We experimentally evolved two S. aureus isogenic variants of USA300 JE2 (Diep et al., 2006), MRSA and MSSA, with or without a host and sub-MIC antibiotic exposure. The ancestral MRSA and MSSA isolates did not differ in terms of total growth without oxacillin, but exhibited a slight decline in sub-MIC oxacillin (Figure 1A and B). For each ancestor, we passaged six independently evolving populations under each condition 12 times, for a total of 48 evolved S. aureus populations (Figure 1C).

To assess the changes in virulence of evolved S. aureus, we measured the mortality of C. elegans infected with evolved pathogens. While the MRSA and MSSA ancestors caused similar levels of host mortality (dotted and dashed lines in Figure 2A), there was a significant treatment effect for evolved pathogens (Figure 2A). For the MRSA genotype, host and oxacillin exposure selected for the greatest virulence. Conversely, pathogens evolved in the absence of either pressure exhibited attenuated virulence. These populations also had significantly greater variance compared to those under selection from host and oxacillin, and those under solely host selection. For MSSA, host and oxacillin exposure similarly favored greater virulence over the other three conditions. There was no difference between variances.

We also characterized the hemolytic activity of evolved populations, which correlates with the ability to secrete extracellular toxins and virulence (Cheung et al., 2012). At least one population from most treatments had lost the ancestral ability to hemolyze sheep’s blood (Figure 2B). Sub-MIC oxacillin maintained hemolytic activity, suggesting that constant exposure to low concentrations of oxacillin favored retention of extracellular toxicity. While hemolysis was not necessary for increased host killing, this ability was more often found in populations causing greater nematode mortality (Figure 2C). As virulence was still heightened in pathogens unable to destroy red blood cells (e.g. MSSA pathogens evolving without host or oxacillin caused greater than 90% mean mortality despite none being hemolysis-positive), other virulence factors may be compensating for the absence of hemolysis.

Natural populations of S. aureus exhibit variation in hemolytic ability (King et al., 2016), even within an individual host (McAdam et al., 2011). We thus hypothesized that a significant selective pressure like oxacillin would favor little variation of this trait within a population. We determined the hemolysis status of isolates from each evolved population (10 isolates × 48 populations)—most populations were in consensus (Figure 2—figure supplement 1). Furthermore, oxacillin still has a significant effect, where hemolysis was maintained when under sub-MIC exposure selection (Figure 2D). Oxacillin favored less hemolytic diversity in MSSA populations, further demonstrating the interaction between antibiotics and virulence, and that antibiotics can shape variation in virulence traits.

We measured the in vitro growth of evolved populations to evaluate how pathogen fitness outside the host had been impacted. We used rich media to replicate the conditions under which S. aureus evolved during the experiment. Importantly, rich media reduced the risk of introducing additional selective pressures than those being tested. In media without oxacillin, there were no significant differences in total growth between treatments (Figure 3A). However, in sub-MIC oxacillin, MRSA populations under selection from both pressures exhibited the greatest growth (Figure 3B), with some achieving more growth than those in the absence of oxacillin. By contrast, exposure to sub-MIC oxacillin alone yielded the lowest growth, suggesting a fitness cost. Similarly, MSSA populations exposed to host and sub-MIC oxacillin exhibited a moderate increase in growth compared to other combinations of selective pressures.

We then determined the MICs of evolved populations to ascertain how host and sub-MIC exposure affected the evolution and maintenance of antibiotic resistance. For the MRSA genotype, all but two populations retained the ancestral level of oxacillin resistance (16 µg/mL oxacillin; Figure 3C). During evolution in hosts without antibiotic selection, two populations lost their resistance and exhibited similar susceptibility as the MSSA ancestor (0.25 µg/mL). For the MSSA genotype, host and oxacillin exposure selected for decreased antibiotic sensitivity, up to eightfold the ancestral MIC. These results suggested sub-MIC exposure combined with host factors potentiated the increase in antibiotic resistance in MSSA. One population under selection from solely the host also had an increased MIC, supporting previous evidence showing non-antibiotic selective pressures, such as a host in our study, can select for reduced antibiotic susceptibility (Koch et al., 2014). There was no variation in antibiotic resistance within MRSA and little variation within MSSA across the most genetically diverse populations (Figure 3—figure supplement 1).

