Phenotypic and metabolic profiling of colony morphology variants evolved from Pseudomonas fluorescens biofilms


  • Matthew L. Workentine,

    1. Biofilm Research Group,
    2. Department of Biological Sciences, Faculty of Science, and
    Search for more papers by this author
  • Joe J. Harrison,

    1. Biofilm Research Group,
    2. Department of Biological Sciences, Faculty of Science, and
    Search for more papers by this author
    • Present addresses: Department of Microbiology, University of Washington, 1705 NE Pacific Street, Seattle, WA 98195-7242, USA;

  • Aalim M. Weljie,

    1. Department of Biological Sciences, Faculty of Science, and
    2. Metabolomics Research Center, University of Calgary, 2500 University Dr NW, Calgary, AB, Canada, T2N 1N4.
    Search for more papers by this author
  • Vy A. Tran,

    1. Department of Biological Sciences, Faculty of Science, and
    Search for more papers by this author
  • Pernilla U. Stenroos,

    1. Department of Biological Sciences, Faculty of Science, and
    Search for more papers by this author
    • HydroQual Laboratories Ltd., #4, 6125-12th Street S.E. Calgary, AB, Canada, T2H 2K1;

  • Valentina Tremaroli,

    1. Biofilm Research Group,
    2. Department of Biological Sciences, Faculty of Science, and
    Search for more papers by this author
    • §

      Sahlgrenska Center for Cardiovascular and Metabolic Research/Wallenberg Laboratory and Department of Molecular and Clinical Medicine, University of Gothenburg, S-413 45 Gothenburg, Sweden.

  • Hans J. Vogel,

    1. Department of Biological Sciences, Faculty of Science, and
    2. Metabolomics Research Center, University of Calgary, 2500 University Dr NW, Calgary, AB, Canada, T2N 1N4.
    Search for more papers by this author
  • Howard Ceri,

    1. Biofilm Research Group,
    2. Department of Biological Sciences, Faculty of Science, and
    Search for more papers by this author
  • Raymond J. Turner

    Corresponding author
    1. Biofilm Research Group,
    2. Department of Biological Sciences, Faculty of Science, and
      E-mail; Tel. (+1) 403 220 4308; Fax (+1) 403 289 9311.
    Search for more papers by this author

E-mail; Tel. (+1) 403 220 4308; Fax (+1) 403 289 9311.


Colony morphology variants isolated from natural and laboratory-grown biofilms represent subpopulations of biofilm cells that may be important for multiple aspects of the sessile lifestyle, from surface colonization to stress resistance. There are many genetic and environmental factors that determine the frequency at which colony morphology variants are recovered from biofilms. One of these factors involves an increased selection for variants in biofilms of Pseudomonas species bearing inactivating mutations in the global activator of cyanide biosynthesis/regulator of secondary metabolism (gac/rsm) signal transduction pathway. Here we characterize two distinct colony morphology variants isolated from biofilms of Pseudomonas fluorescens missing the gacS sensor kinase. These variants produced more biofilm cell mass, and in one case, this was likely due to overproduction of the exopolysaccharide cellulose. Nuclear magnetic resonance (NMR) metabolomics revealed distinct metabolic changes for each of the two phenotypic variants, and these changes involved amino acids and metabolites produced through glutathione biochemistry. Some of these metabolites are hypothesized to play a role in redox and metal homeostasis, and corresponding to this, we show that biofilm populations grown from each of these variants had a different ability to survive when exposed to toxic doses of metal ions. These data suggest that colony morphology variants that evolve during growth of P. fluorescens as a biofilm may have distinct metabolic capacities that contribute to their individual abilities to withstand environmental stress.


Under many laboratory conditions, growth of bacteria in biofilms gives rise to phenotypic variants that have altered colony morphology on solid media (Deziel et al., 2001; Boles et al., 2004; Kirisits et al., 2005; Allegrucci and Sauer, 2007; Davies et al., 2007; Starkey et al., 2009). These colony morphology variants can also be isolated from ‘real world’ environments in which microbes are hypothesized to grow as biofilms, such as from the lungs of chronically infected Cystic Fibrosis patients in the case of Pseudomonas aeruginosa (Drenkard and Ausubel, 2002; Häussler et al., 2003) and from the plant rhizosphere in the case of Pseudomonas fluorescens (Sanchez-contreras et al., 2002). Although several colony morphologies have been identified for Pseudomonas species, two have been studied in some detail. First, small colony variants (SCVs or ‘minis’) isolated from P. aeruginosa biofilms are more resistant to antimicrobial treatment than the parental strain, are hyper-adherent and have significantly reduced motility (Davies et al., 2007). Second, laboratory and clinically isolated rugose small colony variants (RSCVs or wrinkly spreaders, WS) are highly resistant to antimicrobials (Drenkard and Ausubel, 2002), and for P. aeruginosa, have physiological changes indicative of adaptation to growth in the lung environment (Starkey et al., 2009). In laboratories, these types of colony morphology variants are recovered at increasing frequencies when biofilms are exposed to stressors such as oxidative agents, antibiotics and metal ions. This suggests that colony morphology variants might have an important role in survival of a biofilm population when it is exposed to environmental stresses (Boles et al., 2004; Davies et al., 2007; Harrison et al., 2007; Boles and Singh, 2008).

