Microbial Ecology

, Volume 60, Issue 4, pp 873–884

Determining the Effects of a Spatially Heterogeneous Selection Pressure on Bacterial Population Structure at the Sub-millimetre Scale

Authors

  • Frances R. Slater
    • The Centre for Ecology and Hydrology
    • Pharmaceutical Sciences DivisionKing’s College London
    • The University of Queensland, Advanced Water Management Centre (AWMC)
  • Kenneth D. Bruce
    • Pharmaceutical Sciences DivisionKing’s College London
  • Richard J. Ellis
    • NERC Centre for Population Biology, Division of BiologyImperial College London
    • Molecular Pathogenesis and GeneticsVeterinary Laboratories Agency
  • Andrew K. Lilley
    • The Centre for Ecology and Hydrology
    • Pharmaceutical Sciences DivisionKing’s College London
    • The Centre for Ecology and Hydrology
    • CEH Wallingford
Methods

DOI: 10.1007/s00248-010-9687-5

Cite this article as:
Slater, F.R., Bruce, K.D., Ellis, R.J. et al. Microb Ecol (2010) 60: 873. doi:10.1007/s00248-010-9687-5

Abstract

A key interest of microbial ecology is to understand the role of environmental heterogeneity in shaping bacterial diversity and fitness. However, quantifying relevant selection pressures and their effects is challenging due to the number of parameters that must be considered and the multiple scales over which they act. In the current study, a model system was employed to investigate the effects of a spatially heterogeneous mercuric ion (Hg2+) selection pressure on a population comprising Hg-sensitive and Hg-resistant pseudomonads. The Hg-sensitive bacteria were Pseudomonas fluorescens SBW25::rfp and Hg-resistant bacteria were P. fluorescens SBW25 carrying a gfp-labelled, Hg resistance plasmid. In the absence of Hg, the plasmid confers a considerable fitness cost on the host, with µmax for plasmid-carrying cells relative to plasmid-free cells of only 0.66. Two image analysis techniques were developed to investigate the structure that developed in biofilms about foci of Hg (cellulose fibres imbued with HgCl2). Both techniques indicated selection for the resistant phenotype occurred only in small areas of approximately 178–353 μm (manually defined contour region analysis) or 275–350 μm (daime analysis) from foci. Hg also elicited toxic effects that reduced the growth of both Hg-sensitive and Hg-resistant bacteria up to 250 μm from foci. Selection for the Hg resistance phenotype was therefore highly localised when Hg was spatially heterogeneous. As such, for this model system, we define here the spatial scale over which selection operates. The ability to quantify changes in the strength of selection for particular phenotypes over sub-millimetre scales is useful for understanding the scale over which environmental variables affect bacterial populations.

Introduction

Many natural environments, such as soil and plant root and leaf surfaces, are highly physicochemically variable at the microscale [1]. This heterogeneity is thought to be in large part responsible for the high bacterial diversity associated with these environments, with between 4,000 and 830,000 species estimated per gramme of soil [2, 3]. However, quantifying the effects of environmental heterogeneity on bacterial diversity is inherently problematic due to the complexities involved with measuring multiple different environmental variables, such as nutrient levels, pH, salinity and heavy metals, as well as considering the legacies of historical events and neutral drift [4].

In the face of such complexities, the use of microbial model systems can help to elucidate the role of individual variables on community structure and dynamics [5]. In the current study, the effects of a single, spatially heterogeneous selection pressure, mercuric ions (Hg2+), on the dynamics of Hg-sensitive and Hg-resistant populations of Pseudomonas fluorescens SBW25 (hereafter referred to as SBW25) were investigated. Previous studies of the prevalence of Hg-resistant and Hg-sensitive bacteria have indicated that Hg selection effects in natural environments such as soils may be spatially heterogeneous at the scale relevant to the bacterium, with Hg-sensitive and Hg-resistant bacteria co-existing in even Hg-impacted soils [68] (although this is not always the case [9]). This suggests that there are areas within soil particulates where the strength of Hg selection is not so strong and Hg-sensitive bacteria are afforded a relative degree of protection from its effects.

In the current study, Hg resistance is conferred to the cell by the carriage of pQBR103, a large (425 kb) plasmid that was originally isolated from the phytosphere [10]. pQBR103 carries a mer operon, which encodes a suite of enzymes for intracellular reduction of Hg2+ to Hg0 [11]. This reduces the toxicity of Hg as the Hg0 diffuses from the cell and is removed from the local environment via volatilisation. The toxicity of HgCl2 to pQBR103-free and pQBR103-carrying cells is highly dependent on growth conditions. For example, in an experiment using the bioluminescent strain SBW25::luxCDABE, 50% relative toxicity occurred in 0.7 × R2A broth culture for HgCl2 concentrations of ∼5 μM, but in 0.7 × R2A plate culture it occurred only at ∼17 μM [12]. However, it is known that carriage of pQBR103 allows SBW25 to grow on 0.7 × R2A agar supplemented with 25 µM HgCl2, whereas under these conditions plasmid-free SBW25 cannot grow [12].