We conducted whole-genome sequencing of populations to identify mutations arisen from host and antibiotic selection. Below, we focus on mutations that had swept to fixation (Figure 4, Supplementary file 1; Figure 4—figure supplements 1 and 2) summarizing mutations that were below 100% frequency in each population. Populations evolved from the MSSA ancestor had significantly fewer mutations (excluding those intergenic or synonymous) than MRSA populations (Figure 4—figure supplement 3; χ12\begin{document}$\chi _{1}^{2}$\end{document} = 5.28, p=0.022), potentially due to the slightly reduced growth of MSSA in the presence of oxacillin (Figure 1B). Sub-MIC oxacillin selection also resulted in more mutations than in its absence (χ12\begin{document}$\chi _{1}^{2}$\end{document} = 5.92, p=0.015), although this is likely driven by MRSA populations. While this result is consistent with the role of antibiotics in increasing mutation rates in bacteria (Revitt-Mills and Robinson, 2020), there were only two mutations in DNA and mismatch repair genes (mutL and recA), suggesting repair genes were not the sole mechanism involved. Of the 32 indels detected, 19 were in host-associated populations. Across all populations, many mutations occurred in the same gene (e.g. agr, gdpP, graSR, pbpA, and saeRS) but not the same amino acid (Supplementary file 2), indicating parallel evolution at the gene level.

Between pathogens with the highest virulence—those under selection from both selective pressures—populations arisen from MRSA had mutations in 25 genes and intergenic regions, while those from MSSA had mutations in 12 genes and intergenic regions. Both treatments had mutations in virulence regulators codY (Brinsmade, 2017) and saeRS (Montgomery et al., 2010), suggesting some shared pathways to heightened virulence among the two treatments, but mutations in these genes did not occur across all replicate populations. These treatments also had mutations in gdpP and pbpA, which have been implicated in resistance to beta-lactam antibiotics (Bilyk et al., 2022; Poon et al., 2022), with gdpP playing a role in MSSA resistance to oxacillin in particular (Giulieri et al., 2020). Five out of six populations evolved from the MSSA ancestor had a mutation in either gene, which may have contributed to the reduced oxacillin sensitivity in these populations.

Regulatory genes were enriched with mutations (Figure 4; one-sample Poisson test p<0.001 for MRSA and MSSA genotypes). These included agr, saeRS, and codY. agr is a quorum sensor that regulates the expression of virulence factors (Novick, 2003). All 22 mutations in agr were found in populations that have lost the ability to hemolyze red blood cells. In contrast to the other two regulators, codY is a transcriptional repressor found in Gram-positive bacteria that controls virulence by monitoring nutrient levels in the environment (Brinsmade, 2017).

Nine populations experienced large-scale deletions that included the entire SCCmec cassette. Seven of the nine populations were host-associated. Deletion of ACME has been shown to enhance the competitive fitness of S. aureus USA300 in vivo (Diep et al., 2008). The two MRSA populations that lost their resistance (Figure 3C) had SCCmec deletions, suggesting that resistance could be more costly when evolving in a host. The loss of SCCmec and ACME was more often identified in populations exhibiting an increase in total growth from the ancestor outside the host (Figure 5A).

We identified mutations arising in specific genes that appeared more often than by chance in measured traits to pinpoint loci potentially underlying evolved phenotypes. We focused on genes where two or more mutations (excluding intergenic or synonymous mutations) had fixed during the experiment regardless of treatment (Figure 5B). Mutations in three genes were regularly identified in populations exhibiting significant increases in virulence from the ancestor: codY, gdpP, and pbpA. Mutations in agr in general were not associated with changes in overall virulence, but MSSA populations harboring mutations in this gene were more likely to exhibit greater virulence compared to MRSA populations (Wilcoxon rank-sum exact test p=0.045).

