Flow Chamber System for the Statistical Evaluation of Bacterial Colonization on Materials

Biofilm formation on materials leads to high costs in industrial processes, as well as in medical applications. This fact has stimulated interest in the development of new materials with improved surfaces to reduce bacterial colonization. Standardized tests relying on statistical evidence are indispensable to evaluate the quality and safety of these new materials. We describe here a flow chamber system for biofilm cultivation under controlled conditions with a total capacity for testing up to 32 samples in parallel. In order to quantify the surface colonization, bacterial cells were DAPI (4`,6-diamidino-2-phenylindole)-stained and examined with epifluorescence microscopy. More than 100 images of each sample were automatically taken and the surface coverage was estimated using the free open source software g’mic, followed by a precise statistical evaluation. Overview images of all gathered pictures were generated to dissect the colonization characteristics of the selected model organism Escherichia coli W3310 on different materials (glass and implant steel). With our approach, differences in bacterial colonization on different materials can be quantified in a statistically validated manner. This reliable test procedure will support the design of improved materials for medical, industrial, and environmental (subaquatic or subaerial) applications.


Introduction
Prevention of microbial contamination and resulting biofilm formation on surfaces is one of the biggest challenges in biomedical and industrial settings. The use of indwelling medical devices such as pacemakers, prosthetic joints, and venous-or urinary catheters poses risks of bacterial infections. In fact, 95% of urinary tract infections are attributed to the usage of urinary catheters, and 87% of bloodstream infections are associated with intravascular devices [1]. Microbial contamination is also problematic in a large number of industrial processes and causes high costs for cleaning and maintenance [2]. For instance, the overall costs associated with hull fouling on a mid-sized vessel were estimated to be $56 M per year [3]. The development of biofilms takes place as a result of a series of events: microbial surface attachment, cell proliferation plus the formation of microcolonies, and the production of extracellular polymers [4,5]. Among these, bacterial adhesion is a key step in biofilm formation and numerous factors like surface charge, roughness, hydrophobicity, and predisposition

Design of the Flow Chamber System
The main constituents of the flow chamber system presented in this study were four cylindrical glass vessels (200 mm length, 50 mm in diameter), each equipped with a stainless steel tray with the dimensions of 153 × 29 mm 2 (total area: 44.4 cm 2 ) to serve as the specimen holder ( Figure 1). The tray has a capacity for eight coupons 37 × 14 × 1 mm 3 in size, but could be loaded with specimens of different dimensions and geometrical shapes with a maximum height of 5 mm. The flow chambers, which could be used individually or in parallel, were connected with silicon tubes to medium supply and waste containers. To assure an equal distribution of fresh medium to all chambers used in parallel, a combination of a membrane pump and magnetic valve system was chosen. The air pressure was generated in the container for medium supply by the membrane pump. At defined times, the valve system opened and allowed fresh medium to flow into the chambers. Frequency and duration of medium injection was selected by the user. At the same time, used medium was drained into the waste containers.
The rapid and homogenous distribution of injected fresh medium was verified with dye experiments (Appendix A, Figure A1). Different combinations of pulse frequency, pulse length, and stirrer rotation were tested. These experiments revealed that a stirrer at the bottom of the chamber was necessary to assure homogenous distribution approximately four min after medium injection. We chose a slow flow rate of 16  Mono-cell-layer adhesion was rated as bacterial colonization and assessed by epifluorescence microscopy, which is more distributed in laboratories than a confocal microscopy. Pretests showed that nutrient reduced mineral salt medium fulfilled the requirement for bacterial colonization and a diluted (1:25) mineral salt medium MM63 [15] was used for the colonization experiments.
The dissolved oxygen content (DO) and pH of the cultivation media were measured throughout the whole experiment. The dissolved oxygen level did not fall below 80% during the whole incubation time, which demonstrated that oxygen supply was sufficient under the applied conditions. The mineral salt medium was buffered and only little changes in the pH value (<0.4) occurred (Appendix B, Figure B1).
The viability of the E. coli cells in the cultivation chamber was checked and has proved to be stable at end of the experiment (2.1-4.1 × 10 3 CFU·mL −1 ). Mono-cell-layer adhesion was rated as bacterial colonization and assessed by epifluorescence microscopy, which is more distributed in laboratories than a confocal microscopy. Pretests showed that nutrient reduced mineral salt medium fulfilled the requirement for bacterial colonization and a diluted (1:25) mineral salt medium MM63 [15] was used for the colonization experiments.
The dissolved oxygen content (DO) and pH of the cultivation media were measured throughout the whole experiment. The dissolved oxygen level did not fall below 80% during the whole incubation time, which demonstrated that oxygen supply was sufficient under the applied conditions. The mineral salt medium was buffered and only little changes in the pH value (<0.4) occurred (Appendix B, Figure B1).
The viability of the E. coli cells in the cultivation chamber was checked and has proved to be stable at end of the experiment (2.1-4.1 × 10 3 CFU·mL −1 ).