Other than Hg resistance, pQBR103 is largely cryptic. Of 478 predicted coding sequences, only 20% had homology with known proteins or functional domains and only those constituting the mer operon (Hg resistance), and the rulAB homologues (UV resistance) were likely to encode host-beneficial phenotypes [11].

pQBR103 confers a considerable fitness cost on its host in the absence of Hg, with the maximum specific growth rate, µmax, of Hg-resistant SBW25 only 0.66 relative to Hg-sensitive SBW25 in well-mixed broth [13]. This burden is also apparent in spatially structured (biofilm) conditions; when Hg-resistant SBW25 and Hg-sensitive SBW25 were co-inoculated onto membrane filters in an approximately 1:1 ratio in the absence of selection, the area occupied by Hg-resistant SBW25 relative to Hg-sensitive SBW25 was only 0.33 after 3 days [13]. In the presence of Hg, however, pQBR103 carriage confers a fitness advantage on SBW25 in both well-mixed and spatially structured conditions [12, 13]. The strength of this benefit is frequency dependent, with Hg-resistant SBW25 receiving greater fitness benefits when rare in a population of mixed Hg-sensitive and Hg-resistant SBW25 [12, 13]. When Hg-resistant SBW25 and Hg-sensitive SBW25 in a biofilm were exposed to a spatially heterogeneous-Hg selection pressure delivered by cellulose fibres imbued with 12,500 µg HgCl2 g−1, there was evidence of increased selection for Hg-resistant SBW25 around the Hg foci [13]. In the current study, a fine-scale analysis of the distance in micrometres over which Hg elicits a selection effect when it is spatially heterogeneous was undertaken. In order to operate at this scale, novel epifluorescence microscopy image analysis methods were developed and validated.

A number of image analysis programmes, such as CMEIAS [14], COMSTAT [15] and daime [16], have been specifically developed to quantify features of microorganisms within complex environments. The analysis software used in the current study, daime, is of particular utility as it possesses a feature for the quantitative assessment of spatial patterns of bacterial communities, which may reflect specific phenotypic adaptations. For example, clustering of two populations may be indicative of a mutualistic relationship whereas repulsion may be indicative of an antagonistic relationship.

daime has been used to quantitatively demonstrate clustering of ammonia-oxidising bacteria and nitrite-oxidising bacteria in nitrifying activated sludge [16, 17]. Two common dental biofilm bacteria, Streptococcus mutans and Veillonella parvula, were also demonstrated to show increased clustering following exposure to the antimicrobial chlorhexidine using daime [18]. These example studies used daime to quantify the spatial relationship of two bacterial populations relative to each other. In assessing the distances over which Hg-resistant bacteria are at a relative fitness advantage around Hg foci, the current study uses daime for a novel purpose: to quantify the spatial patterns of bacterial populations about a single, central feature. In doing so, we define the sub-millimetre spatial scale over which selection operates in this system.

Materials and Methods

Bacterial Strains and Preparation of Filter Membranes

The bacterial strains used in this study and the experimental conditions for preparation of filter membranes have been described previously [13]. Briefly, a spontaneous rifampicin-resistant mutant of SBW25 (sensitive to mercuric ions) was labelled by random insertion of a constitutively expressed red fluorescent protein cassette (rfp) into chromosomal DNA. SBW25 carrying the plasmid pQBR103 (resistant to mercuric ions) was labelled with a constitutively expressed green fluorescent protein cassette (gfp) on the plasmid. These two reporter strains have been used previously for non-disruptive, in situ study of the effects of plasmid carriage on community dynamics using fluorescence microscopy [12, 13]. Polycarbonate filter membranes (0.2-µm pore size, 25-mm diameter; Millipore, Watford, Herts, UK) were inoculated with approximately 105 cells in a ratio of 1:1 (vol/vol) Hg resistant to Hg sensitive. Membranes were each sprayed with 20 mg of carboxymethyl cellulose fibres (25 to 60 µm; Sigma-Aldrich, Gillingham, Dorset, UK) imbued with either sterile distilled water (SDW; no-Hg treatment) or 12,500 μg HgCl2 g–1 cellulose (to produce spatially heterogeneous foci of the Hg selection pressure; hereafter referred to as heterogeneous-Hg treatment) and then dried and incubated on 0.7 × R2A agar (Oxoid, Basingstoke, Hants, UK) at room temperature (21 to 23°C) for 3 days. Each treatment was replicated five times over two different experimental runs carried out on at different times, with mixed Hg-resistant and Hg-sensitive inoculums prepared and applied to filter membranes on each occasion. The 2-D spatial distribution of Hg-sensitive and Hg-resistant bacteria on membranes was analysed using epifluorescence microscopy using an Eclipse E600 microscope fitted with a 110-W Hg lamp as previously described [13]. Briefly, GFP was visualised by excitation with a 465–495-nm source via a dichroic mirror and barrier filter with pass band centres of 505 and 515–555 nm, respectively. RFP was visualised by excitation with a 510–560-nm source via a dichroic mirror and barrier filter with pass band centres of 575 and 590 nm, respectively. For each membrane, images of at least ten fields of view (1.07 mm2, equivalent to 0.22% of the total membrane area) were captured such that, when a cellulose fibre was at the centre of the field of view, no other fibres could be seen. The density of fibres resulting from a 20-mg spray ensured that this was most often the case; in instances where one or more fibres other than the central fibre could be seen, the image was rejected. The mean (±standard error) total number of cells on membranes after 3 days of incubation for one experimental run was found to be 1.05 × 1010 (±2.34 × 109) [13]. As a field of view represented 0.22% of the total membrane area and contained 1.37 × 106 pixels, it may be assumed that there was an average of approximately 16.80 (±3.74) cells per pixel. For the two types of analysis described below, fluorescence (and by inference, community composition) was assessed at the level of the pixel and, therefore, the analysis in the current study is at the scale of tens of cells rather than of the individual cell.

Confocal Microscopy

The 3-D spatial distribution of Hg-sensitive (RFP) and Hg-resistant (GFP) bacteria on filter membranes was analysed by confocal microscopy using an UltraView ERS confocal dual spinning-disc microscope (Perkin Elmer, Waltham, MA, USA). GFP was visualised by excitation with a 488-nm laser source via a four-wavelength dichroic and emission discrimination filter with a 527-nm pass band centre. RFP was visualised by excitation with a 568-nm laser source via a four-wavelength dichroic and emission discrimination filter with a 615-nm pass band centre. The lenses were Plan-Apochromat 20×/0.75 (air) and Plan Fluor 40×/0.75 (air). A stack of 40 images in the xy plane, of fields of view (1,000 µm × 850 µm, equivalent to 0.17% of total membrane area) at increments of 0.2 µm in the z dimension, were captured using an Electron multiplier CCD cooled digital camera controlled by UltraView imaging software (Perkin Elmer, Waltham, MA, USA). 3-D projection images were rendered using Improvision Velocity Software (version 3; Perkin Elmer, Waltham, MA, USA).