Mutations in specific genes were often found in populations able to hemolyze red blood cells (Figure 5C). There was a greater proportion of populations with agr mutations unable to hemolyze red blood cells compared to alr, argR, codY, gdpP, and purR. There were also significant differences between the mutations regularly identified in oxacillin-resistant populations evolved from the MSSA ancestor (Figure 5D), where a greater proportion of populations with mutations in codY had reduced oxacillin susceptibility compared to brnQ1. Exposure to oxacillin maintained hemolysis in evolved populations (Figure 2B), suggesting that low concentrations of oxacillin exerted negative selection on the agr locus. By contrast, mutations in agr were often in populations exhibiting loss of hemolytic activity, consistent with previous findings (Schwan et al., 2003).

Because mutations in codY appeared to be important for both virulence (Figure 5B) and antibiotic resistance (Figure 5D), we hypothesized that traits controlled by codY were responsible for the traits observed in MSSA populations under selection from both pressures. A potential mechanism underlying changes in virulence and antibiotic sensitivity in MSSA may involve biofilm formation. Biofilm is implicated in S. aureus pathogenesis, as well as in dampening the efficacy of antibiotic treatment (Long et al., 2023). In C. elegans, S. aureus biofilm enhances virulence and protects the pathogen from host innate immune defenses (Begun et al., 2007). Both sub-MIC levels of beta-lactam antibiotics and null mutants of codY induce robust biofilm formation and hemolysis activity (Chen et al., 2021; Kuroda et al., 2007; Majerczyk et al., 2008). Taken together, exposure to sub-MIC levels of oxacillin and acquisition of potentially deleterious mutations in codY may underlie both increased virulence and reduced antibiotic susceptibility. Our findings partially supported this hypothesis: biofilm formation did not significantly differ for the MRSA genotype. By contrast, biofilm production differed between treatments for the MSSA genotype (Figure 5E), with populations exposed to both selective pressures forming more biofilm than those evolved without host or either pressure, but not those evolved in solely sub-MIC oxacillin. While host and sub-MIC oxacillin selection favored robust biofilms in MSSA, oxacillin by itself also increased biofilm formation, suggesting that low antibiotic concentrations contributed significantly to the evolution of drug-sensitive populations. The loci underlying biofilm formation may be different in these two treatments, since mutations in codY did not appear in populations under solely oxacillin selection (biofilm is a polygenic trait in S. aureus; Zapotoczna et al., 2016).

We then determined whether mutations that arose in our experiment exist in natural S. aureus isolates. We conducted BLAST searches against a dataset of 83,383 high-quality public S. aureus whole-genome assemblies (Raghuram et al., 2024; Figure 6A) and found matches against the majority of our mutations. We hypothesized that these mutations may be important for S. aureus adaptation to different environments. Indeed, for many genes implemented in virulence and antibiotic resistance, a greater proportion of natural isolates containing our mutations were found in blood and systemic infections compared to skin/nose/throat colonization than expected (Figure 6, statistics shown in Supplementary file 3). Mutations arisen in MRSA populations under selection from both host and sub-MIC oxacillin had the most matches to natural isolates (Figure 6B), further highlighting the importance of the interaction between these two selective forces. Overall, our results demonstrate that evolution experiments in an invertebrate model can give rise to clinically relevant mutations in a major bacterial pathogen.

We conducted principal component analysis on the evolved traits (virulence, growth in media with or without sub-MIC oxacillin, and biofilm production) in MRSA and MSSA populations to determine the impact of selective pressures on overall pathogen phenotypic evolution (Figure 7A and B). We found a significant effect of treatment for the MRSA genotype, where populations exposed to host and sub-MIC oxacillin clustered together, largely separating from all other treatments (p<0.05 for all pairwise comparisons). The MSSA genotype had a marginally non-significant effect of treatment. Populations evolved from the MRSA ancestor had similar trajectories for growth in sub-MIC oxacillin and virulence, whereas biofilm production and virulence were more correlated for populations evolved from the MSSA ancestor. For MRSA populations, biofilm production and growth without oxacillin also appeared to be positively correlated.

