Bacillus subtilis NCIB 3610, E. coli UTI89 and P. aeruginosa PAO1 were used in this study. LB (10 g of tryptone, 5 g of yeast extract and 5 g of NaCl per liter) or solid LB medium supplemented with 1.5% agar were used for the routine growth of all bacteria. Sodium ascorbate was purchased from Sigma–Aldrich.

For growth analysis, an overnight bacterial culture was diluted 1:100 (1 × 10⁷ CFU) in LB medium with 1% glycerol for B. subtilis, plain LB medium for P. aeruginosa and E. coli, with varying concentrations of sodium ascorbate (neutral pH form). The diluted cultures were incubated at 37°C with continuous agitation (200 rpm) and the absorbance of the culture was measured periodically at 600 nm for 9 h with intervals of 1 h. B. subtilis biofilms were formed in LBGM medium (LB medium containing 1% glycerol; 1 mM MnSO₄). 2–5 × 10⁶ CFU/mL of bacterial culture was inoculated into 5 mL of LBGM medium and incubated at 37°C for 24 h without agitation. E. coli and P. aeruginosa biofilms were formed on 24 well plates. 2–5 × 10⁶ CFU/mL of an overnight bacterial culture was inoculated into a 24 well plate containing 2 mL of LB broth and incubated for 24 h without agitation.

For the biofilm analysis, 24 h old biofilms, grown in the presence of various concentrations of sodium ascorbate, were removed and sonicated at 10 W for 30 s to homogenize the biofilm. The homogenized suspension (5 mL) was used to determine the biomass, colony forming units (CFU), polysaccharides, protein and eDNA. Briefly, for the determination of biomass, the homogenized suspension was washed three times (5000 g for 20 min) with sterile water, lyophilized and weighed. For the determination of viability, an aliquot (100 μL) from the homogenized suspension was diluted serially and plated on LB agar plates to count colonies. Water insoluble polysaccharide was extracted from the lyophilized sample by using 1 N sodium hydroxide (300 μL/mg biomass for three times) and quantified by using a phenol-sulfuric acid assay as described previously (Pandit et al., 2011). For protein quantification, biofilms were collected and homogenized in 1 N NaOH (300 μL/mg biomass for three times). The supernatant from the homogenized suspensions were collected after centrifugation (5000 g, 20 min) and protein content was quantified by using the Bradford assay. For quantification of eDNA in the EPS matrix, the filtered supernatant from the homogenized suspension was used. eDNA was extracted by using a DNA extraction kit (Thermo Fisher Scientific) and the quantity was measured using a nanodrop UV-Vis spectrophotometer (NanoDrop 2000, Thermo Scientific).

The effect of sodium ascorbate on B. subtilis biofilms was analyzed by simultaneous labeling of the bacterial cells and the polysaccharides in the biofilm. Briefly, 10 μg/mL of Alexa flour^® 633-labeled wheat germ agglutinin conjugate (absorbance/fluorescence emission maxima 632/647 nm; Molecular Probes Inc., Eugene, OR, United States) and 50 μg/mL of Concanavalin A, Tetramethylrhodamine conjugate (555/580 nm; Molecular Probes) was added to the culture medium during biofilm formation. The toxicity of these fluorescence probe toward the bacterial cells in biofilms was evaluated by comparison of viability and biomass with control samples. After 24 h, the biofilms were exposed to 2.5 μm of SYTO^® 9 green-fluorescent nucleic acid stain (480/500 nm; Molecular Probes) for 30 min. The stained biofilms were transferred to a glass slide and laser scanning confocal microscope imaging of the biofilms was performed using an LSM 710 NLO (Carl Zeiss) equipped with argon-ion and helium neon lasers. Three independent experiments were performed and image stacks from five sites per experiment were collected (n = 15). EPS biovolume was quantified from confocal stacks by COMSTAT (Heydorn et al., 2000). Biovolume is defined as the volume of the biomass (μm³) divided by substratum area (μm²). For the detection of reactive oxygen species (ROS) in biofilm cells, 24 h old B. subtilis biofilms grown with or without presence of vitamin C were stained with DAPI and CellRox deep ROS sensor (Life Technologies) as described previously (Durmus et al., 2013). Briefly, biofilms were stained with 5 μM of CellRox deep red stain for 30 min, washed with sterile water and counter- stained with DAPI for 20 min. The stained biofilms were visualized by fluorescence microscopy. To visualize the live/dead cells in B. subtilis biofilms grown with and without presence of vitamin C, biofilms were stained with 6.0 μM SYTO 9 and 30 μM propidium iodide (LIVE/DEAD BacLight bacterial viability kit L13152, Invitrogen, Molecular Probes, Inc., Eugene, OR, United States). Imaging was performed with a fluorescence microscope (Axio Imager 2. Carl Zeiss, Zena, Germany).