Quantification of the Initial Bacterial Adhesion
On the basis of the free open source software g'mic, we developed a semi-automated analysis tool for the quantification of bacterial colonization on material surfaces. The script and a flow chart of the processing steps ( Figure S1) are provided in the Supplementary Materials and thus this analysis tool is available for usage and adaptation according to demand. This semi-automated analysis procedure allows for a fast and accurate analysis of a high amount of coupons.
For an easy and quick assessment, overview images were generated from each coupon sample to elucidate the colonization characteristics of the material under investigation (Figure 2a). automatically calculated by the triangle algorithm [16]. In short, this algorithm draws a line between the highest and the lowest signal of the pixel intensities detected in the histogram. The distance between this line and the histogram is calculated for every point, and the threshold is set at the point where the distance is at its maximum. This algorithm was chosen because the histogram of the pixel intensities showed a unimodal distribution. Segmentation was done for each image separately. Detected objects, which were smaller than six pixels (1.4 μm 2 ) or larger than 400 pixels (94.48 μm 2 ), were considered as artefacts and removed automatically. Afterwards, overview images were generated that included the results of the segmentation and the original pixels for one certain coupon ( Figure 2b). These overview images allowed for a quick evaluation of image quality and segmentation results and provided the possibility to recognize and remove blurred images and false detected objects quick and easily.
The bacterial surface coverage, given in percent, was estimated after thresholding. The area of segmented bacteria (A) and the reference area (Aref) were measured. The surface coverage (A/Aref) on a certain material coupon was calculated for every image and for the total coupon. Results were given in percent (A/Aref * 100) (Figure 2c).

Determination of the Minimum Image Number
To address the question of how many pictures are necessary to reach statistically rigorous results, the cumulative variation coefficient was calculated for each coupon of the experiments. One hundred and five images were necessary to lower the coefficient to 13% or below ( Figure 3); for a variation coefficient below 20% at least 75 images were necessary. At first, segmentation of the cells from the background was done by a threshold which was automatically calculated by the triangle algorithm [16]. In short, this algorithm draws a line between the highest and the lowest signal of the pixel intensities detected in the histogram. The distance between this line and the histogram is calculated for every point, and the threshold is set at the point where the distance is at its maximum. This algorithm was chosen because the histogram of the pixel intensities showed a unimodal distribution. Segmentation was done for each image separately. Detected objects, which were smaller than six pixels (1.4 µm 2 ) or larger than 400 pixels (94.48 µm 2 ), were considered as artefacts and removed automatically. Afterwards, overview images were generated that included the results of the segmentation and the original pixels for one certain coupon (Figure 2b). These overview images allowed for a quick evaluation of image quality and segmentation results and provided the possibility to recognize and remove blurred images and false detected objects quick and easily.
The bacterial surface coverage, given in percent, was estimated after thresholding. The area of segmented bacteria (A) and the reference area (Aref) were measured. The surface coverage (A/Aref) on a certain material coupon was calculated for every image and for the total coupon. Results were given in percent (A/Aref * 100) (Figure 2c).

Determination of the Minimum Image Number
To address the question of how many pictures are necessary to reach statistically rigorous results, the cumulative variation coefficient was calculated for each coupon of the experiments. One hundred and five images were necessary to lower the coefficient to 13% or below ( Figure 3); for a variation coefficient below 20% at least 75 images were necessary.