To determine how host and sub-MIC antibiotics shaped pathogen genomic evolution, we built phylogenies using mutations that arose from the MRSA and MSSA ancestors (Figure 7C and D). We compared Euclidean distances calculated from the frequency of each mutation to determine the genetic distance from the ancestor—how fast each population was evolving—and the genetic distance between populations within each treatment—how much each population had diverged from one another (Betts et al., 2018; Ford et al., 2017). There was no significant difference between treatments for MRSA populations in terms of genetic distance to the ancestor (Figure 7—figure supplement 1A), suggesting that populations generally exhibited similar evolutionary rates. By contrast, MSSA populations had a significant effect of treatment (Figure 7—figure supplement 1B), where populations under selection from solely the host had greater genetic distance from the ancestor than populations under selection from solely oxacillin (Tukey’s post hoc test p<0.001) or under neither selective pressure (p=0.004).

For the MSSA group, populations within each treatment tended to exhibit more genetic similarity (Figure 7C and D). For MRSA, treatments varied in how divergent populations were (Figure 7—figure supplement 1C), where populations exposed to both selective pressures diverged more from each other than populations in the other three treatments (Dunn’s post hoc test p<0.05 for all pairwise comparisons). Likewise, MSSA populations also differed in how much they varied (Figure 7—figure supplement 1D). Populations under solely host selection had greater divergence from each other than all other treatments (Tukey’s post hoc test p<0.001 for all pairwise comparisons). By contrast, populations under selection solely from sub-MIC oxacillin diverged the least (p<0.05 for all pairwise comparisons). These findings altogether suggest that even concentrations well below the MIC differentially impact sensitive and resistant pathogens. As MRSA and MSSA only differed in the presence of an intact mecA gene at the start of the experiment, accessory genes may play important roles in shaping bacterial evolution (Jackson et al., 2011).

S. aureus USA300 JE2 (NR-46543) and USA300 JE2 mecA-tn (Transposon Mutant NE1868) from the Nebraska Transposon Mutant Library were acquired from BEI resources (https://www.beiresources.org/). To ensure true isogenic ancestors, the mecA-tn was transduced into the USA300 JE2 background, selecting for the erythromycin resistance marker on the Mariner transposon, and the insertion site was confirmed by polymerase chain reaction. Transposon maintenance was confirmed by testing for resistance to erythromycin (5 µg/mL) or by genome sequencing when available. In cases where transposon loss was verified by genome sequencing, strains were tested and shown to maintain sensitivity to oxacillin. As the primary function of the transposon was to knock out mecA function, loss of the transposon should not confound subsequent analyses as antibiotic sensitivity was maintained.

C. elegans strain N2 Bristol and E. coli strain OP50 were provided by the Caenorhabditis Genetics Center, which is funded by the NIH Office of Research Infrastructure Programs (P40 OD010440). Nematodes were maintained on Nematode Growth Medium Lite (US Biological, Swampscott, MA, USA) according to WormBook protocols (Stiernagle, 2006).

Selection for increased virulence of S. aureus (USA300 JE2 or USA300 JE2 mecA-tn) in C. elegans was performed by passaging S. aureus from dead hosts 24 hr post-infection. Half of the S. aureus populations were additionally subjected to antibiotic pressure during the passages before and after C. elegans selection with sub-MIC oxacillin exposure (0.03125 µg/mL). This concentration was tested against both ancestral strains and did not substantially inhibit the growth of the methicillin-sensitive USA300 JE2 mecA-tn strain.

