Image Analysis Software Based on Color Segmentation for Characterization of Viability and Physiological Activity of Biofilms

Abstract

The novel image analysis software package bioImage_L was tested to calculate biofilm structural parameters in oral biofilms stained with dual-channel fluorescent markers. By identifying color tonalities in situ, the software independently processed the color subpopulations and characterized the viability and metabolic activity of biofilms.

Automated and semiautomated digital image processing methods extract quantitative data about the structure and topographical distribution of a biofilm in two and three dimensions (1, 3, 4, 8, 12-14). To date, only a few software packages have emphasized analyzing color image data to determine the distribution and structure of microbial subpopulations (3, 8). Typically, most color image analysis methods use the original confocal image data file to perform a monochromatic segmentation of the image.

This report's main purpose is to present bioImage_L, which enables in situ color segmentation without prior transformation of micrographs into monochrome channels. The applicability of the software was tested to determine the baseline physiology of dental plaque grown in a mini-flow cell system and the changes to the physiological parameters when dental plaque was subjected to different stress conditions.

Biofilm preparation and image acquisition.

Dental plaque samples were taken from the buccal and lingual surfaces of lower first and second molars with a dental probe. Samples were taken from the gingival margin and supragingival surfaces and suspended in vials containing 0.2 ml peptone-yeast extract-glucose (5) supplemented with a 10 mM potassium phosphate buffer. Biofilms were created in triplicate in the mini flow-chamber system μ-Slide VI for live cell analysis (Integrated BioDiagnostics, Munich, Germany) as in a previous study (2). Briefly, 100 μl of the plaque samples was inoculated into the flow chamber system and incubated in an atmosphere of 5% CO₂ in air at 37°C for 24 h. The baseline physiology of the biofilms was determined by staining them with four fluorescent stains: the BacLight Live/Dead stain (Molecular Probes, Eugene, OR) to measure cell integrity, carboxy-SNARF-1 (Molecular Probes) to measure the intracellular pH, 5-cyano-2,3-ditolyl-tetrazolium chloride (CTC) to measure the dehydrogenase activity, and fluorescein diacetate (FDA) to measure the esterase activity.

The biofilms were examined with an Eclipse TE2000 inverted confocal scanning laser microscope (CLSM) (Nikon Corporation, Tokyo, Japan). The images were automatically acquired with the MultiPoint series macro as a supplement to Nikon's CLSM interface software EZ-C1 version 3.40, build 691 (Nikon Corporation, Tokyo, Japan). CLSM images were acquired with a 60× oil immersion objective with a numerical aperture of 1.4, and the confocal pinhole was set to a diameter of 30 μm. Images were acquired with a zoom factor of 1.0, a pixel resolution of 0.42 μm/pixel, and a field resolution of 512 by 512 pixels. Each stack had a substratum coverage field area of 215 μm by 215 μm. In all cases, the z step for images in a stack was 2 μm, and 10 stacks, composed of 10 two-dimensional (2-D) images, were acquired from each biofilm chamber. The image stacks were serially transformed from the CLSM format Image Display Subsystem to the tiff format using a macro in the EZ-C1 software. The acquired images were processed through the general user interface of bioImage_L.

GUI.

The general user interface (GUI) of bioImage_L was created with the Matlab guide tool (The MathWorks Inc., MA). The GUI's main purpose is to allow easy interaction with the implemented image analysis tools, which primarily support input file preparation and output file displays, as well as fast data preprocessing and processing, structural calculations of biofilm populations, and graphical displays of individual color-based subpopulations with graphic outputs of the results (see http://www.bioimageL.com/get_bioimage_L). The program was created with Matlab 7.4 (R2007a) using the Windows XP SP2 operating system on a computer with a 1.24-GHz central processing unit and 1 GB of random access memory.

2-D cell counting and in situ color segmentation.

The 2-D cell counting routine simultaneously segments red and green classes in CLSM color microphotographs and is the basic unit of analysis for other functions in bioimage_L. To test this function, an image of a biofilm section stained with CTC (red) and Syto24 (green) was processed (Fig. 1). After inputting the image scale (0.42 μm/pixel), the user was asked to select the noise-reducing factor (NRF). The NRF is a standard deviation sigma included in a scalar averaging filter as a square matrix (default: 3 by 3 pixels), which complements a Gaussian low-pass filter. After the NRF is selected, the threshold is determined by an automatic method proposed by Otsu (9), in which an intermediate point in the pixel intensity histogram of the image that corresponds to the threshold value is identified. From the threshold segmentation, the percentage of the area covered by cells can be calculated (Fig. 1a).

FIG. 1.

Color segmentation method implemented in bioImage_L. (a) 2-D section of a 24-h biofilm stained with the CTC metabolic marker (red fluorescence) and the Syto24 counterstain (green fluorescence). This method segments the original color image into four green pseudochannels (b to e) and four red pseudochannels (g to j) (see Table 1 for further specification of the pseudochannel codes and proportional ratios). The total green mask that results from the merging of the four green pseudochannels is shown in panel f. This represents 74% of the total population, with no detectable metabolism by CTC. The red mask is shown in panel k. This mask represents CTC-active cells, accounting for 24% of the total population. Bar, 50 μm.