Determination of the Minimum Image Number
To address the question of how many pictures are necessary to reach statistically rigorous results, the cumulative variation coefficient was calculated for each coupon of the experiments. One hundred and five images were necessary to lower the coefficient to 13% or below ( Figure 3); for a variation coefficient below 20% at least 75 images were necessary.

Assessment of Bacterial Adhesion in Dependence of the Coupon Position in the Flow Chamber System
In order to compare the ability of different materials to attract bacterial cells, it has to be ensured that bacterial colonization is independent of the coupon position in the flow chamber system. Therefore, a high number (105) of images for each coupon were captured to analyze a large area of the coupons. This number of images proved to be just sufficient enough to generate fingerprints of each coupon, with each fingerprint depicting the colonization pattern of one coupon. These fingerprints revealed no dependency of the bacterial colonization either on the position of the material coupon or on the flow direction in the chamber ( Figure 4). There were differences between the coupons, but these differences were randomly distributed (no obvious variation).

Assessment of Bacterial Adhesion in Dependence of the Coupon Position in the Flow Chamber System
In order to compare the ability of different materials to attract bacterial cells, it has to be ensured that bacterial colonization is independent of the coupon position in the flow chamber system. Therefore, a high number (105) of images for each coupon were captured to analyze a large area of the coupons. This number of images proved to be just sufficient enough to generate fingerprints of each coupon, with each fingerprint depicting the colonization pattern of one coupon. These fingerprints revealed no dependency of the bacterial colonization either on the position of the material coupon or on the flow direction in the chamber (Figure 4). There were differences between the coupons, but these differences were randomly distributed (no obvious variation).

Reproducibility of Results
To assess the reproducibility of the results obtained with the flow chamber system, colonization experiments with glass coupons were carried out. Since there was no specific pattern recognized on the coupons, the surface coverage (%) of the whole coupon was calculated for the analysis of differences between glass and implant steel. The results for each flow chamber system experiment (each with four flow chambers, each flow chamber equipped with up to eight coupons) were taken together and checked with the Shapiro-Wilks-, the Kolmogorov-Smirnov-, the Lilliefors-, and the Anderson-Darling test for normal distribution. All data sets were normally distributed (α = 0.05). Afterwards, one way analysis of variance (ANOVA) was performed with these data sets. No significant differences were observed between the four trials with glass coupons, when the significance level was set to 0.1%. When the significance level was set to a lower level, significant differences between the first and the last glass trial were recognized, but nevertheless the differences between the mean values were less than 0.1% (Figure 5a). Furthermore, the confidence intervals of the four trials were overlapping, which underlines that the results were highly reproducible.

Reproducibility of Results
To assess the reproducibility of the results obtained with the flow chamber system, colonization experiments with glass coupons were carried out. Since there was no specific pattern recognized on the coupons, the surface coverage (%) of the whole coupon was calculated for the analysis of differences between glass and implant steel. The results for each flow chamber system experiment (each with four flow chambers, each flow chamber equipped with up to eight coupons) were taken together and checked with the Shapiro-Wilks-, the Kolmogorov-Smirnov-, the Lilliefors-, and the Anderson-Darling test for normal distribution. All data sets were normally distributed (α = 0.05). Afterwards, one way analysis of variance (ANOVA) was performed with these data sets. No significant differences were observed between the four trials with glass coupons, when the significance level was set to 0.1%. When the significance level was set to a lower level, significant differences between the first and the last glass trial were recognized, but nevertheless the differences between the mean values were less than 0.1% (Figure 5a). Furthermore, the confidence intervals of the four trials were overlapping, which underlines that the results were highly reproducible.

Reproducibility of Results
To assess the reproducibility of the results obtained with the flow chamber system, colonization experiments with glass coupons were carried out. Since there was no specific pattern recognized on the coupons, the surface coverage (%) of the whole coupon was calculated for the analysis of differences between glass and implant steel. The results for each flow chamber system experiment (each with four flow chambers, each flow chamber equipped with up to eight coupons) were taken together and checked with the Shapiro-Wilks-, the Kolmogorov-Smirnov-, the Lilliefors-, and the Anderson-Darling test for normal distribution. All data sets were normally distributed (α = 0.05). Afterwards, one way analysis of variance (ANOVA) was performed with these data sets. No significant differences were observed between the four trials with glass coupons, when the significance level was set to 0.1%. When the significance level was set to a lower level, significant differences between the first and the last glass trial were recognized, but nevertheless the differences between the mean values were less than 0.1% (Figure 5a). Furthermore, the confidence intervals of the four trials were overlapping, which underlines that the results were highly reproducible.   Figure 4).
Finally, the data sets for the bacterial surface coverage for glass and implant steel coupons were compared (Figure 5b). The surface coverage was almost doubled on the implant steel in comparison to the glass material. Both data sets were normally distributed (α = 0.05). The t-test revealed significant differences between glass and implant steel coupons (α = 0.001).

