Growth rates in exponential phase were measured in a Bioscreen C reader (Oy Growth Curves Ab) at 37°C. Bacteria from ON cultures were diluted to a final concentration of 10⁶ CFU/ml in 1 ml TSB or in 1 ml TSB supplemented with gentamicin (aaa12 MIC of the parent SCV strain). Three hundred microliter was transferred to each well of the bioscreen plate with measurements taken every 5 min at 600 nm wavelength. The relative growth rates were calculated as the growth rate of the mutant divided by the growth rate of the WT strain. Three biological replicates were performed for each mutant, with three technical replicates per biological. (ii) colony forming unit (CFU) counts at stationary phase after 24 h growth was determined by growing cells in 1 ml TSB in a 10 ml falcon tube with constant shaking at 200 rpm. Cells from ON cultures were diluted to a starting inoculum of 10⁶ CFU/ml. CFU quantification was determined on TSA plates. Three biological replicates were performed for each strain.

Assessment of variations in membrane potential was estimated using a flow cytometry assay based on the BacLight Bacterial Membrane Potential Kit (LifeTechnologies). Cells from ON cultures were inoculated in 30 ml TSB in 300 ml Erlenmeyer flasks and grown to an OD₆₀₀ of 0.2. Fifteen microliter culture was transferred to 1 ml filtered phosphate-buffered saline (PBS). To each cell solution 10 μl of fluorescent membrane potential indicator dye, DiOC₂(3), was added and cells were stained for 30 min at room temperature. Data was recorded on a BD Biosciences Accuri C6 flow cytometer (Becton, Dickinson and Company), with emission filters suitable for detecting red and green fluorescence. Settings on the flow cytometer were as follows: 50000 recorded events at a FSC threshold of 15000 and medium flow rate. Gating of stained cell population and analysis of flow cytometry data were performed in CFlow^® (BD Accuri). As an indicator of membrane potential the ratio of red to green fluorescence intensity was calculated. The assay was verified using the two SCV strains with transposon insertions in hemB (NE1845) and menD (NE1345).

Significant difference was calculated by 1-way ANOVA, with a post-hoc analysis of Tukey’s Multiple Comparison Tests (^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001). Statistical analysis was performed in GraphPad Prism 4 (GraphPad Software, Inc.).

Sequence reads from all isolates are deposited in the European Nucleotide Archive under study accession no. PRJEB15409.

Spontaneous gentamicin resistant mutants were selected on TSA plates containing 4 μg/ml gentamicin, corresponding to 2x MIC of WT S. aureus JE2. All resistant mutants displayed a small colony phenotype and were selected with an average frequency of 2 × 10⁻⁶. Five mutants showing a stable small colony phenotype when restreaked on antibiotic-free plates were whole genome sequenced (WGS) in order to identify the resistance conferring mutations. Only a single mutation was identified for each mutant (Table 2). Four mutations were located in the menaquinone (MenA Asn198Lys and MenD Ala413fs) and hemin (HemH Glu263* and HemB Pro240Leu) biosynthesis pathways. The fifth strain contained a deletion in the SAUSA300_1683 gene (SAUSA300_1683 Leu274fs), encoding for the bi-functional enzyme 3-deoxy-7-phosphoheptulonate synthase [EC: 2.5.1.54]/chorismate mutase [EC: 5.4.99.5]. This enzyme is required for the biosynthesis of chorismate, the first precursor molecule of the menaquinone biosynthesis pathway (Wakeman et al., 2012). All five mutants displayed a significant reduction in membrane potential, when assayed with the indicator dye, DiOC₂(3) (Table 2). The reduction in membrane potential of the SCVs correlated with 8–16 times increased gentamicin MIC (16–32 μg/ml), compared to the MIC of the WT strain (2 μg/ml). The menD and SAUSA300_1683 displayed menadione auxotrophy, while the hemH and hemB displayed hemin auxotrophy. Growth complementation with menadione could not be achieved for menA (Lannergård et al., 2011) (Table 2).

