High-Throughput and Quantitative Procedure for Determining Sources of Escherichia coli in Waterways by Using Host-Specific DNA Marker Genes

E. coli reference strains and molecular markers.Reference strains used in this study were previously isolated from the feces of geese (Go66, Go90, Go126, Go172, and Go206) or humans (Hu51, Hu130, Hu132, Hu188, and Hu252) using selective and differential growth media (6, 10, 17). Genomic DNAs from these strains were previously used in suppression subtractive hybridizations to identify goose/duck-specific molecular marker probes. When combined, these probes identified 76 and 73% of the goose and duck isolates tested, respectively (10). Two of the marker gene probes, GB2 and GE11, each hybridized with about 50% of the 135 goose isolates tested and on average cross-hybridized with less than 4% of E. coli from humans and most other animal hosts.

Sample collection and concentration.Offshore lake water samples were collected using standard procedures as described previously (3). Water samples used in spiking experiments were collected from Lake Como (St. Paul, MN) in August 2005. Geese were not present at this lake at the time of sampling. The water samples used for field evaluation were collected in September 2005 from two urban lakes suspected of being contaminated by goose feces, St. Louis Bay in Lake Superior (Duluth, MN) and Lake Calhoun in Minneapolis, MN. In addition, Lake Hartwell (Clemson, SC), a relatively pristine lake reportedly free of fecal contamination from geese, was sampled in September 2006. Bacteria in water samples were concentrated by membrane filtration at 25°C using 0.45-μm (47-mm-diameter) Nuclepore polycarbonate membranes (Whatman, Florham Park, NJ); multiple membranes were successively used to facilitate filtration and circumvent membrane clogging. Filters were washed with 10 ml of phosphate-buffered saline (PBS) (20 mM sodium phosphate, 15 mM sodium chloride, pH 7.2), and bacterial cells were released by gentle agitation for 10 min using a sterile magnetic stir bar. The solution and membranes were transferred to sterile plastic culture tubes and vortexed for 10 min to further remove bacterial cells from membranes.

Enumeration of E. coli.The number of E. coli cells in water samples was determined by using the modified mTEC membrane filtration method (38), except that 5-bromo-4-chloro-3-indolyl-β-d-glucuronic acid (X-Gluc) (500 μg/ml) was used as the chromogenic indicator. Concentrated cell suspensions were diluted in sterile PBS prior to enumeration.

Spiked samples.Bacterial cells in approximately 20 liters of Lake Como water were concentrated by membrane filtration as described above and suspended in 20 ml of PBS. The suspensions contained about 3,000 CFU of E. coli per ml, as determined by using the membrane filtration method and modified mTEC agar medium. E. coli strain Go66 was grown overnight at 37°C in LB medium and centrifuged at 10,000 × g for 10 min at 4°C, and the pellet was washed and resuspended in PBS. Washed Go66 cells were added to the concentrated lake water suspension to obtain 0, 20, 100, 200, 400, or 800 spiked cells per ml. Triplicate samples were prepared for each treatment.

Field studies.The relative contributions of geese to fecal contamination in Lake Superior (St. Louis Bay, Duluth, MN) and Lake Calhoun (Minneapolis, MN) were determined by picking and arraying 1,000 E. coli isolates from each water sample onto membranes as described above. Ducks and geese are frequent inhabitants of both lakes. Samples (4 liters) of near-shore water were concentrated to 5 ml as described above. Colony hybridizations (31) were done using the ³²P-labeled combined GB2 and GE11 probes and analyzed as described below. Fecal E. coli counts in water samples were determined, prior to concentration, by using the modified mTEC membrane filtration method as described above.

Automated isolation and picking of E. coli colonies.Approximately 2,000 E. coli cells from each environmental or spiked water sample were spread plated onto the surface of modified mTEC agar medium containing X-Gluc (500 μg/ml) in 22-by-22-cm Qtray plates (Genetix, Boston, MA). Plates were incubated at 35°C for 2 h, followed by incubation at 44.5°C for 22 h or until colonies reached 1 to 2 mm in diameter. Plates were stored at 4°C overnight to allow further development of blue pigment in colonies. This allowed differentiation of E. coli from other coliform and gram-negative bacteria.