For the biofilm formation assay, 24 h old biofilms of E. coli and P. aeruginosa were rinsed three times with sterile water to remove the loosely adherent bacteria and dried for 30 min at room temperature. Dried biofilms were then stained with 1% crystal violet for 5 min without agitation. All the biofilms were washed at least five times with sterile water to remove the excess stain and dried for 1 h at room temperature. Absolute ethanol (1 mL) was added to the dried stained biofilm and agitated vigorously for 15 min to dissolve the stain. Optical density was measured at 600 nm.

All experiments for MS analysis were carried out in biological triplicates. 24 h biofilms of B. subtilis were collected and centrifuged to obtain a cell pellet. Cell lysis was performed by re-suspending the cell pellets in an SDS lysis buffer containing 4% SDS in 100 mM triethylammonium bicarbonate pH 8.5, 5 mM β-glycerophosphate, 5 mM sodium fluoride, 5 mM sodium orthovanadate and 10 mM ethylenediaminetetraacetic acid, along with a protease inhibitor cocktail (Roche). The cell extracts were boiled at 90°C for 10 min followed by sonication. The cell debris was removed by centrifugation at 13400 rpm for 30 min and the crude protein extracts were cleaned up by chloroform/methanol precipitation. Dried protein pellets were dissolved in denaturation buffer containing 8 M urea in 10 mM Tris-HCl pH 8.0. The protein lysate was reduced with 1 mM dithiothreitol and alkylated with 5.5 mM iodoacetamide in the dark, for 1 h each at room temperature. Proteins were then subjected to overnight digestion with an endoproteinase Trypsin (1:100, w/w; Pierce^TM). The reaction was stopped by acidification with 10% trifluoroacetic acid and stage-tipped before injecting the samples into the mass spectrometer (Ishihama et al., 2006). Samples were analyzed on an Q Exactive mass spectrometer coupled to an Easy-nLC 1200 (both Thermo Fisher Scientific, Inc., Waltham, MA, United States). Chromatographic separation was performed using an in-house constructed pre-column (45 mm × 0.075 mm I.D) and analytical (200 mm × 0.075 mm I.D.) column set up packed with 3 μm Reprosil-Pur C18-AQ particles (Dr. Maisch GmbH, Ammerbuch, Germany). Peptides were injected onto the column with solvent A (0.2% formic acid in water) at a flow rate of 300 nL/min and 500 bars. Peptides were then eluted using a segmented gradient of 7–27% B-solvent (80% acetonitrile with 0.2% formic acid) over 45 min, 27–40% B over 5 min, 40–100% B over 5 min with a final hold at 100% B for 10 min. The mass spectrometer was operated on a data-dependent mode. Survey full-scans for the MS spectra were recorded between 400 and 1600 Thompson at a resolution of 70,000 with a target value of 1e6 charges in the Orbitrap mass analyzer. The top 10 most intense peaks from the survey scans of doubly or multiply charged precursor ions were selected for fragmentation with higher-energy collisional dissociation (HCD) with a target value of 1e5 in the Orbitrap mass analyzer in each scan cycle. Dynamic exclusion was set for 30 s. Triplicate injections (technical replicates) were carried out for each of the samples for label free quantitation (LFQ).

Acquired MS spectra were processed with the MaxQuant software suite (version 1.5.3.30) (Cox et al., 2009), integrated with an Andromeda search engine. Database search was performed against a target-decoy database of B. subtilis 168 downloaded from UniProt (taxonomy ID 1423), containing 4,195 protein entries, and additionally including also 248 commonly observed laboratory contaminant proteins. Endoprotease Trypsin/P was set as the protease with a maximum missed cleavage of two. Carbamidomethylation (Cys) was set as a fixed modification. Label free quantification was enabled with a minimum ratio count of two. A false discovery rate of 1% was applied at the peptide and protein level individually for filtering identifications. Initial mass tolerance was set to 20 ppm. In case of the main search, the peptide mass tolerance of precursor and the fragment ions were set to 4.5 and 20 ppm, respectively. Downstream bioinformatics analysis was performed using Perseus version 1.5.3.2 (Tyanova et al., 2016). Grouping of proteins with similar expression profiles was achieved by hierarchical clustering analysis. Log10 transformation of mean LFQ intensities of proteins was performed for all the tested conditions. Missing values were replaced from the normal distribution via imputation. Hierarchical clustering was performed on Z-score transformed values using Euclidean as a distance measure and Average linkage cluster analysis. Significance B (p ≤ 0.05) was calculated to identify significantly regulated proteins in each of the ascorbate treatment conditions relative to the control.