We estimated the fitness cost of the SCV mutations using two parameters (i) exponential growth rate and (ii) CFU in stationary phase. The relative growth rate of the five SCV mutants were 0.59–0.72 in TSB, while being 0.49–0.62 in TSB supplemented with gentamicin compared to the WT strain grown in TSB (Table 2; Figure 1). Furthermore, the SCVs experienced 10–30 fold lower CFUs in stationary phase, when grown in TSB (Figure 2).

To investigate if fitness could be ameliorated in the SCVs and address the compensatory mechanisms involved, we conducted an evolution experiment with and without sub-lethal concentrations of gentamicin (aaa12 MIC), selecting for mutants with improved fitness. For each SCV mutant we passaged five lineages without and five lineages with gentamicin and at regular intervals screened for higher fitness by visual examination of bacterial colony size on TSA plates. To separate potential adaptation to the growth medium from SCV specific compensatory events we passaged five WT lineages for 500 generations in TSB in parallel. None of the evolved WT lineages displayed improved fitness in terms of exponential growth rate or stationary CFU count, indicating that S. aureus JE2 is generally well adapted to growth in TSB (Figures 1 and 2).

Out of the 50 passaged lineages, only 12 reverted to WT colony size (Supplementary Table S2). All of the 12 lineages originated from SCVs containing point mutations (hemH, menA or hemB) and the reversion occurred only when passaged without gentamicin. Interestingly, in a single passage (∼10 generations) revertants with WT colony size accounted approximately for 50% of the total colonies in the population (tested upon plating on antibiotic-free plates), where no revertants had been observed at the former passage. This fast take over cannot be explain solely with increased growth rates but might also indicate that revertants utilize nutrients that the SCV mutant cannot, which is supported by the observation that revertants grow to approximately 10-fold higher cell densities than SCVs (Figure 2). Reversion to WT colony size proceeded within 50–310 generations for the 12 lineages. As the reverted clones in a single passage constituted 50% of the population, the rate for reversion to WT colony phenotype was treated as a stochastic event. The rate of reversion was estimated to be approximately 10⁻⁸–10⁻⁹ per cell per generation (Maisnier-Patin et al., 2002), similar to previous reversion rates of SCV mutants with point mutations (Lannergård et al., 2008; Pränting and Andersson, 2011). Fitness in terms of exponential growth rate and stationary CFU of the 12 lineages increased to WT levels (Figures 1 and 2). The relative growth rates for the compensated strains ranged from 0.89–1.00, compared to 0.60–0.72 of the ancestral SCVs (Figure 1). The lineages also gained sensitivity toward gentamicin (Figure 3) and an assessment of fitness was therefore not possible to do in TSB supplemented with gentamicin (Figure 1).

None of the 10 lineages derived from deletion SCV mutants (menD and SAUSA300_1683) and evolved without gentamicin reverted to WT colony phenotype, even though they displayed improved fitness in terms of growth rate and CFU in stationary phase (Figures 1 and 2). The relative growth rates in TSB of menD lineages were 0.65–0.73 compared to 0.63 of the ancestral menD, while the relative growth rates of SAUSA300_1683 lineages were 0.65–0.76 compared to 0.59 of the ancestral SAUSA300_1683. Only three of the lineages displayed twofold reduced MIC compared to respective ancestral SCVs, while the remaining seven lineages retained the same MIC as respective ancestral SCVs (Figure 3). The increased growth rates of the 10 lineages resulted in a trade-off with significantly reduced growth rates for the majority of the lineages, when grown in TSB supplemented with gentamicin (Figure 1).

None of the 25 lineages evolved with gentamicin restored colony phenotype to WT levels. Nonetheless, the majority of lineages (n = 22/25) displayed improved growth rates in TSB supplemented with gentamicin, while the growth rates were increased for 15/25 in TSB (Figure 1). Increased levels of resistance to gentamicin (2–8 fold) were observed for 20/25 lineages (Figure 3).