The automated picking of blue E. coli colonies was done using a Genetix QBot robot (Boston, MA), running Qsoft XP version 6000.11 software. Colony-bearing agar plates were illuminated from below using incandescent lighting, and grayscale digital pictures were taken from the top. Translucent blue paper, which separates the hybridization membranes (Genetix, Boston, MA), was placed under Qtray plates to enhance contrast and facilitate picking of blue-colored E. coli colonies. The color of the blue contrast paper was found to be 107:151:167 (red:green:blue), as determined by using the RGB mode of Photoshop CS version 8.0 software (Adobe Systems, San Jose, CA). The translucent paper had 22.5% transmittance at 600 nm, as determined using a Beckman DU-70 spectrophotometer (Fullerton, CA). The parameters used by the QBot for colony identification included colony diameter, roundness, axis ratio, proximity, and overlaps, which were determined on a plate-by-plate basis. Mean pixel intensities of colonies were calculated using grayscale analysis; the upper-threshold value was set to 255, and the lower-threshold values ranged from 100 to 140. Selected blue colonies were picked and placed, in random order, in 384-well microplates containing mTEC medium supplemented with 4.4% glycerol but without X-Gluc. Microplates were incubated at 44.5°C for 12 h and stored at −70°C before use. The taxonomic identities of 384 randomly selected E. coli isolates were verified by growth and reaction on selective and differential laboratory media and by biochemical tests, as described previously (6).

Robotic arraying of isolates.Aliquots (0.5 μl) of E. coli isolates in each microplate well were spot inoculated onto positively charged, 22-by-22-cm Performa nylon membranes (Genetix, Boston, MA) using the QBot robotic platform. The membranes were wetted with sterile water, and excess water was subsequently removed using dry sterile Whatman 3MM filter paper to prevent colonies from running together. Membranes were divided into 6 sections, each section contained 384 subunits, and each subunit consisted of 9 individual spots. Using this gridding format, each membrane was arrayed with 20,736 E. coli colonies, although it is possible to array 36,864 colonies on the membrane using an alternate grid pattern. Replication and standardization was performed by spotting each isolate in three different sections of the membrane and placing one of the positive and negative reference E. coli strains described above in each subunit. Arrayed membranes were placed on the surfaces of mTEC agar plates (without X-Gluc), incubated in an inverted position at 44.5°C for 8 to ∼10 h, and stored at 4°C overnight or until used.

Colony hybridizations.Colony hybridizations were done to determine the reactivities of the E. coli isolates with the mixed goose/duck-specific DNA probes GB2 and GE11 as described previously (10). Extra caution was taken during the cell lysis step to prevent mixing of the lysed colonies. Goose/duck-specific DNA probes were labeled with [³²P]dCTP using the Random Primer DNA labeling system (Invitrogen, Carlsbad, CA) according to the manufacturer's protocol. Membranes were hybridized overnight at 68°C and washed under high stringency at 65°C in 0.1× SSC (1× SSC is 0.15 M NaCl and 0.015 M sodium citrate) containing 0.1% sodium dodecyl sulfate.

Quantitative image analysis.Quantitative image analysis was done as described previously to determine positive and negative signals on colony hybridization membranes (10). Images were captured using a STORM 840 densitometer (Molecular Dynamics, Piscataway, NJ) and analyzed using ScanAlyze version 2.50 software (http://rana.lbl.gov/EisenSoftware.htm ). The normalized intensity of each spot was calculated by subtracting the median background intensity from the mean intensity of each spot, and data were plotted using Sigma Plot version 8.0 software (Systat Software, Point Richmond, CA). A cutoff value was assigned based on normalized mean intensities of negative control spots plus three times the standard deviation.