One of the genetic pathways linked to the process of phenotypic variation in both P. fluorescens and P. aeruginosa is the global activator of cyanide biosynthesis/regulator of secondary metabolism (gac/rsm) signal transduction system. Found in many γ-proteobacteria, this system regulates many and diverse processes involved in biofilm formation, virulence, biocontrol and motility (Lapouge et al., 2008). Studies with both P. aeruginosa (Davies et al., 2007) and P. fluorescens (van den Broek et al., 2003; 2005a,b; Martínez-Granero et al., 2005; Martinez-Granero et al., 2006) have implicated GacS and GacA in the process of phenotypic variation. Pseudomonas fluorescens is known to contain two site-specific recombinases that target gacS and gacA genes (Martínez-Granero et al., 2005). During rhizosphere growth, biocontrol strains of P. fluorescens accumulate mutations in gacS and gacA (Bull et al., 2001). Data to date suggest mutations in gacA or gacS might be favoured in the rhizosphere, as co-culture of the mutants with the parental strain increase the ability of P. fluorescens to colonize plant tissues. Here, a fitness trade-off might come at the expense of resistance to environmental stressors, as gacS mutants have a decreased capacity to withstand many antimicrobial agents (Davies et al. 2007; and H. Ceri, unpubl. data). Observations that biofilms of P. aeruginosa PA14 missing the gacS sensor kinase produce SCVs at high frequency fit with an idea that a counter balance might be increased selective pressure for secondary mutations (Davies et al., 2007). The aim of the present manuscript was to characterize the metabolism of isogenic colony morphology variants in order to pave the way to future hypothesis-driven studies examining the ecological roles of these phenotypically diverse cell lines.

In the current study we identified and characterized two colony morphology variants that evolved during growth of P. fluorescens as a biofilm: an SCV and a WS. In accordance with previous studies, variants with similar colony morphologies emerged at increased frequency in a strain bearing a markerless deletion of the gacS sensor kinase. In an effort to better understand the nature and function of different colony morphology variants, we pursued a bottom-up approach to examining cell physiology: rather than characterizing genetic changes we used nuclear magnetic resonance (NMR) metabolomics to obtain snapshots of the variant cells' physiological states.

Metabolomic technologies are in their infancy relative to other ‘omic’ approaches and are limited by the sheer number and diversity of small molecular weight compounds in the cell. However, recent progress has allowed several analytical platforms to be developed to study microbial metabolism on a large-scale basis (van der Werf et al., 2007; 2008; Coucheney et al., 2008; Garcia et al., 2008; Tremaroli et al., 2009). Although limited in sensitivity when compared with mass spectrometry, NMR can precisely quantify metabolic compounds in complex metabolite mixtures such as cell extracts (Weljie et al., 2006). Identification can be accomplished by comparing 1H NMR spectra from a metabolite extract to reference spectra for individual metabolites from a database (Weljie et al., 2006). In combination with multivariate analysis, this technique can reveal precise differences between microbiological samples. Previously, we have used this methodology to reveal mechanisms of tellurite (K2TeO3) tolerance and toxicity in a hyper-tolerant mutant of Pseudomonas pseudoalcaligenes (Tremaroli et al., 2009).

Using NMR metabolomics we demonstrate that the morphological variants isolated from P. fluorescens biofilms are metabolically distinct from their ancestral strain and from each other. Most of these metabolic differences could be accounted for by amino acids as well as by metabolites involved in glutathione biochemistry. We found that biofilms of the variants showed different levels of survival when exposed to toxic metal ions and microbial survival could be correlated to the cellular metabolite concentrations of the variants. All together, these findings suggest that colony morphology variants isolated from P. fluorescens biofilms can evolve diverse physiologies and that this can impact the ability of the individual morphotypes to withstand environmental stress.


Phenotype characterization of P. fluorescens colony variants isolated from biofilms

In previous work with P. aeruginosa we were able to demonstrate that a strain missing the gacS sensor kinase could produce colony morphology variants at high frequency when cultured as a biofilm and exposed to certain antibiotics, metal ions or hydrogen peroxide (Davies et al., 2007). Here, it was hypothesized that a similar phenomenon might occur for P. fluorescens. To test this, biofilms of P. fluorescens CHA0 (wild-type) and CHA19 (ΔgacS) (Zuber et al., 2003) were grown in the Calgary Biofilm Device (CBD) and then exposed to a variety of metal ions. When biofilms were disrupted and plated onto agar, two distinct colony morphotypes were observed in the ΔgacS strain that were distinct from the ancestral smooth colony morphotype: (i) a small colony variant (SCV) and (ii) a wrinkly spreader (WS) that was similar to that previously identified by Rainey and Travisano (1998) (Fig. 1A). By plating on LB containing Congo red (CR), the WS phenotype was easily separated from the SCV due to dye binding, which turned the WS colonies red. The WS morphotype also bound calcofluor, a dye that fluoresces when bound to cellulose and cellulose-like polymers (Fig. 1D). This suggests that WS is very similar the wrinkly spreader morphotypes isolated from static microcosms (Rainey and Travisano, 1998; Spiers et al., 2002; Spiers et al., 2003; Spiers and Rainey, 2005; Goymer et al., 2006; Bantinaki et al., 2007). The SCV did not bind calcofluor, neither did the ancestral smooth colony morphotypes of the wild-type nor ΔgacS strains.