For the point mutation SCVs evolved in TSB, Sanger-sequencing of the respective SCV conferring genes (hemH, menA or hemB) revealed intracodonic mutations in 12/15 lineages (Table 3). We identified reversion to WT sequence on nucleotide level in eight lineages and for two lineages on amino acid level. In two lineages, compensatory mutations substituted the SCV conferring amino acid with a non-synonymous amino acid compared to the WT sequence. The acquisition of intracodonic compensatory mutations in the 12 lineages correlated with reversion of membrane potential to the level of the WT (Figure 4) and gentamicin re-sensitization (Figure 3). In the remaining three lineages, we did not identify any intragenic mutations in the SCV conferring genes. These lineages indeed displayed lower levels of fitness amelioration compared to the lineages carrying intracodonic compensatory mutations. The two lineages (E68 and E69) retained membrane potentials as ancestral SCV and remained resistant to gentamicin, while lineage E47 displayed partial growth rate and membrane potential restoration and gentamicin re-sensitization (Supplementary Table S2).

To identify compensatory mechanisms in the lineages that did not ameliorate the fitness cost via restoration of the membrane potential, we WGS all lineages derived from deletion SCV mutants (menD and SAUSA300_1683) exhibiting improved fitness, as well as the five WT control lineages. All of the evolved SCV lineages still contained the original SCV mutation (Figure 5; Supplementary Table S3). We identified multiple genes that acquired mutations in parallel lineages, suggesting selection for such changes. No mutations from the control lineages were identified in the evolved SCV lineages, suggesting that mutations in the SCV lineages were specific to the effect of the SCV mutations and not due to general adaptation to the growth environment.

From the menD and SAUSA300_1683 lineages evolved without gentamicin, 7/10 acquired mutations in genes related to the σ^B stress response [spoVG (n = 4), rsbU (n = 1) and rpoF (n = 2)]. The spoVG gene encodes for the downstream modulator SpoVG of the σ^B general stress response (Schulthess et al., 2011), rsbU encodes for the RsbU activator of σ^B (Hecker et al., 2007), while rpoF encodes for the alternative sigma factor SigB (σ^B) (Hecker et al., 2007). None of the seven sequenced lineages evolved with gentamicin had acquired mutations related to the σ^B stress response. To determine whether inactivation of the σ^B response confers a fitness advantage in TSB we inactivated rpoF in MV118 with a transposon insertion derived from the Nebraska Transposon Mutant Library (Fey et al., 2013). The exponential growth rate of this mutant (MV216) was 7% higher in TSB than ancestral MV118, however, displayed a significant growth defect in TSB supplemented with gentamicin (Figure 6). This correlates well with the observed growth rate trade-off in the majority of menD and SAUSA300_1683 lineages evolved without gentamicin (Figure 1). Furthermore, three of the lineages evolved without gentamicin displayed non-synonymous missense mutations in the sensor histidine kinase srrB of the staphylococcal respiratory response two-component system SrrAB (Kinkel et al., 2013). Two of these lineages did not contain any σ^B-related mutations.

Independent of whether menD and SAUSA300_1683 lineages evolved with or without gentamicin, 12/17 lineages acquired mutations in the gene SAUSA300_1252, encoding a alanine/glycine:cation symporter. Premature stop-codons and INDELs causing frameshifts suggest that the symporter is non-essential and that the mutations cause a loss of function. We inactivated SAUSA300_1252 with a transposon insertion in SCV strain MV118 (MV218) to assay the fitness effect in TSB and TSB supplemented with gentamicin. We could not identify any difference in terms of exponential growth rate between MV118 and MV218 in TSB or TSB supplemented with gentamicin (Figure 6).

Increased levels of gentamicin resistance were detected in 20/25 of the lineages evolved with gentamicin compared to respective ancestral SCV mutants (Figure 4). We WGS seven menD and SAUSA300_1683 lineages evolved in the presence of gentamicin. All of the seven lineages contained mutations in either fusA and/or SAUSA300_0749, while none of the lineages evolved without gentamicin contained mutations in these genes (Figure 5). The gene fusA encodes the elongation factor G (EF-G) and mutations were identified in three lineages (E72, E75, and E76). The combination of fusA and a SCV conferring mutations has previously been described for in vitro selected kanamycin resistant mutants in S. aureus (Norström et al., 2007; Lannergård et al., 2011). The gene SAUSA300_0749 encodes a product of unknown function and has not previously been associated with gentamicin resistance. A premature stop-codon in strain E74 (Gln3^∗) suggests that the gene is non-essential and that loss-of-function confer the resistance.