Discussion
In this study, we present a flow chamber system designed for the testing of materials regarding their susceptibility to bacterial adherence. This test system combines convenient handling with great flexibility concerning shape and size of the material specimens. Microbial strains, and medium and culture conditions can be selected according to the user's demand.
Two general approaches for bioreactors suitable for biofilm testing are known from literature. In the classical flow chamber systems, or "plug flow reactors", the specimens are exposed to bacterial cultures in a chamber with an inlet and an outlet on opposite sides. Typical examples are drip flow reactors, classical flow chambers, and modified Robbins devices. All these systems have in common that the turbulence is low, depending on the flow velocity. However, conditions inside these chambers are not constant and change progressively throughout the reactor [17,18]. Mixing occurs only by diffusion, which we had considered as the main drawback of these systems. In the second approach, specimens are placed in common bioreactors, ensuring constant conditions in the whole liquid phase by mixing with a stirrer. Typical examples are the CDC (Centers for Disease Control), the annular, and the rotating disc biofilm reactor [17,18]. The use of a stirrer enhances shear forces and turbulences in the liquid phase, which can influence the adhesion behavior of bacteria [19][20][21]. On the other hand, biofilm formation can be reduced under static compared to dynamic conditions [19,21].
We combined the advantages of both systems. The addition of fresh medium was performed regularly to allow limited growth in the flow cells. Medium inlet and outlet are located at opposite sides of the bioreactor. Constant conditions were ensured by mixing with a stirrer centered on the bottom of the flow chamber. Moreover, the specimens were located on a tray above the stirrer to reduce the influence of shear forces. The presence of two outlets offers the opportunity to drain the medium periodically. This is a valuable tool to simulate and analyze subaquatic and also subaerial biofilms which are frequently encountered in nature.
This specific design of the flow chamber system allows (i) homogenous medium supply; (ii) application-related biofilm cultivation; (iii) investigation of specimen with different geometries; and (iv) analysis of a statistically relevant number of samples.
We decided to use a diluted minimal medium and bacteria which were not in the exponential stage of growth for adhesion experiments. In nature, exponential growth is rather an exception and bacteria have to cope with a reduced supply of energy. Under these conditions, the establishment of a biofilm is promoted because it offers a fitness advantage [22,23]. Furthermore, it is reported that biofilm formation increased in diluted growth medium [24,25] and was described as faster when the growth rate of planktonic bacteria was lower [25].
Although biofilm formation is a big challenge in medical and industrial applications, an easy and fast direct quantification method which allows for a reproducible and statistically reliable analysis is still missing [13,26]. Quantification of bacteria directly on the surface reveals more information compared to the indirect methods currently in use. Therefore, we developed a direct method based on a semi-automated analysis for rapid discrimination of materials concerning their susceptibility for bacterial adhesion. The generation and pre-analysis of overview images also facilitate the evaluation of the materials.
A commonly used indirect method is the dislodgement of bacterial cells from the specimen by sonication, followed by plate spreading and colony counting (colony forming units = CFU). The detachment procedure is the crucial step during analysis and the application of different cell detachment procedures hamper the comparability of the results [13,27]. A weak detachment procedure does not ensure a complete detachment of the cells [13], whereas an inappropriate power density of the sonication regime leads to the thinning of the cell membrane, which decreases the number of detected colonies [28]. Furthermore, different sonication times are applied, eventually resulting in different CFU numbers [27,[29][30][31]. Moreover, when the detached bacterial cells appear in aggregates, the detected CFU will be too low [13]. Cell count determination by microscopic analysis directly on the material surface usually results in higher values compared to the results obtained by sonication and determination of CFU [32]. Besides, CFU methods are labor-intensive, time-consuming, and have a high material throughput [18]. Colorimetric and fluorescent methods are faster and more convenient [33], but nevertheless, most colorimetric methods are done in microtiter plates, which are not suitable for testing bacterial colonization on different surfaces with different geometries. With indirect methods only the total number of adhered bacterial cells is detectable and it is not possible to focus on weak points of a certain specimen like threads or grooves on screws or stents, for example, or the interface of two combined materials.
Various methods for the quantification of bacterial colonizations are known from literature and a "gold standard" does not exist, since each method will give an answer to a specific question [14]. The specific application of the material has to be considered when choosing the right cultivation and quantification method [18,21]. Nevertheless, the choice of the biofilm cultivation method (static or flow conditions) seemed to be critical for adhesion behavior, and there is obviously a need for standardized methods and agreed protocols if different material surfaces will be compared. The developed flow chamber combined with the quantification step is ideally suited for this purpose.
Variations were recognized within a certain material specimen, which can be caused by material inhomogeneities or biological fluctuations. However, we showed that with a high number of images per material specimens statistically rigorous results were possible. The variation coefficients fell below 13%, which underlined the importance of a high number of images to gain reliable statements about material differences. We present here a method with which already small changes between materials can be detected, because of its high statistical resolution.
In sum, we developed a flow system which is cost efficient and easy to handle. It can be used for material specimens of various sizes and is not only appropriate for the testing of different materials and coatings, which are usually provided as flat coupons, but also for the testing of specimens such as whole implant devices like screws, fixators, or dental prostheses.