Subsequently, bioImage_L applies an in situ color segmentation routine that automatically segments the color image into individual pseudochannels, and the areas and percentages of each identified color subpopulation are calculated and presented. The principle of the color segmentation routine (available for download at http://www.mathworks.com/matlabcentral/fileexchange) relies on the color addition theory (6) and classifies each pixel of the image into a predefined color class, resulting in the generation of pseudochannels. The color to be segmented is identified by red, green, and blue codes (Rs, Gs, and Bs), which are subtracted from each pixel in the image. The absolute differences (R1diff, G1diff, and B1diff) between all three components in each pixel are represented as follows: R1diff = R1 − Rs for the red component, G1diff = G1 − Gs for the green component, and B1diff = B1 − Bs for the blue component. Subsequently, the three obtained differences are checked to determine whether they are within the standard tolerance value (i.e., less than the tolerance default value of 0.001).

If all three differences are within this tolerance value, it is determined that the selected color is present in that pixel. The Matlab code returns a pseudochannel of the image with the pixels that have passed the color segmentation test. The color segmentation routine in bioImage_L has been implemented with a simultaneous segmentation of red and green classes, since these colors are used by most common fluorescent labels (segmentation of blue color is also implemented; see http://www.bioimageL.com/get_bioimage_L). As shown in Fig. 1b to e, the image is segmented into four green pseudochannels and then merged as the total green subpopulation area in Fig. 1f. In this example, the inactive (green) subpopulation area represents 76% of the original population. Similarly, four different tonalities of red are segmented in Fig. 1g to j, resulting in a total CTC-active (red) area (Fig. 1k) that represents 24% of the total population in the sample biofilm. The corresponding pseudochannel codes and ratios are presented in Table 1.

TABLE 1.

Red, green, and blue color coding and corresponding percentages of identified pseudochannels in green and red segments of Fig. 1a

Segment	Fig. 1 panel showing pseudochannel	Color coding (red, green, blue)	% in segment
Green	b	118, 238, 0	13
	c	50, 205, 50	39
	d	0, 205, 0	34
	e	0, 139, 69	87
Red	g	238, 44, 44	47
	h	255, 0, 0	38
	i	192, 0, 0	59
	j	139, 0, 0	67

Surface and volume distribution.

One of the main advantages of imaging biofilms with a confocal microscope is that a series of z axis scans is produced, which enables the reconstruction of 3-D profiles. The surface and volume distribution function in bioImage_L was tested to compare the biofilm structure and distribution of independent subpopulations of cells grown on smooth uncoated polystyrene surfaces with those of cell subpopulations grown on surfaces coated with salivary mucins (gelMUC5) (11). Biofilms were created on surfaces preconditioned with gelMUC5, incubated in an atmosphere of 5% CO₂ in air at 37°C for 24 h, and stained with CTC. Biofilms were counterstained with Syto24, which stained all cells fluorescent green.

The first step of the routine uses a parsing algorithm that ranks images by the stack name and identifies the last two numbers in each image name, which correspond to its z position. After this parsing, the user is asked to input the scale (μm/pixel), NRF, and distance between the z layers. Then the user is asked to select one stack, and the program runs the 2-D cell counting routine for each image in the stack, while a subpanel shows the analyzed images. After completion of 2-D cell counting, a subpanel shows the results for the total population and the subpopulations. The parameters in these results are biovolume, mean height, and substratum coverage. In addition, green and red segments are reconstructed in three dimensions (Fig. 2). This 3-D reconstruction is achieved with a modified version of the Matlab code, vol3d (available for download at http://www.mathworks.com/matlabcentral/fileexchange/). As shown in Fig. 2a to c, on the uncoated surface, dental plaque bacteria covered 88% of the substratum surface. However, 91% of the population showed no metabolic activity (green biovolume) (Fig. 2a). In addition, it appeared that the few metabolically active cells (red subpopulation) were allocated in the upper layers of the biofilm (mean height of 17.3 μm), probably where nutrients were more easily accessible (Fig. 2b). In contrast, the salivary mucin-coated surface showed an uneven distribution of cells on the substratum, with substratum coverage of 52% (Fig. 2d to f). However, the presence of mucins on the surface apparently activated the cells' metabolism, with 42% of the population stained fluorescent red by CTC (Fig. 2e). Similar vertical distributions between the red and the green metabolically inactive subpopulations were also seen, with mean heights of 11.6 μm and 11.2 μm, respectively.

FIG. 2.