Figure 1.

Phenotype of variants isolated from metal exposed biofilms.
A. Colony images of P. fluorescens CHA0, CHA19, small colony variant (SCV) and wrinkly spreader (WS).
B. Swimming motility was assayed; the SCV and WS had significantly reduced swimming motility diameters, checked with a Student's t-test.
C. Swarming motility.
D. Colonies were grown on media containing calcofluor and visualized under UV light.
E. Biofilm formation as assayed by crystal violet staining.

For each metal ion exposure the frequencies at which variants were recovered from CHA19 biofilms are shown in Table 1. Only CHA19 is shown, as under the reported conditions, no variants could be isolated from the CHA0 biofilms. We observed a large deviation in the numbers of variants produced by individual replicates and we could not correlate chemistry of the metal ions to the frequencies at which variants were detected (M.L. Workentine, data not shown). Furthermore, the control culture that was not exposed to metals had a number of variants indicating that metal toxicity was not solely responsible for selecting the variant phenotypes.

Table 1.  Frequency of phenotypic variants produced by P. fluorescensΔgacS strain CHA19.
IonConcentration (mM)No. of survivors (log10 cfu peg−1)Log-survival (log10 cfu peg−1)Frequency of morphological variants (%)
LB (growth control)N/A4.50 ± 0.060.44 ± 0.658
SiO32−55.25 ± 0.72−0.32 ± 0.7260
Ca2+0.56.02 ± 0.74−1.09 ± 0.7446
Rb+254.52 ± 0.440.41 ± 0.4443
F-204.12 ± 0.451.64 ± 0.4542
Cu2+14.76 ± 0.350.17 ± 0.3541
Dy3+2.54.16 ± 0.230.76 ± 0.2338
Ru3+254.88 ± 0.300.05 ± 0.3037
MoO42−2004.02 ± 0.371.74 ± 0.3735
Pb2+35.31 ± 0.741.69 ± 0.5433
SeO32−2.53.24 ± 0.542.09 ± 0.3025
Os4+13.67 ± 0.302.11 ± 0.3922
Ni2+13.66 ± 0.390.71 ± 0.5221
Al3+14.22 ± 0.52−0.38 ± 0.7419
Rh3+0.254.48 ± 0.652.99 ± 0.6014
I-252.77 ± 0.60−0.74 ± 0.4613
Pt4+105.66 ± 0.461.26 ± 0.0612
Ga3+0.54.09 ± 0.451.67 ± 0.2512
Zn2+14.09 ± 0.25−0.20 ± 0.4110
Li+505.12 ± 0.411.67 ± 0.227
K+2004.76 ± 0.611.70 ± 0.186
Ag+0.011.33 ± 0.910.16 ± 0.616
Na+2004.10 ± 0.221.67 ± 0.456
Fe2+12.54.06 ± 0.181.21 ± 0.175
Sr2+254.55 ± 0.17−0.23 ± 0.525
AsO2−2.51.53 ± 1.771.97 ± 0.684
Co2+0.55.16 ± 0.521.79 ± 0.604
Au3+104.20 ± 0.151.86 ± 0.303
Cs+253.14 ± 0.602.21 ± 0.441
TeO32−0.43.90 ± 0.302.19 ± 0.200
Ba2+253.87 ± 0.221.39 ± 0.570
CrO42−33.55 ± 0.442.92 ± 0.780
Y3+13.79 ± 0.681.70 ± 0.520
Ir4+13.54 ± 0.572.95 ± 0.600
Hg2+12.01 ± 0.782.01 ± 0.400
La3+13.23 ± 0.522.07 ± 0.220

When these morphological variants were subcultured we found that they were stable. In some instances, a colony resembling a WS would lose its distinct wrinkly appearance and look more like the SCV. When the WS and SCV were propagated in planktonic culture they maintained their phenotype and both variants produced thick mats on the surface of both shaking (Fig. 1E) and static cultures (data not shown). Interestingly, biofilm cell numbers remained fairly consistent between the wild type, ΔgacS and the variants, despite the robust pellicles formed by the variants on top of liquid cultures. After 24 h of growth in the CBD, the cell counts were 5.6 ± 0.4, 5.3 ± 0.6, 5.4 ± 0.5 and 4.8 ± 0.3 log10 cfu peg−1 for CHA0, CHA19, SCV and WS respectively. The biofilm structure of these variants was compared with the ancestral strains by using confocal scanning laser microscopy (CLSM) to examine biofilms stained with the DNA intercalator acridine orange. In contrast to P. aeruginosa PA14, deletion of gacS did not result in defective surface growth, and biofilms of the SCV were approximately the same thickness as both the wild-type and ΔgacS ancestral strains. In contrast, the WS had increased thickness and more pronounced structural features (Fig. 2). Based on the visible cell mats that the SCV and WS formed in liquid culture (Fig. 1E), it was expected that biofilms grown on the peg of CBD would be considerably thicker; however, since biofilms had similar cell numbers, it is likely that these thick mats are matrix-associated DNA or other extracellular material, such a polysaccharides.