Bacterial Strain and Growth Media
Escherichia coli W3110 (kindly provided by Prof. R. Hengge, Freie Universität Berlin (now Humboldt-Universität zu Berlin), Berlin, Germany) was used for the experiments. A colony was picked from an agar-plate (MOPS, 0.033 M KH 2 PO 4 , 0.25 M Glucose, 20 g·L −1 Agar) and cultivated overnight in the minimal medium MM63 [15]. After reaching the stationary phase, the culture was diluted to OD 540 = 1 with fresh medium and used for inoculation of the flow chamber system.

Test Coupons
Glass material was purchased from Marienfeld, Lauda-Königshofen, Germany and cut into the appropriate size of 37 × 1 × 1 mm 3 by hand. Coupons made of implant steel, also 37 × 13 × 1 mm 3 , were purchased from Königsee Implantate GmbH, Allendorf, Germany. Coupons were washed in detergent solution, disinfected for 20 min in 70% ethanol solution, air dried, and transferred aseptically to the flow chambers.

Operation of the Flow Chamber System
Cylindrical glass vessels (20 cm, 5 cm in diameter, Ochs, Bovenden/Lenglern, Germany) were used as biofilm reactors. The reactors were closed by silicone rubber plugs equipped with one inlet (V4A steel, diameter: 3.5 mm) and two outlets (both glass, diameter: 3 mm) Figure 1. The upper outlet limited the maximum fluid level to approximately 140 mL, the lower one was used for draining the reactor. The components of the flow chamber system (medium supply, bioreactor, and waste) were connected via silicon tubes. Silicone rubber plugs (already equipped with inlet and outlets), the specimen holders, and the tubing plus electronically controlled magnetic valves needed for medium distribution were purchased from Durakult, Berlin, Germany. Distribution of fresh medium to the individual flow chambers was achieved by a combination of air pressure driven medium flow and magnetic valve systems as described in Section 2.1. Continuous mixing of the medium was achieved by a magnetic stirrer (10 × 3 mm 2 ) centered at the bottom of the flow chamber. The system was run at 37 ± 2 • C in an incubator (Innova 44 Lab Shaker Series, New Brunswick Scientific's, Nürtingen, Germany).