Spatial Arrangement Analysis Using Manually Defined Contour Regions

Ten images of fields of view from 2-D epifluorescence microscopy analysis from each of two membranes subjected to either the no-Hg or heterogeneous-Hg treatment (20 mg of 12,500 μg HgCl2 g−1-treated cellulose; Supplementary Information Figure S1) were each cropped using Adobe Photoshop CS2 (Adobe Systems Inc., San Jose, CA, USA) to generate six, non-overlapping, concentric contour regions at about (a) 1–100 pixels (∼0–88 μm), (b) 101–200 pixels (∼89–177 μm), (c) 201–300 pixels (∼178–265 μm), (d) 301–400 pixels (∼266–353 μm), (e) 401–500 pixels (∼ 354–442 μm) and (f) 501–600 pixels (∼443–529 μm) from the perimeter of the cellulose fibre (e.g. Supplementary Information Figure S2). The perimeter of the cellulose fibre was first defined manually with the aid of the lasso tool in Adobe Photoshop CS2 and then serially expanded in increments of 100 pixels to give seven contours. Neighbouring pairs of contours were then used to generate the six 100-pixel-wide contour regions at increasing distances from the perimeter of the cellulose fibre.

The area of each contour region occupied by Hg-resistant bacteria, \( {A_{H{g^r}}} \), Hg-sensitive bacteria, \( {A_{H{g^s}}} \) and any areas with very little bacterial cover (seen as black) were measured using Simple PCI 6 imaging software (Compix Inc. Imaging Systems, Sewickley, PA, USA) as previously described [13]. The ratio of \( {A_{H{g^r}}} \) to \( {A_{H{g^s}}} \) in each contour region, \( {a_{H{g^r}}} \), was then calculated.

Spatial Arrangement Analysis Using daime: Method Development

The algorithm implemented in daime is one adapted after Reed and Howard [19], whereby the covariance of two object populations is assessed over a range of distances [16]. Essentially, a number of isotropic, random dipoles of length r units are extended in a semicircle from each pixel into the image space. Dipoles with both ends touching objects from different populations are recorded as “hits” whilst all other dipoles, i.e. those with one or both ends touching empty space or both ends touching objects from the same population, are recorded as “misses”. Dipoles that extend beyond the image space are not recorded. An estimate of the covariance of the two populations, P(r), is then the proportion of all recorded dipoles that are hits. If the two populations are randomly distributed, then P(r) is dependent solely on the densities of the two populations. In daime, P(r) is normalised with the densities of the two populations to give the estimated pair cross-correlation function, g(r). Values of g(r) will therefore approach unity when populations are randomly distributed, even if the populations are at different densities. Deviations from unity are indicative of nonrandom distributions of the populations, with g(r) greater than unity indicating clustering, and g(r) less than unity indicating repulsion, at any given distance r.

As discussed in the “Introduction”, daime has previously been used for quantitative assessments of the spatial relationships of two bacterial populations relative to each other. As the intention in the current study was to assess the spatial relationships of two bacterial populations with a single, central feature, an initial assessment of the utility of daime for this purpose was undertaken. Control images of two populations of randomly oriented, different coloured objects were generated. Fluoresbrite® Yellow Green (hereafter referred to as green) and Yellow Orange (hereafter referred to as red) Carboxylate Microspheres (6.00 µm diameter; Polysciences Inc., Warrington, PA, USA) were each diluted by dispensing one drop into 0.5 mL of SDW. These suspensions were mixed in a 1:1 (vol/vol) ratio, and a 20-µL volume of the resultant mixture was placed between a glass microscope slide and coverslip. The microspheres were then imaged by epifluorescence microscopy, and images of ten fields of view chosen at random were captured as previously described [13]. A white rectangle (to represent a cellulose fibre equivalent to 74 × 57 µm or 1/16th of total field of view area) was generated using Adobe Illustrator CS (Adobe Systems Inc., San Jose, CA, USA) and superimposed in the centre of each microsphere image. Images were pre-processed in daime (version 1.2) to produce three monochrome component images of green microspheres, red microspheres and the central feature. Firstly, each image was automatically split into red, green and blue channels in daime. Operations to further differentiate between groups of objects, such that in each component image there were green microspheres, red microspheres or the central feature only, were performed by applying the image calculator function (with the green channel as series 2 when the red channel was series 1 and vice versa; daime user manual: http://www.microbial-ecology.net/daime/daime-manual.asp). Images were converted to object masks and segmented using default settings. The spatial arrangement analysis (two populations) function was applied to firstly, the green microspheres and red microspheres and, secondly, the red microspheres and central feature, again using default settings.

Spatial Arrangement Analysis Using daime: Experimental Images

The ten field-of-view images from each of the two membranes that were analysed using the manually defined contour region analysis were also analysed using daime. All images were pre-processed in the same way as for the Fluoresbrite® microsphere images to each generate three-component images of Hg-resistant bacteria, Hg-sensitive bacteria or the cellulose fibre only. The spatial arrangement analysis (two populations) function was then applied to, firstly, the Hg-resistant component image and cellulose component image and, secondly, the Hg-sensitive component image and cellulose component image, again using default settings.