For each passage of experimental evolution, S. aureus was grown in Brain Heart Infusion (BHI) broth with 4 µg/mL colistin (to prevent E. coli contamination from the nematode intestine)±0.03125 µg/mL oxacillin and incubated at 37°C overnight, shaking at 250 rpm. The next day, 200 µL of the overnight S. aureus culture was plated onto BHI agar and incubated at 28°C overnight. Approximately 1500 C. elegans were plated and allowed to consume S. aureus for 24 hr before 30 dead nematodes were picked from each plate. Nematodes were identified as dead by a lack of response to tapping by a platinum wire (Ford et al., 2017; Hoang et al., 2024a). Bacterial extraction methods were adapted from Vega and Gore, 2017. In brief, picked dead nematodes were washed once with M9 buffer supplemented with 0.1% Triton X-100, once with M9 buffer, then subjected to a 1:1000 bleach-M9 buffer solution for 15 min. Nematodes were then washed twice with M9 buffer and ground with a motorized pestle to extract S. aureus from the host intestine. S. aureus populations evolving without hosts were obtained from nematode-free bacterial lawns with a loop. Populations of S. aureus were grown as previously described before freezing and storage at –80°C. Twenty-five percent of each frozen culture was used as inoculum for the next round of experimental evolution. Preliminary experiments determined colony-forming units per milliliter dropped by approximately 50% after freezing. To maintain population diversity, the inoculum amount (25%), adjusted for freezing, was chosen to be within the dilution ratio presented in Wahl et al., 2002, which minimizes the chance that rare beneficial mutations are lost. Twelve rounds of passage were completed for 2 S. aureus genotypes × 2 host treatments × 2 antibiotic treatments × 6 replicates = 48 populations. Whole populations were frozen for subsequent analyses. We conducted all assays to evaluate changes in evolved S. aureus at the population level instead of single isolates to represent the context under which S. aureus evolved in our experiment and generally how pathogens exist in nature and in hosts (Cordero et al., 2012; Launay et al., 2021; McAdam et al., 2011; Mei et al., 2021; Paterson et al., 2015). Population-level assays also allow for potential interactions between genotypes that give rise to phenotypes such as virulence (Ruiz-Bedoya et al., 2023) and antibiotic resistance (Azimi et al., 2020; Cordero et al., 2012) that would otherwise not be captured in single-isolate assays.

Nematodes were age-synchronized as in Penley and Morran, 2018, and grown on E. coli OP50 until L4 larval stage (48 hr at 20°C). Nematodes were subsequently washed off, counted, and approximately 200 individuals were plated on S. aureus lawns on BHI plates. Another set of nematodes were plated on E. coli OP50 to estimate the total number of nematodes. Plates of S. aureus were prepared 24 hr prior by plating 200 µL of an overnight population culture on BHI agar and grown at 28°C. After 2 days at 20°C, plates were scored by counting live nematodes. Host mortality data were analyzed using a generalized linear model with a binomial distribution and followed by Tukey’s multiple-comparison tests to determine pairwise differences. To compare variances across treatments, we conducted Levene’s test for homogeneity of variance. All statistical analyses, including those below, were conducted in R (version 4.2.0).

Populations of S. aureus were streaked onto Trypticase Soy Agar II with 5% sheep’s blood plates and incubated overnight at 37°C. Bacteria were assessed as either hemolysis positive or negative depending on whether there was a complete clearing around the colonies. We did not cross-streak with RN4220, which produces beta-hemolysin, and allows for detection of delta-hemolysin. Therefore, our assay mainly detects alpha-toxin and phenol-soluble modulins. Hemolysis data were analyzed using a generalized linear model with a binomial distribution. To determine whether there was an association between hemolysis status and host mortality, we compared the host mortality means of hemolysis-positive vs. hemolysis-negative pathogens using a Kruskal-Wallis test.

To determine whether there was variation in hemolysis ability within populations, we streaked 10 colonies from each of the 48 evolved populations onto Trypticase Soy Agar II with 5% sheep’s blood plates and incubated overnight at 37°C. To compare the number of isolates able to hemolyze, we conducted a generalized linear model with a binomial distribution. To compare variances, we conducted Levene’s test for homogeneity of variance.