Figure 2.

Confocal laser scanning microscopy (CLSM) of 24 h biofilms of P. fluorescens CHA0 (top left), CHA19 (top right), SCV (bottom left), WS (bottom right). Image dimensions are 238 × 238 µm. Cross-sections through the x- and y-axis are shown next to the averaged image stacks.

A change in the motility of the variants with respect to the parental strains was also observed (Fig. 1B). Both the SCV and WS showed significantly reduced swimming motility as compared with wild-type and ΔgacS. Swarming motility was also assayed; however, swarming was abolished in the ΔgacS strain and was not restored in either of the two variants (Fig. 1C).

Metabolic profiling of P. fluorescens morphological variants

In an effort to better understand the physiological differences of P. fluorescens variants we utilized a novel NMR metabolomics technique to look at the metabolism of the wild-type, ΔgacS, SCV and WS strains. Our goal was to identify patterns of metabolic change that were associated with the altered phenotype and corresponding physiology. We hypothesized that the variants would have a significantly different metabolism from the parental strains. To test this, we grew planktonic cultures of all the samples and harvested the cells using a cold methanol quenching procedure (see Experimental procedures). After this, the metabolites were extracted into methanol and 1H NMR spectra were obtained for each sample, which were used to identify and quantify the metabolites.

The lower limit of detection for NMR metabolomics allows for the quantification of the most abundant intracellular metabolites, and here, 32 metabolites were identified in the 1H NMR spectra, including amino acids, alcohols and organic acids. Multivariate analysis was used to analyse relative metabolite concentrations in samples from the WT, ΔgacS, SCV and WS. The initial analysis was performed using principle component analysis (PCA), which is an unsupervised method that separates the samples based on their differences (principle component). This unsupervised analysis was used primarily to ensure that there were no significant outliers, but also provided an indication that there were clear differences between all the strains (data not shown). Partial least squares discriminate analysis (PLS-DA) is a supervised method that separates the samples based on pre-assigned classes, in this case the four different strains. This analysis showed clear separation in three components indicating that each of the four strains was metabolically distinct from the others (Fig. 3). The R2 value for this model (0.754, Table 2) is quite high for biological samples and the cross-validation statistic, Q2 (0.699, Table 2), is also quite high and indicates this model is representative of true differences in the data (i.e. the model is not being over fit) (Eriksson et al., 2001). To further characterize these differences each strain was compared with the others using orthogonal partial least squares discriminate analysis (OPLS-DA), which minimizes the differences between the replicates and maximizes the differences between two specified groups (scores and loadings plots are published as supporting material on the publisher's website as Figs S1 and S2). Six models were generated and the statistics for these models are shown in Table 2. A good separation can be obtained for each pair-wise comparison (as indicated by the high R2 and Q2 values in Table 2), which tells us that the phenotypic variants have distinct metabolic profiles that are not just different from the CHA0 and CHA19 strains but are different from each other as well.

Figure 3.

Partial least squares discriminate analysis (PLS-DA) scores plot of the metabolite concentrations of P. fluorescence CHA0 (black), CHA19 (red), SCV (green) and WS (blue). Each data point represents a single extract and the position determined as a linear combination of 32 metabolite concentrations obtained from the 1H NMR spectra. The four strains could be separated along three components.

Table 2.  Model statistics for the multivariate analysis of the metabolite concentrations obtained through 1H NMR metabolomics.
DescriptionModel typeComponentsModel statisticsa
  • a. 

    R2 and Q2 values represent the goodness of fit and the fraction of data correctly predicted in model cross-validation respectively.