Statistical Analysis

Statistical tests of data from manually defined contour regions and whole fields of view, including the F-test for homogeneity of variances, the Anderson–Darling test for normality of error and general linear model procedures, were calculated using Minitab software (version 15; Minitab Ltd., Coventry, Warks, UK). The response variable for all tests was the ratio of \( {A_{H{g^r}}} \) to \( {A_{H{g^s}}} \), \( {a_{H{g^r}}} \), in either contour regions or whole fields of view. Data points where values of \( {A_{H{g^r}}} \) and/or \( {A_{H{g^s}}} \) were equal to 0 (limited to five contour regions with very little bacterial cover at 0–88 μm from the cellulose fibres in the heterogeneous-Hg treatment images) were excluded from the analysis. Response variable data were log10-transformed and, where necessary, data points with large standardised residuals were excluded from the analysis to ensure homogeneity of variances and normality of error. Removal of data points with large standardised residuals was only necessary in one instance: when testing for a difference in \( {a_{H{g^r}}} \) between no-Hg and heterogeneous-Hg treatments (categorical) for data from whole fields of view from all membranes, i.e. five membranes per treatment over two experimental runs. In this test, run (random, categorical) and membrane replicate (random, categorical, nested within Hg treatment) were also included as variables because significant differences had been found previously [13]. Of 132 data points, 16 were removed. The effect of Hg treatment was similarly significant before and after removal of data (see “Results” section for after-removal F and p values), but the effect of membrane replicate changed from not significant before removal (F4, 126 = 1.25, p = 0.292) to significant after removal (F4, 110 = 4.76, p = 0.001). Run was found to have a similarly significant effect both before and after removal of data (after removal: F1, 110 = 74.29, p < 0.001). Statistical tests for differences in \( {a_{H{g^r}}} \) between the heterogeneous-Hg and no-Hg treatment (categorical) were performed separately for contour regions at different distances from the cellulose fibre. A statistical test for a difference in \( {a_{H{g^r}}} \) between the heterogeneous-Hg and no-Hg treatment (categorical) was performed for data from whole fields of view from the two membranes used to generate contour regions. A statistical test for a difference in \( {a_{H{g^r}}} \) between the heterogeneous-Hg and no-Hg treatment (categorical) was also performed for data from the whole fields of view from all membranes subjected to heterogeneous-Hg and no-Hg treatments, i.e. five membranes per treatment over two experimental runs. As significant differences between experimental runs due to the small differences in starting ratios of Hg-resistant-to-Hg-sensitive bacteria had been found previously [13], run (random, categorical) was included as a variable. As significant differences between membrane replicate had also been found previously [13], membrane (random, categorical, nested within Hg treatment) was also included as a variable. Both run and membrane replicate were found to have significant effects on \( {a_{H{g^r}}} \) (run: F1, 110 = 74.29, p < 0.001; membrane: F4, 110 = 4.76, p = 0.001) but the effect of Hg treatment remained significant when these effects had been taken into account.

The spatial arrangement of Hg-resistant and Hg-sensitive bacteria about cellulose fibres was assessed by calculating the mean estimated pair cross-correlation function, g(r), over distances r of 0–550 μm of ten images each from heterogeneous-Hg and no-Hg treatment membranes using daime. Values of g(r) at distances r greater than 550 μm fluctuated widely and so were not considered. Statistical support for g(r) is weaker at further distances as estimation of g(r) is from the fewer dipoles of longer length r for which both ends remain within the bounds of the image space [16, 19].

Results

Confirmation of Appropriateness of Area Comparisons Between Hg-Resistant and Hg-Sensitive Regions

Inspection of the surfaces of filter membranes using confocal microscopy revealed 3-D features of the biofilms that had developed over 3 days of incubation. Biofilms had an approximate thickness of 0–40 µm with considerable variation in topography; depth increased in areas proximal to cellulose fibres and decreased in areas proximal to areas with very little bacterial cover (Supplementary Information Figure S3a, b and c). N.B. it is possible that there were in fact some bacterial cells on those parts of the membrane that appear black but that the microscopic magnification employed in the current study was not sufficient to detect them. However, these cells would be few and insignificant in the context of the large numbers of cells in the visible biofilm. Bacteria colonising the membrane formed clearly delineated, contiguous areas of Hg-resistant and Hg-sensitive cells, with strong vertical boundaries between the two cell types at all depths of the lawn. N.B. again, it is possible that there was in fact a small amount of mixing of the two cell types at these boundaries that the microscopic magnification was not sufficient to detect. However, again, these cells would be few and would not significantly affect the calculations of \( {A_{H{g^r}}} \) and \( {A_{H{g^s}}} \). Therefore, calculation of the ratio of \( {A_{H{g^r}}} \) to \( {A_{H{g^s}}} \), \( {a_{H{g^r}}} \), from 2-D epifluorescence microscopy images was considered a reasonable approximation of the ratio of the number of Hg-resistant bacteria relative to Hg-sensitive bacteria in the biofilm. Furthermore, previous work has demonstrated that, for membranes exposed to spatially uniform Hg selection, trends in Hg-sensitive to Hg-resistant cell number ratios closely match those of \( {a_{H{g^r}}} \) [13]. However, there were small areas immediately proximal to cellulose fibres that comprised mixed Hg-resistant and Hg-sensitive bacteria rather than the clearly delineated, contiguous areas of Hg-resistant and Hg-sensitive bacteria that covered the rest of the membrane (Supplementary Information Figure S3d). These areas were visible as a bright orange-red “halo” of cells around cellulose fibres in epifluorescence microscopy images. This phenomenon was apparent for both heterogeneous-Hg and no-Hg treatments and was therefore attributed to the presence of the cellulose fibre itself rather than any effects of the Hg. Cellulose fibres exhibit hydrophilic behaviour, with highly polar surfaces due to the presence of hydroxyl groups [20]. An excess of water may have formed a meniscus around cellulose fibres, resulting in areas of free-swimming cells and consequently a loss of integrity of biofilm structure.