We measured growth in BHI to simulate conditions under which pathogens were grown in our experiment. Rich media allowed us to determine growth when conditions are optimal for evolved pathogens. Measuring growth in minimal media, while being more similar to conditions that S. aureus may face in its environment, would introduce another selective pressure. Overnight cultures of each evolved population were diluted 1:1000 in BHI broth, then 200 µL of each dilution was added to 96-well plates, covered with a Breathe-Easy sealing membrane (Sigma) to minimize evaporation. The optical density at 600 nm at 37°C with shaking at 282 cycles per minute was recorded every 10 min for 16 hr (1000 min) using a BioTek Synergy HTX Multimode Reader. We randomly assigned each of the 48 evolved populations to the inner wells of the 96-well plates, with at least three technical replicates per population. Each plate also included technical replicates of each ancestor. Each ancestral strain was assayed with four technical replicates per plate across four plates. These steps were repeated for media with 0.03125 µg/mL oxacillin. We used the R package gcplyr (Blazanin, 2023) to calculate the area under the curve of density, a metric that quantifies the overall bacterial growth. We compared the growth of populations across treatments using the mean area under the curve of the technical replicates for each population. We used a linear model followed by Tukey’s multiple-comparison tests to determine pairwise differences between the ancestors, and linear mixed models for evolved populations grown with or without oxacillin.

Antimicrobial susceptibility testing for oxacillin was done according to Clinical and Laboratory Standards Institute (CLSI) standard protocols for broth microdilution (CLSI, 2013). In brief, populations were streaked from frozen stock and re-streaked the next day. From the second day culture, populations were resuspended in normal saline to a 0.5 McFarland standard. Then, these cultures were diluted 1:20 in normal saline before 10 µL was added to 90 µL of CAMHB+2% NaCl with the appropriate concentration of oxacillin. Plates were incubated at 35°C without shaking and read at 24 hr. For statistical analysis of populations evolved from the MSSA ancestor, we set a threshold such that anything above the ancestral susceptibility level (0.25 µg/mL) was considered a decrease in oxacillin sensitivity. We then compared the proportion of populations exhibiting a decrease in oxacillin sensitivity with those that had not using Fisher’s exact test with false discovery rate-adjusted (FDR) p-values.

To determine whether there was variation in antibiotic susceptibility within evolved populations, we sampled 10 colonies from each of the four populations with the most number of mutations (two from MRSA and two from MSSA), and two additional randomly selected populations. We argued that variation would be the most observable in populations with the most genetic diversity. We then followed the standard protocol for a Kirby-Bauer disk diffusion susceptibility test (Hudzicki, 2009) using oxacillin. Briefly, we inoculated Mueller-Hinton agar plates (Hardy Diagnostics) of overnight cultures with standardized OD₆₀₀ and placed an oxacillin-impregnated disk (HardyDisk AST Oxacillin, OX1) onto each plate. We measured the diameter of the zone of inhibition after incubation at 37°C overnight.

Overnight cultures of each population were diluted 1:40 in BHI broth containing 0.5% glucose, then 100 µL of each dilution was added to the inner wells of 96-well plates (NEST Biotechnology, flat-bottom, tissue culture-treated). Plates were incubated for 24 hr at 20°C, the temperature that nematodes experience during infection. Each well was washed three times with 200 µL of phosphate-buffered saline (PBS) to remove non-adherent cells. We then stained cells with 125 µL of 0.1% crystal violet for 20 min, followed by washing each well twice with 200 µL PBS. We then resuspended the crystal violet in 125 µL of 30% acetic acid. After incubation for 10 min, we measured the optical density at 595 nm using a BioTek Synergy HTX Multimode Reader. We randomly assigned each of the 48 evolved populations to the inner wells of the 96-well plates, with at least three technical replicates per population. Each plate also included technical replicates of each ancestor. We standardized the mean of technical replicates for each population against the respective ancestor within each plate to control for variation across plates. We used mixed linear models followed by Tukey’s multiple-comparison tests to assess pairwise differences between treatments.

Bacterial genome DNA was extracted from whole populations using Wizard Genomic DNA Purification Kit (Promega). Population sequencing allows us to capture the genetic variation within populations, which is critical to understanding how pathogens evolve (Azimi et al., 2020; Ford et al., 2017; Ford et al., 2016; Hoang et al., 2024a; Hoang et al., 2024b; Mahrt et al., 2021). Genome sequencing of these DNA samples was performed at the M-Gen center using a paired-end library preparation method based on Illumina Nextera and sequenced on the NextSeq 550.