CHA0, CHA19, SCV and WSPLS-DA30.7540.699
CHA0 and CHA19OPLS1 + 10.6910.900
CHA19 and SCVOPLS1 + 10.6800.964
CHA0 and SCVOPLS1 + 10.7640.935
CHA0 and WSOPLS1 + 10.7220.709
CHA19 and WSOPLS1 + 10.6590.950
SCV and WSOPLS1 + 10.7490.966

In order to get a clear picture of the metabolic profiles in each strain, the variable influence on projection (VIP) was calculated for each metabolite, which indicates quantitatively how much each metabolite contributes to the OPLS model. The VIPs from one model were plotted against another; termed VIP-shared and unique structure (VIP-SUS) (Wiklund et al., 2008) plots and these indicate which variables are important for multiple models and which ones are not. For the interpretation of the data here, a VIP value of 1 or more was defined as being statistically significant (Eriksson et al., 2001). These plots are shown in Fig. 4 and were used to identify metabolic patterns for the wild type, ΔgacS mutant and the phenotypic variants. Therefore we can identify metabolites that are important only in the WS models (Fig. 4A, top right quadrant), metabolites that are important only in the SCV models (Fig. 4B, top right quadrant) and metabolites that would be important in both models. Valine, phenylalanine and glycine are the metabolites important only for the WS models whereas acetate, pyruvate, aspartate, proline and glutamate are the metabolites important for the SCV models. Tryptophan makes very strong contributions to both models. The relative concentrations for each of these metabolites are shown as box plots in Fig. 5. All of the aforementioned metabolites had significantly reduced levels as compared with the other strains. Interestingly, pyruvate, acetate and aspartate all showed a very similar pattern of concentrations; that is they all were lower in SCV and progressively higher in CHA19, WS and CHA0 respectively (Fig. 5C). Similar patterns between multiple metabolites are an indication pathways containing these metabolites may be altered. These patterns are also evident in the loadings plot (Fig. S2) where these metabolites cluster together.

Figure 4.

Variable influence on projection-shared and unique structure plots (VIP-SUS). These plots are used to compare multiple OPLS-DA models to determine if there are similar metabolites important for both models.
A. Models with the WS are compared.
B. Models with SCV are compared.
Metabolites in the upper right quadrant are important for both models. Lines were drawn at a VIP value of 1 and metabolites with a value higher than this were considered significant.

Figure 5.

Box and whisker plots of the important metabolites identified using the OPLS-DA modelling and VIP-SUS plots. The concentrations are determined directly from the NMR spectra. Median values are represented by the line in the middle of the box with the ends showing the 25th and 75th percentiles. Upper and lower ends of the lines represent the maximum and minimum values respectively.

Colony morphology variants have altered metal susceptibility profiles

Several of the metabolites that were identified in the metabolome analysis have been implicated in the response to toxic metal stress. For example, glycine is intricately involved in glutathione metabolism and valine may be involved in reactive oxygen species detoxification (Tweeddale et al., 1999; Tremaroli et al., 2009). Moreover, SCVs from P. aeruginosa are known to be more tolerant to silver and copper cations than their ancestral strains (Davies et al., 2007), and therefore, we chose to test biofilms of P. fluorescens CHA0, CHA19, SCV and WS for their susceptibility to AgNO3, CuSO4 and NiSO4 (Fig. 6). Quantitative evaluation of bacterial cell survival reveals that the ΔgacS mutant had increased susceptibility to metal toxicity relative to the other strains. In contrast, both variants had wild-type susceptibility to NiSO4, the SCV was tolerant to CuSO4 and the WS was tolerant to AgNO3. These findings indicate that biofilms of different morphological variants have different abilities to survive metal toxicity relative to their ancestral parent (the ΔgacS mutant), and suggest that under certain circumstances, certain cell types might have a survival advantage over others.

Figure 6.

Killing curves of P. fluorescens CHA0 (open squares, □), CHA19 (filled squares, inline image), SCV (closed circles, ●) and WS (open circles, ○). Biofilms of each of the strains were exposed to a series of metal concentrations for 4 h followed by viable cell counting. Shown are the log-killing values, which are the number of cells killed following the exposure to metal. Error bars represent standard deviation calculated from four replicates. Average cell counts for the initial unexposed controls were 5.38 ± 0.47, 4.89 ± 0.45, 5.23 ± 0.39, 5.17 ± 0.47 log10 cfu peg−1 for CHA0, CHA19, SCV and WS respectively.

Differences in metal sensitivity correlate to differences in metabolism

In order to determine if there was a connection between the different metabolites and sensitivity to metal toxicity, we included the metal susceptibility data in the metabolic OPLS models as a simple ‘true’ or ‘false’ variable as it relates to the wild-type level. For example, both the ΔgacS and WS strains were more susceptible to CuSO4 than wild type, so these were assigned a value of 1 (or ‘true’). On the other hand, the SCV was no more susceptible to CuSO4 than the wild type and so it was assigned a value of 0 (or ‘false’). This type of statistical analysis has limited power; however, we were able to obtain significant correlations to the metabolic data for sensitivity to CuSO4 and AgNO3. Figure 7 shows the coefficients from the OPLS models for AgNO3 (x-axis) and CuSO4 (y-axis) plotted against each other. If a metabolite was farther from the origin on the x-axis then it correlated higher with AgNO3 sensitivity. In contrast, if a metabolite was farther from the origin on the y-axis, then it correlated with CuSO4 sensitivity. Values far from the origin on both axes correlated with susceptibility to both CuSO4 and AgNO3. Tryptophan, glutathione, methionine, adenosine and glucose correlated with sensitivity for both metals whereas proline was strongly correlated to sensitivity for copper and lactate and NAD+ are more strongly correlated with sensitivity to silver. These findings suggest that the metabolic state of the variants may be one of the factors contributing to an altered ability to survive environmental stress, which in this case was metal toxicity.