Manually Defined Contour Region Analysis Demonstrated Changes in Structure About Hg Foci

In all contours of the no-Hg treatment, Hg-sensitive bacteria dominated the community (Fig. 1a). The structure was broadly similar between all contour regions at different distances from cellulose fibres, with the exception of the contour region closest to the cellulose fibre (0–88 μm). The five contour regions furthest from the fibre were occupied by bacteria in ratios of between 3:1 and 4:1 Hg-sensitive to Hg-resistant bacteria, which is in good agreement with results from analyses of whole fields of view from no-Hg treatments in a previous study [13]. The marked difference in the areas occupied by the two strains is a result of the considerable burden of plasmid carriage incurred by the Hg-resistant bacteria, as discussed in the “Introduction”. The structure in the contour region closest to the cellulose fibre (0–88 μm) was slightly different, with a ratio of over 5:1 Hg-sensitive to Hg-resistant bacteria. This was likely due to an overestimation of \( {A_{H{g^s}}} \) in the closest region. It was not possible to discriminate between the orange-red halo comprising both Hg-sensitive and Hg-resistant bacteria and the red areas comprising Hg-sensitive bacteria only using the automated segmentation feature of Simple PCI image analysis software. It has been noted previously that automated segmentation features in image analysis programmes are prone to this lack of accuracy [21], although this will likely be improved in future releases of software. A slight, systematic overestimation of the proportional area of Hg-sensitive bacteria, \( {A_{H{g^s}}} \) (and therefore a slight underestimation of \( {a_{H{g^r}}} \)) was therefore likely in contour regions closest to the cellulose fibre where the halo region occurred. At all distances, the contour regions were almost completely occupied by bacteria, with mean areas with very little bacterial cover of less than 2%.
https://static-content.springer.com/image/art%3A10.1007%2Fs00248-010-9687-5/MediaObjects/248_2010_9687_Fig1_HTML.gif
Figure 1

Community structure differed according to proximity to Hg foci. Areas of manually defined contour regions occupied by Hg-resistant bacteria (hatched) and Hg-sensitive bacteria (black) and areas with very little bacterial cover (open) were broadly similar at all distances from no-Hg cellulose fibres (a) but differed at close proximities to Hg cellulose fibres (b). Error bars represent SE of the mean from ten epifluorescence microscopy images. *Contour regions in which values of \( {a_{H{g^r}}} \)were significantly different between no-Hg and heterogeneous-Hg treatments at the 5% level (see Table 1)

For the heterogeneous-Hg treatment, there was greater variability in the structure between contour regions at different distances from cellulose fibres (Fig. 1b). As for the no-Hg treatment, Hg-sensitive bacteria dominated the community in all contours. This implies that, even when Hg is present to select for Hg-resistant bacteria, the burden imposed by carriage of the plasmid is still high. However, there was an increase in \( {a_{H{g^r}}} \) for the heterogeneous-Hg treatment relative to the no-Hg treatment in two of the contour regions, at distances of 178–265 and 266–353 μm from the cellulose fibre (Table 1), indicative of increased fitness of Hg-resistant bacteria. There was no significant difference in \( {a_{H{g^r}}} \) for the heterogeneous-Hg treatment relative to the no-Hg treatment in the contour regions at 89–177, 354–442 and 443 – 529 μm from the cellulose fibre. A test for a difference between the no-Hg and heterogeneous-Hg treatment could not be performed for the closest contour region at 0–88 μm from the cellulose fibre. This was as a result of five data points being excluded from the analysis due to very little bacterial cover which left a small dataset of unequal sample size and also because the variances of the heterogeneous-Hg and no-Hg data were not homogeneous (F-test statistic = 0.02, p = 0.041).
Table 1

Statistical analysis of the effects of distance from Hg foci on bacterial growth

Distance from cellulose (µm)

Mean (±SE) \( {a_{H{g^r}}} \) (n = 10)

F1, 18-statistic

p value

No-Hg

Hg

89–177

0.32 (0.05)

0.36 (0.08)

0.01

0.944

178–265

0.27 (0.04)

0.49 (0.09)

5.72

0.028b

266–353

0.32 (0.03)

0.45 (0.04)

6.33

0.022b

354–442

0.31 (0.03)

0.38 (0.05)

1.37

0.258

443–529

0.29 (0.02)

0.38 (0.04)

4.07

0.059

Whole fields of view

0.27 (0.02)

0.35 (0.04)

3.19

0.091

Whole fields of view from all membranes

0.31 (0.01)

0.50 (0.03)

19.89

0.011b

n = 65

n = 52

F1, 110a

There were significant differences in the ratio of \( {A_{H{g^r}}} \) to \( {A_{H{g^s}}} \), \( {a_{H{g^r}}} \), between the no-Hg and heterogeneous-Hg treatment membranes analysed using contour region analysis at distances of 178–265 and 266–353 µm from cellulose fibres. These differences were not reflected in comparisons of \( {a_{H{g^r}}} \) from whole fields of view from these two membranes but were reflected in comparisons of \( {a_{H{g^r}}} \) from whole fields of view from all membranes subjected to these treatments. Response variable data were log10-transformed to ensure normality of error. Data for the closest contour region at 0–88 μm from the cellulose fibre are not shown due to inaccurate data resulting from the halo effect as described in the text

aSixteen data points excluded from analysis

bSignificant at the 5% level

Greater variability in structure in the heterogeneous-Hg treatment than in the no-Hg treatment was further demonstrated by changes in the area with very little bacterial cover between contour regions. The size of these areas decreased with increasing distance from cellulose fibres, with a mean (±SE) of 57.92 (±15.20)%, 17.33 (±17.33)% and 4.16 (±2.92)% of contour region area at distances of 0–88, 89–177 and 178–265 μm, respectively. These data demonstrate that there was an Hg gradient produced around the cellulose fibres in the heterogeneous-Hg treatment membranes, with higher levels of Hg at closer proximities to the cellulose fibre.