Log10 LFQ protein intensities were used to generate sodium ascorbate concentration specific models by mapping the protein intensities to the genome-scale metabolic model of B. subtilis (Bs-iYO844) (Oh et al., 2007). Log10 LFQ protein intensities from biological replicates (three replicates for each sodium ascorbate concentration and 2 for the control) were averaged and used to constrain the fluxes in the associated reaction using Gene Inactivity Moderated by Metabolism and Expression (GIMME) (Becker and Palsson, 2008). GIMME was run using 90% of the objective function threshold and 50th percentile of proteins expression level. Since properly constrained reactions do not demonstrate uniform distributions of feasible steady-state fluxes, the range and distribution of feasible metabolic flux for each reaction were determined by using Markov Chain Monte Carlo (MCMC) sampling (Lewis et al., 2012). To do this, a large number of feasible sets of metabolic fluxes were randomly moved within the solution space until they were well mixed, thereby sampling the entire solution space (Lewis et al., 2012). This sampling process yielded a distribution of feasible steady-state fluxes for each reaction. Sampling was done using gpSampler with default settings from the COBRA toolbox (Schellenberger et al., 2011) using Gurobi (Gurobi Optimization, Inc., Houston, TX, United States) and MATLAB^® (The MathWorks Inc., Natick, MA, United States). Averaged sampled predicted flux distributions for each reaction at each sodium ascorbate concentration were compared to those from the control model in order to identify reactions (and their associated genes) that have significantly altered flux rates upon adding sodium ascorbate to B. subtilis.

The data are presented as the mean ± standard deviation. Intergroup differences were estimated by one-way analysis of variance (ANOVA), followed by a post hoc multiple comparison (Tukey) test to compare the multiple means. Differences between values were considered to be statistically significant when the P-value was <0.05.

Five–ten millimeter doses of vitamin C were previously reported to completely exterminate mycobacteria (Vilcheze et al., 2013). By contrast, it has been reported that vitamin C is not bactericidal toward opportunistic pathogens such as E. coli and P. aeruginosa, but it renders them more susceptible to antibiotics and some physical treatments (Kallio et al., 2012; Khameneh et al., 2016; Helgadóttir et al., 2017). To clarify this effect of vitamin C on non-mycobacterial species, we used the opportunistic pathogens E. coli and P. aeruginosa. Since we previously hypothesized that the synergistic effects of vitamin C may be related to biofilms (Helgadóttir et al., 2017), we included also B. subtilis, the model organism for biofilm development (Vlamakis et al., 2013). We exposed these bacterial strains to a concentration range of vitamin C from 10 to 40 mM (sodium ascorbate, neutral pH form), assessing both the survival and growth in liquid media and biofilms. Up to 40 mM vitamin C did not significantly inhibit the planktonic growth of B. subtilis, E. coli, or P. aeruginosa (Supplementary Figure S1). However, in the same concentration range, biofilm formation was impaired for all three species (Figure 1 and Supplementary Figure S2). Since the vitamin C effect was most pronounced on biofilms, we focused on B. subtilis for an in-depth study of the mechanism behind this effect. B. subtilis biofilm is the most robust and easiest to analyze in terms of structure and composition, and the mechanisms leading to its formation are well characterized (Vlamakis et al., 2013). The normal B. subtilis pellicle (control, 0 mM) exhibits wrinkled and folded architecture (Figure 1C). Vitamin C treatment abolished this wrinkled architecture and visibly attenuated the pellicle in a concentration-dependent manner (Figure 1C). This effect was accompanied by a linear decrease in biofilm biomass (Figure 1A). However, the viability of B. subtilis in the pellicle was not significantly affected, suggesting that the reduced biofilm biomass could be due to loss of EPS and not the loss of cells.