Figure 7.

Coefficient plot from OPLS modelling when the metal susceptibility data are included. The farther the coefficient value of each metabolite is from the origin, the more correlation it has to the metal sensitivity. Metal sensitivity was defined relative to wild type (see Results for details).


In the present study, we have used 1H NMR to show that significant changes in metabolism occur in colony morphology variants evolved from laboratory grown biofilms. We found that two colony morphology variants, the SCV and the WS, have distinct metabolic profiles. The most significant metabolic changes were associated with numerous amino acids as well as several other central metabolites. The metabolic changes that we have observed by direct measurement of intracellular metabolite concentrations correlate very well with metabolic changes previously observed in the P. fluorescens large spreader wrinkly spreader (LSWS) morphotype, as determined by proteomics (Knight et al., 2006) and growth in Biolog plates (Maclean et al., 2004). In previous studies, the ability of the LSWS genotype to colonize the air-liquid interface was associated with disadvantaged carbon metabolism (Maclean et al., 2004). Interestingly, proteome changes in the LSWS were not associated with proteins involved in surface mat formation, but rather with amino acid catabolism and their uptake transport proteins (Knight et al., 2006). These same pathways were implicated by Biolog analysis (Maclean et al., 2004). Multiple valine, phenylalanine and glycine degradation proteins were upregulated in the LSWS and here we find that intracellular concentrations of all three of these metabolites are reduced relative to the ancestral strain and contribute significantly to the separation of the WS from the other stains (as indicated by a VIP value > 1, Fig. 4A). Upregulation of degradation pathways could account for the reduced intracellular concentrations of these metabolites observed here.

Changes in proline and glutamate catabolism and transport were also apparent in the LSWS (Maclean et al., 2004; Knight et al., 2006). Here we found that the SCV had significantly decreased levels of both of these metabolites and that they were both important for the multivariate models separating the SCV from the other strains. Our study indicates that the SCV and WS (and LSWS) are significantly different; however, it is conceivable that there would be some metabolic similarities. We observed that tryptophan levels were significantly decreased in both the WS and SCV, which contributed to significant differences to the OPLS-DA models (Fig. 4A and B). Impaired amino acid metabolism has also been identified in P. aeruginosa RSCVs through Biolog analysis (Starkey et al., 2009). Transcriptomics studies reveal, among other things, that metabolic changes are also apparent on the level of gene expression (von Götz et al., 2004; Kirisits et al., 2005; Starkey et al., 2009). Taken together with the data presented in this study, it is clear that global metabolic adaptations are involved in the physiology of colony morphology variants.

The wrinkly spreader colony morphology has been attributed to the overproduction of cellulose or other extracellular polysaccharides (Spiers et al., 2003). Although changes in intracellular glucose (Fig. S2) and pyruvate (Fig. 5) concentrations were detected in our metabolic studies for the SCV and WS, it is not clear from the data how this fits within the context of central carbon metabolism. Many of the metabolites involved in bacterial sugar metabolism cannot be resolved using the 1H NMR metabolomics technique employed here. However, these initial findings are clues that levels of glycolytic intermediates are likely altered in the colony morphology variants relative to the parental strain and such changes may account in part for changes in exopolysaccharide production.

Some of the metabolites we identified here have been implicated in oxidative stress. Although it did not emerge as a statistically significant metabolite for all the WS models, glutathione was important when comparing the WS and CHA0 models (VIP value > 1, Fig. 4A) and a VIP very close to 1 when comparing the WS to CHA19 (Fig. 4A). Furthermore, we found that glutathione was one of the metabolites that correlated to both CuSO4 and AgNO3 sensitivity. Enzymes involved in the oxidative stress response such as glutathione oxidoreductase (gor), glutathione synthetase (gshA) and glutaredoxin (grxA) are important for planktonic cell survival upon exposure to CuSO4 or AgNO3 (Harrison et al., 2009) and this is also observed for biofilms (J.J. Harrison, R.J. Turner and H. Ceri, unpubl. data). Valine has been implicated in the P. pseudoalcaligenes response to K2TeO3 and there is some evidence for the involvement of proline and glutamate in the anti-oxidant response of P. fluorescens (Mailloux et al., 2009). It has been suggested that oxidative stress may be a driving factor in selecting for phenotypic variants in Pseudomonas biofilms (Boles and Singh, 2008) and we have previously shown that P. aeruginosa SCVs are quite resistant to hydrogen peroxide, as compared with the wild type (Davies et al., 2007).