When the same ten images per membrane were not cropped to produce contour regions but instead were analysed as whole fields of view of 1.07 mm2, there was no significant difference in values of \( {a_{H{g^r}}} \) between the heterogeneous-Hg and no-Hg treatments (Table 1). However, when whole fields of view from all membranes subjected to 0 and 12,500 µg Hg g−1 cellulose treatments, i.e. five membranes per treatment over two experimental runs (see [13]), were compared, there was a significant difference in values of \( {a_{H{g^r}}} \) between the heterogeneous-Hg and no-Hg treatments (Table 1). Therefore, \( {a_{H{g^r}}} \) values for heterogeneous-Hg treatment membranes were, on average, significantly higher than \( {a_{H{g^r}}} \) values for no-Hg membranes at the scale of the field of view. Whilst the selection elicited by Hg on the particular heterogeneous-Hg treatment membrane chosen for contour region analysis may have been comparatively weak and therefore not apparent at the scale of the fields of view, there was still evidence of a relative fitness benefit for Hg-resistant bacteria at the scale of the contour region.

Spatial Arrangement Analysis of Bacterial Populations About Single, Central Features Was Possible Using daime

In the current study, daime was adapted to assess the spatial relationships of a bacterial population about a single feature rather than the spatial relationships of two bacterial populations relative to each other (Fig. 2). We confirmed that this was possible using images of two differently coloured microspheres (to represent the bacterial populations) superimposed with a single, central oblong feature (to represent the cellulose fibre). The green and red microspheres were expected to have a totally random arrangement with respect to each other (Fig. 3a). This was found to be true at most distances r (15–550 µm) because daime analysis returned values of g(r) that approached unity (Fig. 3b). However, microspheres sometimes clumped together in pairs or threes (either all red, all green or a mixture of red and green). The clumping of red and green microspheres meant that, at short distances r, the two populations were not randomly arranged with respect to each other, resulting in higher values of g(r) from 0 to 15 µm. The red microspheres about the superimposed, central feature had an altogether more non-random spatial arrangement (e.g. Fig. 3c), with a general increase in mean pair cross-correlation, g(r), with increasing distance r (Fig. 3d). This deviation from unity was a result of the central position of the superimposed feature. In daime, the covariance of the two populations, P(r), is the proportion of dipoles of length r extended into the image space that are “hits” [16]. In the red microsphere-central feature images, a hit would have been recorded when one end of a dipole was touching the central feature and the other end was touching a red microsphere. As the feature was positioned centrally in the image and dipoles are extended isotropically in a semicircle, more dipoles with one end touching the feature could be extended into the image space at increasing distance r. Assuming the red microspheres were randomly distributed throughout the image, there would therefore have been an increase in the number of these dipoles that touched a red microsphere and were recorded as hits at greater distance r. The proportion of all dipoles that are hits, P(r), is normalised with the densities of the two object populations in the image to give the estimated cross-correlation function, g(r), which should then approach unity for randomly distributed populations [16]. However, the normalisation does not take into account the inevitable increase in P(r) with increasing r that occurs when one population is a centrally positioned feature. A g(r) approaching unity is therefore not indicative of a population that is randomly distributed around a centrally positioned feature. However, it was still possible to use daime to assess the spatial arrangement of a population of objects around a centrally positioned feature by redefining the model of g(r) that corresponds to a random spatial arrangement. A linear model of the form
$$ g(r) = g{(r)_0} + mr, $$
where g(r)0 is the intercept and m is the slope, was fitted to the mean g(r) as a function of r from spatial arrangement analyses of ten images of red microspheres about centrally positioned features using SigmaPlot software (version 10; Systat Software, Inc., USA). The fit returned a small positive slope value, m = 2.65 × 10−3, an intercept, g(r)0 = 0.70, and a goodness of fit, R2 = 0.44 (Fig. 3d). All subsequent pair cross-correlation functions, g(r), calculated to assess the spatial arrangement of Hg-resistant and Hg-sensitive bacteria about a centrally positioned cellulose fibre, were compared to this reference line.
https://static-content.springer.com/image/art%3A10.1007%2Fs00248-010-9687-5/MediaObjects/248_2010_9687_Fig2_HTML.gif
Figure 2

A schematic to illustrate the use of daime to quantify the spatial arrangements of two bacterial populations (squares and circles) relative to each other (a) and the spatial arrangement of one bacterial population (in this case, circles) relative to a single, central feature (rounded oblong) (b). Isotropic, random dipoles of length r are extended from each pixel in a semicircle. Dipoles with the two ends touching the different populations (or the population and the central feature) are hits (H), those with both ends touching the same population or the central feature, or with one or both ends touching empty space, are misses (M) and those with one end beyond the border of the image are not recorded (X). See text in “Materials and Methods” for further explanation of how the ratio of hits to misses is used to determine the estimated pair cross-correlation function, g(r)

https://static-content.springer.com/image/art%3A10.1007%2Fs00248-010-9687-5/MediaObjects/248_2010_9687_Fig3_HTML.gif
Figure 3

Method development for spatial arrangement analysis of Hg-resistant and Hg-sensitive bacteria about cellulose fibres using daime. Epifluorescence microscopy image of randomly dispersed, red and green microspheres (both of approximately 6 µm diameter) (a). Spatial arrangement of red and green microspheres relative to each other, showing the mean g(r) values of ten images (irregular line) and a standard reference line given by g(r) = 1 (straight line) (b). Epifluorescence microscopy image of red microspheres and a single, central feature (white) (c). Spatial arrangement of red microspheres relative to the central feature, showing the mean g(r) values of ten images (straight line) and a fitted reference line given by g(r) = 0.70 + (2.65 × 10−3)r (irregular line) (d). Epifluorescence microscopy image of Hg-resistant bacteria (green), Hg-sensitive bacteria (red) and halo region of mixed Hg-resistant and Hg-sensitive bacteria (indicated by white ring) around a cellulose fibre (central) with areas with very little bacterial cover also shown (black) (e). Spatial arrangements of Hg-sensitive bacteria including halo region (irregular grey line) and Hg-sensitive bacteria following removal of halo region (irregular black line) about the cellulose fibre, with a fitted reference line (straight line), given by g(r) = 0.70 + (2.65 × 10−3)r (f). Scale bars on epifluorescence microscopy images represent 100 µm