Next, we examined in details the content of various biofilm components in response to the same concentration range of vitamin C. All three major components of the biofilm matrix, polysaccharides, proteins and eDNA, were reduced in the presence of vitamin C (Figures 2A–C). The reduction in protein content vs. vitamin C concentration showed a linear regression coefficient of only 0.88, but the trend of exopolysaccharide and DNA reduction was more closely correlated to increasing concentration of vitamin C and the loss of pellicle biomass (Supplementary Figure S3). The polysaccharide content and bacterial bio-volume in the B. subtilis biofilm were examined by laser scanning confocal microscopy, in the same concentration range of vitamin C (Figures 2E–G). EPS account for over 40% of the mass of the B. subtilis biofilm, but their identification and characterization is not complete. Roux et al. (2015) identified that poly n-acetylglucosamine is the major polysaccharide component of the B. subtilis biofilm matrix (Roux et al., 2015). We therefore used fluorescence probes to visualize EPS components: Alexa flour, WGA conjugate, for n-acetylglucosamine and ConA, Tetramethylrhodamine conjugate, as an unspecific EPS binder for proteins and other polysaccharides. We first established that these probes had no effect on biofilm formation and exhibited no toxicity to B. subtilis cells (Supplementary Figure S4). The n-acetylglucosamine (stained in red) occupied a significant part of the bio-volume in the untreated sample (Figure 2G). With increasing concentrations of vitamin C, the overall thickness of the biofilm decreased, and the content of the poly n-acetylglucosamine (red) and the unspecific EPS matrix (blue) decreased as well (Figure 2G). The decreasing pattern of NAG with vitamin C treatment was consistent with the total polysaccharides content observed by colorimetric assay, where significant inhibition was observed with ≥20 mM of concentration. Meanwhile, the bacterial bio-volume remained constant up to 30 mM vitamin C. The decreasing pattern of bacterial bio-volume was not consistent with other results because vitamin C treatment is mainly affecting the ECM production but not the growth and viability of bacteria at lower concentrations as shown in Figure 1. Viability of cells was examined by live/dead staining (Figure 3A) where it was observed that, at low concentration of vitamin C (10 mM) the cell survival rate was similar to that of untreated control biofilms. By contrast, a significant number of cells were dead in biofilms treated with a higher concentration of vitamin C (40 mM). The killing of cells at the higher concentration of vitamin C coincided with higher levels of detectable oxidative stress (Figure 3B). This confirmed that vitamin C effect on biofilms takes place in two stages: at concentrations of up to 30 mM the cell viability is preserved, but there is a loss of the EPS, primarily exopolysaccharides. Above 30 mM vitamin C, the bacterial cells start dying.

Since vitamin C seemed to inhibit B. subtilis biofilm formation in two stages: reduction of EPS components at up to 30 mM (Figure 2), and killing of cells at 30 mM and higher by inducing the oxidative stress (Figures 1B, 3), we performed an in-depth label-free quantitative proteome analysis of vitamin-C treated biofilms in this critical concentration range. A total of 2056 B. subtilis proteins were identified, of which 1373 were quantified (Supplementary Table S1). Three biological replicates showed a high degree of correlation (Pearson correlation coefficient ≥ 0.9) (Supplementary Figure S5). Hierarchical clustering analysis was employed for grouping similar expression profiles of proteins (Figures 4A–C). Differentially regulated proteins grouped in four clusters (Figure 4B). The majority grouped in clusters 1 (expression reduced upon the addition of vitamin C) and 4 (expression enhanced upon the addition of vitamin C). Clusters 2 and 3 showed variation across the range of vitamin C concentrations. All proteins falling in the four different clusters are listed in the Supplementary Table S1. Proteins for which the expression levels were most strongly affected by vitamin C treatment (p ≤ 0.05) were identified by plotting log2 transformed label-free quantification (LFQ) ratios against log10 transformed LFQ intensities (Figure 4C). Around 100 proteins were found in this category. Among these, at lower vitamin C concentrations, many proteins directly involved in exopolysaccharides synthesis, export and biofilm formation were depleted: notably PtkA, SlrR, SpeA, EpsC, EpsD, EpsE, EpsH, EpsI, EpsO, TuaD, Ugd (Kobayashi, 2008; Gerwig et al., 2014; Mijakovic and Deutscher, 2015). This correlated with disrupted expression of the key regulators controlling synthesis or activity of these proteins, such as ComA, RepC, Spo0A, YmdB, KinC, FloT, and PtkA (Lazazzera et al., 1999; Diethmaier et al., 2011; Yepes et al., 2012). By contrast, at higher vitamin C concentrations, oxidative stress associated proteins (Antelmann et al., 2000; Grimaud et al., 2001; Towe et al., 2007) were strongly overexpressed: MsrA, MsrB, MhqA, MhqD, AzoR2, and RocA. Individual roles of these proteins in biofilm production and oxidative stress response are reviewed in detail in the discussion section. It was evident from this dataset that vitamin C treatment provoked a two-stage global rearrangement of the cellular proteome, which correlated well to our previous observations on EPS content and cell viability. At low concentrations of vitamin C, biosynthetic pathways leading to exopolysaccharide synthesis and export were down-regulated, explaining the observed depletion of biofilm EPS. At higher vitamin C concentrations, the cell started expressing proteins to cope with excessive oxidative stress, which correlates to cell death and loss of the bacterial bio-volume in the biofilm. To assess the specific impact of this global proteome adaptation on redistribution of metabolic fluxes, we used the quantitative proteomics data to generate vitamin C concentration-specific genome-scale metabolic models, by mapping the protein intensities to the available B. subtilis model Bs-iYO844 (Oh et al., 2007) (Supplementary Figure S6). This enabled us to identify several metabolic pathways with flux redistribution provoked by vitamin C, which corroborate our findings (Figure 5). Notably, the model guided analysis showed that pyrroline-5-carboxylate dehydrogenase (P5CDH)/RocA had a significantly upregulated flux in the presence of vitamin C (Figure 5), which indicates that the cells are trying to neutralize ROS.