Although it was promising to find that metabolic data could be linked to the different metal sensitivity profiles, a limitation that cannot be resolved from this type of analysis is whether these effects are direct or coincidental. For example, this technique cannot resolve whether the decreased tryptophan leads to decreased CuSO4 and AgNO3 sensitivity or if the metabolic changes that lead to lower tryptophan levels also lead to CuSO4 and AgNO3 sensitivity. Furthermore some of the small differences in biofilm structure may be contributing to the differences in susceptibility. However, we note that different metabolites seem to be important for tolerance to these different metals and this observation is consistent with recent work with Escherichia coli, in which CuSO4 and AgNO3 were found to exert toxicity through different biochemical routes (Harrison et al., 2009). AgNO3 was shown to exert toxicity primarily through thiol-disulfide chemistry, whereas CuSO4 was predominantly involved in generation of ROS (Harrison et al., 2009).

In summary, we have shown that at least two colony morphology variants – the SCV and WS – evolve during the growth of P. fluorescens as a biofilms and we present further evidence that the deletion of the gac/rsm signal transduction system affects the frequency at which these variants are recovered. Metabolomics is an emerging field that holds much promise for the investigation of microbial physiology and using these new technologies we have demonstrated the SCV and WS variants have distinct metabolic states. This approach provides additional evidence that that these variants have altered amino acid and glutathione metabolism when compared with their ancestral strains, which are findings that corroborate the transcriptomic and proteomic results of previous researchers. Susceptibility testing indicated that SCV and WS morphotypes have different sensitivities to metal toxicity, and this could be correlated with the concentrations of specific metabolites within cells. These findings suggest that biofilm growth can select for phenotypes that have varying abilities to withstand different environmental stressors.

Experimental procedures

Bacterial growth

Pseudomonas fluorescens CHA0 and CHA19 (Zuber et al., 2003) were routinely cultured on LB agar or LB media at 30°C. All strains were stored at −80°C in MicroBank™ vials and subcultured no more than twice prior to experimentation. Biofilms were cultured in LB using the CBD (Ceri et al., 1999), with shaking at 150 rpm, at 30°C and approximately 95% relative humidity. The CBD plates were inoculated with a 1:30 dilution of a 1.0 McFarland standard and mature biofilms were sonciated into 1× PBS or 0.9% saline using a water-bath sonicator (VWR) and serially diluted. Dilutions were spread plated (100 µl) onto LB agar (without salt) containing 0.001% Congo red (Goymer et al., 2006). The Congo red allowed the easy identification of the wrinkly spreader morphotype (Goymer et al., 2006). Both CHA0 and CHA19 biofilms typically reached a cell density of 105 cfu peg−1 after 24 h of growth. Strains were checked by PCR for the presence or absence of the gacS gene, both prior to and after growth as a biofilm to ensure that the strains were correct and to verify that observed colony morphotypes were not contaminants.

Molecular biology

All molecular biology procedures were carried out according to standard protocols. To verify the gacS deletion mutant CHA19 and to ensure the phenotypic variants maintained the deletion the region upstream and downstream of gacS was PCR amplified using the forward primer 5′-GGCTTGGGCAGGTAATCGTC-3′ and the reverse primer 5′-GGCTGCTGACCTTTCCCAAC-3′. Purified PCR product was then sent for DNA sequencing with the University of Calgary Core DNA Services. CHA19 and the variants derived from it gave a 750 bp PCR product that matched the upstream and downstream regions of gacS. The wild-type CHA0 strain gave a 3 kb fragment that matched the full-length sequence of gacS from P. fluorescens Pf-5. To check for complementation the plasmid pME3258 containing CHA0 gacS and upstream promoter region and the empty vector pME6010 (Zuber et al., 2003) were introduced to the variants via electroporation. Electroporation was performed as described (Choi et al., 2006).

Phenotype analysis

For crystal violet staining, 3 ml of cultures of P. fluorescens were grown for 17 h in test tubes, while shaking at ∼150 rpm. The liquid cultures were removed with a pipette, leaving cell mats formed on the surface of the culture and biofilms that had formed on the sides of the test tube. This cell mass was stained with 1% crystal violet, washed 2× with water, and then photographed. Calcofluor binding was assessed in manner similar to that previously described (Solano et al., 2002). Briefly, LB agar plates were prepared containing 200 µg ml−1 calcofluor (Florescent Brightener 28, Sigma) onto which 5 µl of overnight culture was spotted and then incubated for 4 days before visualization under UV light on a standard DNA gel box. Motility was assessed as described previously (Davies et al., 2007). Swimming was assayed on LB plates containing 0.3% agar. Swarming was assayed on a minimal media containing 3 g l−1 KH2PO4, 5 g l−1 Na2HPO4, 0.5 g l−1 NaCl, 0.5 g l−1l-glutamate, 2 g l−1 dextrose, 1 mM MgSO4, 0.01 mM CaCl2 and 0.5% agar.


Microscopy was performed according the protocols outlined previously (Harrison et al., 2006). Briefly, biofilms were grown for 24 h on the pegs of the CBD and subsequently stained with acridine orange. The pegs were examined using a Leica DM IRE2 spectral confocal and multiphoton microscope with a Leica TCS SP2 acoustic optica beam splitter (AOBS) (Leica Microsystems). A 63× water immersion objective used for all the imaging and the image capture was performed using Leica Confocal Software Lite (LCS Lite, Leica Microsystems). Acridine orange images were obtained with 476 nm excitation and monitoring of fluorescence in the green region of the spectra. Image stacks were processed using Imaris 6.3.1 (Bitplane) to generate images for publication.