The Changes in Population Structure About Hg foci that Were Revealed by daime Analysis Were Similar to Those Revealed by Analysis of Manually Defined Contour Regions

As in Simple PCI imaging software, the halo regions of mixed Hg-resistant and Hg-sensitive bacteria (bright orange-red) around cellulose fibres could not be discriminated from regions of Hg-sensitive bacteria only (red) using the automated segmentation features in daime. The spatial arrangement analysis of Hg-sensitive bacteria about the cellulose fibre in one image from the no-Hg treatment (Fig. 3e) was first carried out for Hg-sensitive bacteria including halo bacteria, with the Hg-sensitive component image determined using the automatic segmentation features in daime as described in the methods. It was then also carried out for Hg-sensitive bacteria only, with halo bacteria manually removed from the Hg-sensitive component image using the object editor function (daime user manual; http://www.microbial-ecology.net/daime/daime-manual.asp). The removal of the halo resulted in an area of empty image space and concomitant changes to the spatial arrangement analysis of Hg-sensitive bacteria about cellulose fibres (Fig. 3f). The distances r at which g(r) was below the reference line increased from 0 to <50 to 0 to 150 μm when the halo was removed, due to the change in the proportions of dipoles of these lengths that would have been recorded as hits. All those dipoles that had previously had one end touching the cellulose fibre and the other touching the halo (hits) would have had one end touching the cellulose fibre and the other touching empty space (misses) following removal of the halo. It should be noted that values of g(r) were below the reference line from 0 to <50 µm even before removal of the halo. This was due to the size of the individual areas occupied by Hg-resistant and Hg-sensitive bacteria and the cellulose fibre. A high proportion of dipoles of lengths less than 50 µm would have had both ends within either the cellulose fibre or a same-coloured region of the field of view (misses). Following removal of the halo, there was a slight increase in values of g(r) at all distances greater than 150 μm. This was attributed to the reduction in Hg-sensitive population density. As the covariance of two populations, P(r), is normalised with the densities of the two populations in the image to give the pair cross-correlation, g(r), a reduction in the density of one population will therefore inevitably result in uniformly higher values of g(r), as seen in Fig. 3f.

The spatial arrangement of Hg-resistant and Hg-sensitive bacteria about cellulose fibres was assessed for all ten images from the no-Hg treatment membrane and ten images from the heterogeneous-Hg treatment membrane following removal of the halo from each image. For bacteria on the no-Hg membrane, the spatial arrangements of Hg-resistant and Hg-sensitive bacteria about cellulose fibres were relatively similar (Fig. 4a). Values of g(r) were below the reference line at distances r < 100 μm for both Hg-resistant and Hg-sensitive bacteria for the reasons outlined above for Fig. 3f. At distances r of 100–550 μm, values of g(r) approached those of the reference line, indicating random arrangements of both Hg-resistant and Hg-sensitive bacteria about cellulose fibres. For bacteria on the heterogeneous-Hg treatment membrane, there were differences in the spatial arrangements of Hg-resistant and Hg-sensitive bacteria about cellulose fibres relative to the no-Hg treatment membrane (Fig. 4b), due to Hg selection effects. Values of g(r) for the spatial arrangements of both Hg-resistant and Hg-sensitive about Hg cellulose fibres were close to zero from 0 to 200 μm, approximately 100 μm more than for the no-Hg treatment. This was indicative of larger areas with very little bacterial cover immediately proximal to Hg cellulose fibres. From 275 to 350 μm, values of g(r) for Hg-resistant bacteria were slightly above the reference line and those for Hg-sensitive bacteria were slightly below the reference line. This is indicative of selection for Hg-resistant bacteria in this region, such that they were better able to compete for resources and occupy a greater proportional area than in other regions. This resulted in a concomitant decrease in the area occupied by Hg-sensitive bacteria.
https://static-content.springer.com/image/art%3A10.1007%2Fs00248-010-9687-5/MediaObjects/248_2010_9687_Fig4_HTML.gif
Figure 4

daime analysis of Hg-resistant and Hg-sensitive bacteria about cellulose fibres on no-Hg and heterogeneous-Hg membranes. The spatial arrangements were similar for Hg-resistant (irregular grey lines) and Hg-sensitive (irregular black lines) bacteria on the no-Hg treatment membrane (a). There was increased clustering of Hg-resistant relative to Hg-sensitive bacteria at distances of approximately 275–350 µm on the heterogeneous-Hg treatment membrane (b). Data are mean (irregular thick lines) ± SE (irregular thin lines) g(r) values from ten epifluorescence microscopy images, with a fitted reference line given by g(r) = 0.70 + (2.65 × 10−3)r (straight line)

Discussion

The methods employed in the current study enabled a quantitative and objective assessment of the changes in bacterial biofilm population structure about Hg cellulose fibres at sub-millimetre scales. Two systems of analysis were used (manually defined contour region analysis and daime analysis) to interrogate ten fields of view each from a membrane sprayed with Hg cellulose fibres (heterogeneous-Hg treatment) and a control membrane sprayed with just cellulose fibres (no-Hg treatment).

Analysis of manually defined contour regions of the 20 images indicated that the area where Hg-resistant bacteria were at a relative fitness advantage was in a narrow region approximately 178–353 µm from the Hg focus. However, when the ten heterogeneous-Hg and ten no-Hg fields of view from these two membranes were considered in their entirety (i.e. without splitting each into six contour regions), there was no significant evidence for Hg-resistant bacteria being at an increased relative fitness advantage around Hg cellulose. Therefore, just as millimetre-scale analysis revealed evidence of selection for Hg-resistant bacteria that was not detectable at the centimetre scale of the whole membrane in a previous study [13], sub-millimetre-scale analysis revealed evidence of selection for Hg-resistant bacteria that was not detectable at the millimetre-scale in the current study.