3-D reconstructions of dental plaque biofilms growing in mini-flow cell systems on an uncoated smooth polystyrene surface (a to c) and a saliva mucin-coated surface (d to f). The fluorescent stain used is CTC, which indicates metabolically active cells (red cells) and metabolically inactive cells (green). In panels a and d, the metabolically inactive green subpopulations are shown, while panels b and e show the active red subpopulations. (c and f) 3-D reconstructions of the entire biofilm population. Axis units are μm.

Analysis of biofilm populations.

The function “viability and metabolic activity of biofilms” in bioImage_L is designed to analyze multiple stacks representing one biofilm in a single run. In Fig. 3, after the analysis is completed, the resulting values of biovolume, mean height, and substratum coverage are presented (Fig. 3e), in addition to a graph showing biomass values corresponding to different z levels and plots of the total population and the green and red subpopulations (Fig. 3f).

FIG. 3.

GUI of the “viability and metabolic activity of biofilms” function in bioImage_L. (a) Command button to open a file path where folders corresponding to biofilm(s) are allocated; (b) setting subpanel; (c) subpanel with the list of biofilms found and command buttons to select either one biofilm or all for analysis; (d) image display subpanel; (e and f) result subpanel (e) with a compiling graph on biomass and z level (f).

To determine experimental reproducibility, results obtained from different biofilms were statistically compared with the function “viability and metabolic activity of biofilms (batch),” which analyzes different biofilms and presents the overall results. A two-way analysis of variance (ANOVA) is automatically performed to give the significant differences in the variabilities of the green and red subpopulations.

For three 24-h biofilm populations, the results on the viability of dental plaque bacteria indicated that the biovolume of the subpopulation of microbes with undamaged cell membranes accounted for 96% ± 2% of the total biofilm biovolume (Fig. 4a). No significant variation was detected by two-way ANOVA (P < 0.0001). When three of these populations were exposed to 5% chlorhexidine gluconate for 30 min, the biovolume of the green population was reduced to 77% ± 1%. The cells that were in the upper levels, closer to the surface, were more affected by the chlorhexidine exposure, although the proportion of viable cells in the deeper biofilm layers was still high (Fig. 4b).

FIG. 4.

Baseline characteristics of dental plaque grown in vitro for 24 h in terms of (a) viability as measured with the BacLight Live/Dead stain, (c) intracellular pH as measured with carboxy-SNARF-1, (e) dehydrogenase activity as measured with CTC, and (g) esterase activity as measured with FDA. The effect of a 30-min exposure to 5% chlorhexidine gluconate on viability is shown in panel b, the effect of exposure to pH 3 for 30 min is seen in panel d, and the effects of nutrient deprivation on dehydrogenase and esterase activities are seen in panels f and h, respectively. Error bars denote standard errors of triplicate experiments. PBS, phosphate-buffered saline; EB, ethidium bromide.

In this report, carboxy-SNARF-1, a cell-permeable fluorescent red dye that emits light in the presence of free ions released due to extreme intracellular pH changes (7, 10), was used in dental plaque biofilms. A working solution of carboxy-SNARF-1 was prepared by mixing 1 μl of 25 mM carboxy-SNARF-1 (Molecular Probes, Eugene, OR) with 999 μl of phosphate-buffered saline. This mixture (40 μl) was added to each tested biofilm chamber and counterstained with 1 μl of 1 mM Syto24 (green fluorescence). The data obtained from three different biofilms showed that the biovolume of the subpopulation with acidic intracellular pH (fluorescent red) represented 7% ± 3% of the total population (Fig. 4c). However, after exposure to extreme acid stress (pH 3) for 30 min, the proportion of the total biovolume with low intracellular pH increased only to 35% ± 4% (Fig. 4d).

In this study, the effect of 16 h of nutrient deprivation on dental plaque bacteria was also studied. The effect of nutrient deprivation was measured by determining the levels of the dehydrogenase activity with CTC (Fig. 4e and f) and the esterase activity with FDA (Fig. 4g and h). CTC was inoculated at a concentration of 5 mM, and biofilms were counterstained with green Syto24 as described above. The FDA-ethidium bromide mixture was prepared and inoculated into the biofilm chambers as previously described (2). The biovolume of the dehydrogenase-active subpopulation (red) of 24-h dental plaque bacteria was 52% ± 4% of the total biovolume (Fig. 4e); however, this value was reduced to 17% ± 4% (ANOVA variability not significant, P < 0.001) when the biofilms were deprived of nutrients. Less dramatic was the reduction of the esterase activity, which showed a basal value of 87% ± 6% (Fig. 4g) and was reduced to 71% ± 5% of the total biofilm biovolume (Fig. 4h).

Conclusions.

By means of the novel in situ color segmentation approach included in bioImage_L, the baseline physiology of 24-h dental plaque was effectively monitored, as well as the physiological changes occurring when dental plaque was subjected to different stress conditions. Application of this software as an alternative to monochrome image analysis processing could be of benefit in biofilm research since further implementation of the software includes simultaneous segmentation of multiple color classes, e.g., when using multiple fluorescence in situ hybridization probes within a biofilm population.