Metal susceptibility testing

Biofilms of the wild type, ΔgacS and the two variants were tested for their susceptibility to several cations. Biofilms were grown to a cell density of 1 × 105 cfu peg−1 (mid-log phase), which was 4.75, 5.0, 7.0 and 5.75 h ± 0.25 h for CHA0, CHA19, SCV and WS respectively. Testing was carried out on according to previously described protocols (Ceri et al., 1999; Harrison et al., 2005a,b). Copper (II) sulfate (CuSO4·5H2O, Fisher Scientific), silver nitrate (AgNO3, Fisher Scientific) and nickel sulfate (NiSO4·6H2O, Sigma-Aldrich) stock solutions were made up in sterile distilled water at 5–1000 times the highest concentration used in the assay. After an initial dilution in LB, the metals were serially diluted twofold across the rows of a microtitre plate. Biofilms were exposed to the metal challenge plate for 4 h followed by disruption into a neutralization solution as previously described (Harrison et al., 2005a). At each metal concentration total cfu peg−1 was determined by plating on LB agar.

Metabolite extraction

Metabolite extraction was carried out as described previously (Tremaroli et al., 2009). Briefly, 40 ml of mid-exponential phase culture was added quickly to 160 ml of 60% methanol in water, cooled to below −40°C in a dry ice ethanol bath. After centrifugation at −20°C cell pellets were washed once with 60% methanol and centrifuged again. Cell pellets were re-suspended in 5 ml of 100% cold methanol and sonicated to lyse the cells. The lysate was pelleted and the supernatant containing the metabolites was transferred to a clean tube. Metabolites were further purified by two water:chloroform extractions and the extracts were dried overnight in a vacuum concentrator.

To prepare the samples for NMR analysis the dried pellets were re-suspended in 700 µl of deuterium oxide (D2O). This re-suspended extract was loaded onto 3 kDa cut-off spin columns (3 K Nanosep, Omega), which had been washed more than five times with 500 µl of water and twice with 500 µl of D2O to remove the glycerol used to conserve the columns. The extract was filtered through the column by centrifugation on a bench centrifuge and the extract was stored at −80°C until analysis.

NMR acquisition and data analysis

Immediately prior to analysis samples stored at −80°C were thawed and 200 µl of a stock solution of the chemical shift standard, 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS), in 0.5 M NaH2PO4 buffer, was added. The pH of the samples was adjusted to ∼7.0 with 1 M NaOH or NaCl and the volume was brought up to 1 ml with D2O. Six hundred microlitres of sample was loaded into a 5 mm NMR tube for analysis. NMR spectra were acquired on a 600 MHz Bruker AVANCE II spectrometer. One-dimensional spectra were acquired using a standard noesy-presaturation experiment from the Bruker pulse program library (noesypr1d), with an initial relaxation delay of 3 s, and acquisition time of 2 s for an overall recycle time of 5 s. A delay of 100 ms was used for the noesy mixing time.

The baseline of all the spectra were manually corrected and the metabolites present were identified and quantified by comparison to a library of standard metabolites using Chenomx NMR Suite 4.6 (Edmonton, Canada) (Weljie et al., 2006). Compound identities were confirmed by comparison with reference spectra of the pure compounds. The concentrations were normalized by dividing each individual metabolite concentration by the sum of all the concentrations within a particular NMR spectra. This was done to eliminate global changes in metabolite concentrations due to small differences in cell growth between replicates. The relative concentrations of the metabolites were then subjected to multivariate analysis using the SIMCA-P statistical software (Umetrics AB, Sweden). Model quality was assessed by means of the goodness of fit parameter (R2) and cross-validation predictive ability (Q2). Ideal models will have R2 and Q2 values near one, although in biological systems this is rarely observed due to their variability. Box and whisker plots were generated using GraphPad Prism 4.02 for Windows (GraphPad Software).


This work was supported through discovery grants from the Natural Sciences and Engineering Research Council (NSERC) of Canada to R.J.T. and H.C. Aspects of this work was also supported by the Canadian Institutes of Health Research (CIHR) to R.J.T. NSERC has also provided a Post-Doctoral Fellowship to J.J.H. and a Postgraduate Scholarship (Doctoral) to M.L.W. M.L.W. and J.J.H. were additionally supported by PhD Studentships from the Alberta Heritage Foundation for Medical Research (AHFMR). H.J.V. is the recipient of a Scientist award from AHFMR. The Metabolomics Research Centre is currently supported by grants from the Alberta Cancer Board, the Alberta Sepsis Network and the University of Calgary. CLSM was made possible through a Canadian Foundation for Innovation (CFI) Bone and Join Disease Network grant to H.C. The authors would also like to thank Ping Zhang for technical assistance with NMR acquisition and metabolite profiling.