The raw data was quality-trimmed and adapters removed using illumina-cleanup (https://github.com/rpetit3/illumina-cleanup; Petit, 2024) or with bactopia (Petit and Read, 2020). Mutations were called based on comparison to the reference USA300 JE2 genome (BioProject: PRJNA224116, BioSample: SAMN06677988, Assembly: GCF_002085525.1) using breseq (Deatherage and Barrick, 2014). The presence of large-scale deletions and other genetic changes were confirmed by analysis of de novo assembled genomes performed by the SKESA assembler software (Souvorov et al., 2018). The ancestral populations were almost identical to the reference genome used except for three mutations that were not found in any evolved populations. Another three mutations in bioD1, cls, and ccdC were present in all populations derived from the MSSA ancestor and likely occurred in the first passage from the frozen culture before experimental evolution was performed. We therefore removed these six mutations from the analysis. Raw sequences have been deposited in the NCBI Sequence Read Archive under the BioProject accession number PRJNA1162848.

The number of mutations (excluding synonymous or intergenic regions) across treatments were analyzed using a generalized linear model with a Poisson distribution. For mutation enrichment analysis, we followed the steps as described previously (Lees et al., 2017). Briefly, we calculated the ‘mutation rate’ for each ancestor by dividing the total number of mutations by the reference genome size. We multiplied the mutation rate by the total length of regulatory genes (Ibarra et al., 2013) with mutations in our study to generate the expected number of mutations for these genes. We then conducted single-tailed Poisson tests to determine whether the actual number of mutations in regulatory genes was greater than expected.

For phenotypes vs. mutations, we considered only mutations occurring at least twice throughout the experiment. For host mortality vs. mutations and growth rate vs. SCCmec and ACME, we conducted a one-sample t-test for each evolved pathogen against its respective ancestor. p-Values were adjusted using the FDR. For hemolysis status vs. mutations and oxacillin MIC vs. mutations, we conducted Fisher’s exact tests with Bonferroni corrections.

For each non-synonymous mutation in each gene, we created the FASTA file of the USA300 JE2 protein product and a cognate file of the mutant protein with the amino acid substitution edited manually. We ran tBLASTn (version 2.12.0) searches of both proteins against 83,383 high-quality S. aureus whole-genome assemblies (Raghuram et al., 2024). We filtered out matches to genomes with less than 95% identity. We screened for cases where mutants had a better match to a genome than the ancestral protein (higher bitscore).

We collected metadata for BioSamples accession linked to each genome from NCBI Entrez using a Python script from Jason Stajich (https://github.com/stajichlab/biosample_metadata copy archived at Stajich, 2026) then manually curated the ‘isolation_source’ column to either blood/systemic infection, skin/nose/throat, general host-association, animal, environment, others, and missing (see Supplementary file 4 for specific terms). For BioSamples matching mutants within each treatment, we compared the proportion of blood/systemic infection vs. skin/nose/throat against the expected proportion (all BioSamples in the dataset) using a chi-square goodness-of-fit test.

We conducted a principal component analysis on the phenotypes quantified in evolved populations using the prcomp function in R for each S. aureus genotype. We performed permutational analysis of variances (PERMANOVA, 999 permutations) to test for differences between treatments using adonis2 of the vegan package (Oksanen et al., 2024) and pairwiseAdonis (Martinez Arbizu, 2020).

We generated a matrix of Euclidean genetic distances (the square root of pairwise differences) using the frequencies of mutations that arose for each S. aureus genotype, from which we constructed phylogenies (Betts et al., 2018; Ford et al., 2017). We used pairwise differences between the ancestor and each evolved population to compare rates of evolution across treatments, and pairwise differences between replicate populations within each treatment to compare the degree of divergence across treatments. We compared rates of evolution for MRSA and MSSA treatments, and degree of divergence for MSSA treatments, using linear models followed by Tukey’s multiple-comparison tests. We compared the degree of divergence for MRSA treatments using a Kruskal-Wallis test followed by Dunn’s post hoc test.