Bacterial Attachment Tests
Pretests had shown that the usage of diluted MM63-medium (4% v/v) with a medium exchange rate of 270% per day, and an incubation time of 18 ± 1 h resulted in a mono-cell-layer colonization on glass surfaces, which was required for analysis by epifluorescence microscopy.
The flow system was stopped after 18 ± 1 h of incubation, flow chambers were emptied, and the material coupons were removed, washed with sterile PBS solution (8.0 g NaCl, 0.2 g KCl, 1.44 g Na 2 HPO 4 , 0.24 g KH 2 PO 4 , pH 7.4), fixed with 3.7% formaldehyde (Sigma-Aldrich, Deisenhofen, Germany) in PBS solution for 2 h at 4 • C, and rinsed afterwards with sterile deionized water. Coupons were dried and stored at room temperature.
To test cell viability, colony forming units were determined in the medium phase when the flow system was stopped. Medium solution was diluted with MM63 medium lacking glucose and plated on STD1-agar-plates (Merck, Darmstadt, Germany), each dilution was plated in triplicate. After incubation at 37 • C for 24 h, colony forming units were counted. The cell viability test was performed in quadruplicate.

Image Analysis
The free open source software g'mic version 1.6.8 (http://gmic.eu, accessed on 17 December 2015) was used for digital processing and analysis of the microscopic images. Quantification of the initial bacterial adhesion was done as described in Section 2.2. The script is provided in the Supplementary Materials. Fingerprints of the colonization pattern were compiled with Microsoft Excel (2013) using the conditional formatting tool.

Statistical Evaluation
Results were evaluated with OriginPro 9 (OriginLab, Northampton, MA, USA). Normal distribution was checked with four different tests (Shapiro-Wilk-test, Kolmogorov-Smirnov test, Lilliefors test, and Anderson-Darling test). Multiple comparisons of the results were done with one way analysis of variance (ANOVA) and post hoc analysis for pairwise comparison with Tukey's range test. The significance level was set to 5% for all statistical tests applied (confidence of 95%).
Supplementary Materials: A short instruction to g'mic (BAM.pdf) and the used script (BAM.gmic) are available online at www.mdpi.com/1996-1944/9/9/770/s1. Both can be used with g'mic version equal to or higher than 1.7.2. Additionally, the influence of the flow velocity on the colonization characteristic of Staphylococcus pseudintermedius was investigated. Therefore, S. pseudintermedius was cultivated in the flow chamber system with complex medium for 26 h, at 37 • C with different flow rates. A flow rate of 21 mL·h −1 led to a wave like pattern on the coupon ( Figure A2a). When the flow rate was decreased to 16 mL·h −1 this pattern was not recognized ( Figure A2b).

Appendix B DO and pH Measurement
Dissolved oxygen content (DO) and pH of the cultivation media were measured throughout the whole experiment. Therefore, the bioreactor was equipped with non-invasive sensors (PreSens, Regensburg, Germany) for pH and dissolved oxygen (DO). Sterile single use sensor spots for pH measurement and reusable DO spots were glued inside the glass vessel. DO spots were calibrated prior to each experiment according to the manufacturer's instructions. The dissolved oxygen level did not fall below 80% during the whole incubation time, which demonstrated that oxygen supply was sufficient under the conditions applied. The mineral salt medium was buffered and, therefore, only little changes in the pH value (<0.4) occurred ( Figure B1).

Appendix B DO and pH Measurement
Dissolved oxygen content (DO) and pH of the cultivation media were measured throughout the whole experiment. Therefore, the bioreactor was equipped with non-invasive sensors (PreSens, Regensburg, Germany) for pH and dissolved oxygen (DO). Sterile single use sensor spots for pH measurement and reusable DO spots were glued inside the glass vessel. DO spots were calibrated prior to each experiment according to the manufacturer's instructions. The dissolved oxygen level did not fall below 80% during the whole incubation time, which demonstrated that oxygen supply was sufficient under the conditions applied. The mineral salt medium was buffered and, therefore, only little changes in the pH value (<0.4) occurred ( Figure B1).

Appendix B DO and pH Measurement
Dissolved oxygen content (DO) and pH of the cultivation media were measured throughout the whole experiment. Therefore, the bioreactor was equipped with non-invasive sensors (PreSens, Regensburg, Germany) for pH and dissolved oxygen (DO). Sterile single use sensor spots for pH measurement and reusable DO spots were glued inside the glass vessel. DO spots were calibrated prior to each experiment according to the manufacturer's instructions. The dissolved oxygen level did not fall below 80% during the whole incubation time, which demonstrated that oxygen supply was sufficient under the conditions applied. The mineral salt medium was buffered and, therefore, only little changes in the pH value (<0.4) occurred ( Figure B1).