When whole fields of view from further replicate membranes were included in the analysis (a total of 65 fields of view from five heterogeneous-Hg membranes and 52 fields of view from five no-Hg membranes), there was significant evidence for Hg-resistant bacteria being at an increased relative fitness advantage around Hg cellulose. This indicates that there was variation in the strength of Hg selection from one heterogeneous-Hg membrane to another and that there was particularly weak Hg selection on the heterogeneous-Hg membrane selected (at random) for manually defined contour region analysis and daime analysis in the current study.

The scale and extent of the toxic effects of Hg was demonstrated by the areas in which there was very little bacterial growth (of even Hg-resistant bacteria) around the Hg foci. These areas were largest at close proximity to Hg foci and decreased to the equivalent of background levels over distances of approximately 250 µm. There was therefore a partial overlap of areas where there was an increase in the relative fitness of Hg-resistant bacteria and areas where there were toxic effects on all bacteria. Taken together, this suggests that the fitness advantages for Hg-resistant bacteria are highly localised when Hg is spatially heterogeneous.

Even in regions about the Hg cellulose fibre where there was evidence for selection for Hg-resistant bacteria, the Hg-sensitive cells were at an overall fitness advantage as they always occupied a greater proportional area of the membrane. This is indicative of a high burden of plasmid carriage and a relatively small benefit of Hg-resistance. This may be a common scenario for any analogous assessment of bacterial fitness in terrestrial environments. Firstly, bacterial plasmids from terrestrial environments often carry mercury resistance genes [22]. Secondly, there is a high proportion of bacterial plasmids in terrestrial environments that are large, with the percentage of plasmids isolated from these environments over 100 kb being higher, at 46.8%, than the equivalent values for plasmids isolated from animal hosts (27.6%) and aquatic environments (19.2%) (values for all sequenced plasmids in the Plasmid Genome Database: http://www.genomics.ceh.ac.uk/plasmiddb/, [23]). This is relevant here as there is some evidence that large Hg-resistant plasmids confer greater fitness burdens on their hosts than small Hg-resistant plasmids [24].

The changes in effects of Hg on bacterial populations with increasing distance from the Hg foci may be explained in this system by diffusion-limited kinetics of Hg through the agar. Agar has previously been shown to limit the toxicity of Hg as a higher concentration is required in solidified R2A to achieve equivalent toxicity as seen in otherwise identical liquid media [12]. This also has parallels to the situation in terrestrial environments; similarly to agar, the bioavailability of Hg2+ in soils and vegetation is reduced as a result of adsorption to high-affinity sites, with the rate and degree to which this occurs varying according to the diffusion properties and chemical composition of the particular substrate [25].

Analysis using daime permitted a more fine-scale quantification than was possible using manually defined contour region analysis. The daime analysis indicated that the region where Hg-resistant bacteria were at a relative fitness advantage was 275–350 μm from Hg foci. That was approximately 100 µm narrower than the region derived from manually defined contour region analysis, discussed above. Therefore, daime may allow an accurate quantitative assessment of the arrangements of bacterial populations around a single, central feature. daime analysis also had the advantage of being rapid and straightforward to perform (following the initial phase of redefining the reference line) whereas analysis of contour regions required more time-consuming manual pre-processing of each image. The development of algorithms for specific spatial arrangements, such as objects clustering around a point or preferentially dispersing along a gradient, in daime or similar image analysis programmes would be a great benefit.

The ability to obtain a quantitative assessment of the spatial arrangement of populations of bacteria around a feature has applications for microbial ecology. For example, bacteria have been found to cluster around point-source releases of nutrients from lysed or excreting protozoa [26] or oxygen and nutrients from phytoplankton [26, 27] in seawater. There is considerable interest in investigating the extent to which these associations are species specific and the implications of these associations for bacterial and phytoplankton community dynamics [28]. In terrestrial environments, rhizosphere bacteria adhere differentially to plant roots due to the variety of attachment mechanisms, bacterial surface polysaccharides and plant factors involved, as well as the influence of extracellular environmental conditions [29]. For example, two mucoid derivatives of the biocontrol strain P. fluorescens CHA0 display greater aggregation and attachment to plant roots and fungal structures than the wild-type strain [30]. There is particular interest in identifying plant-growth-promoting strains that effectively colonise and form stable relationships with plant root surfaces as this process is known to be important for the efficacy of biocontrol [31]. In many different natural environments, changes in bacterial gene expression may be mediated by quorum-sensing signals, such as acyl homoserine lactones (AHLs), from other bacteria. The spatial scale at which AHL signals may be effective has been quantified in P. putida using in situ image analysis [32]. This approach demonstrated that effective signalling could take place between individual bacteria and that the distance at which signalling was effective could be up to 78 µm.

To conclude, the method described here allowed the spatial scale over which selection operated in this system to be elucidated. As such, this study demonstrates the impact of model systems in this context. There is no doubt that the application of the techniques described in the current study to more complex, field-relevant situations such as the rhizosphere presents numerous technical challenges. Moreover, further studies will be needed to resolve wider issues surrounding the impact of environmental heterogeneity on both diversity and fitness.

Acknowledgements

F.R.S. was funded by a Natural Environment Research Council Algorithm studentship. We are grateful to Holger Daims for useful advice in the early stages of the study, Stewart Houten for useful discussions in the later stages of the study and six anonymous reviewers for constructive comments on the manuscript.

Supplementary material

248_2010_9687_MOESM1_ESM.doc (7 mb)
ESM Figures S1(DOC 7140 kb)

Copyright information

© Springer Science+Business Media, LLC 2010