Equations for estimating relative fecal contributions.Targeted goose- or duck-derived environmental E. coli can have various levels of reaction to host-specific DNA marker probes, and reactions of E. coli from other host animal sources can also vary (10). Therefore, for reliable host source determination and quantification, a variety of probes targeting different host source E. coli strains are required. The relative contributions of fecal E. coli strains from different host sources [C_host(i)] can be estimated based on the following equations. where P_probe(i) is the percentage of environmental E. coli isolates hybridizing with the DNA probes specific for host i; S_(i) is the sensitivity of probes toward E. coli from the target host; and K_{[probe(i),host(i)]} are the rates of cross-hybridization with E. coli strains from other host sources. However, since only one set of goose/duck-specific probes was used in this study, simplifying assumptions were made to treat E. coli strains from other host sources as one entity, and thus, an average cross-hybridization rate was used. Equation 2 was derived and used to estimate the contribution of fecal E. coli from geese/ducks in field studies. $$mathtex$$\[C_{\mathrm{goose/duck}}{=}(P_{\mathrm{probe}}{-}K_{(\mathrm{probe,animals})})/(S_{\mathrm{goose/duck}}{-}K\ _{(\mathrm{probe,animals})})\]$$mathtex$$(2) For the GB2/GE11 mixed probes, the K_{(probe,animals)} and S_goose/duck values were previously determined to be 5.1 and 48.1%, respectively (10).

Statistical analysis.Statistical significance of data was determined by using the analysis of variance (ANOVA) and Tukey's honestly significant difference (HSD) routines, at an α value of 0.01, of R program version 2.0.1 (http://www.r-project.org/ ). Standard statistical analysis parameters were obtained by using Excel 2002 (Microsoft, Redmond, WA). Method linearities for different sample sizes were determined by plotting the number of probe-positive isolates against the number of E. coli cells spiked into water samples, and r values from linear regression were determined.

Robot-assisted isolation of environmental E. coli.The automated picking of colonies by using the QBot system drastically decreased the labor intensiveness associated with obtaining environmental E. coli isolates, relative to manual picking methods. Using this system, 20,736 colonies can be scanned, picked from E. coli-bearing agar plates, and inoculated into master microtiter plates in approximately 6.5 h. Only limited human intervention is required for this procedure, mainly for adjusting the robot scanning parameters (colony diameter, roundness, axis ratio, and proximity) to select target E. coli colonies. This, however, needs to be done only once for each initial batch of Q-tray plates in each series. While some of the instruments involved in this procedure (i.e., the colony-picking robot) are not present in most individual laboratories, many researchers have access to these resources in shared genomic facilities located across the United States.

The efficacy of E. coli isolation and picking was determined by the specificity of modified mTEC agar in differentiating E. coli colonies from other waterborne microbes, by uniformity in spread plating, and by cell growth characteristics. Modified mTEC agar medium has previously been shown to be effective in the isolation of E. coli from water, soil, and fecal samples (14, 17). Preferential E. coli colonies used in this study were 1 to 2 mm in diameter and dark blue in color. Smaller colonies were found to increase colony picking error due to a limited target size, as were larger colonies that had white edges, most likely due to depletion of X-Gluc in the area surrounding colonies. The parameters of roundness, axis ratio, and proximity we specified reduced the risk of picking non-E. coli colonies; typical values used in this study were 0.9, 0.8, and 0.3 mm, respectively. In addition, the blue-colored background sheet used in our studies greatly facilitated robotic picking of colonies. Without this blue background, there was insufficient contrast to allow accurate colony identification.

The accuracy of colony scanning and picking was tested by examining the taxonomic identities of 384 randomly selected isolates. Microbiological and biochemical assays indicated that 99.7, 100, 98.7, and 100% of the isolates were correctly identified as E. coli by their reactions on ChromaAgar, MacConkey agar, and the methyl red and indole tests, respectively. By combining all of these tests, only a limited number of false-positive or -negative isolates would be chosen using the robot picking system. The overall performance of the automated picking system was comparable to that of nonautomated procedures, of which the false-positive rates for environmental water samples were reported as ≤1% (38).

Automated colony arraying.The application of robotic automation also allowed high-density colony arraying, owing to size uniformity of the gridding pins. This allowed for highly accurate spot positioning and subsequent uniform colony growth, due to small deviations in inocula applied to membranes. This also benefited subsequent hybridization and data analyses. The contact area of the individual arraying pin was 0.24 mm², and the area of each spot was 2.2 mm², allowing for the spotting of 20,736 colonies on a single 22-by-22-cm membrane. This large number of colonies allowed for proper replication of unknown isolates and inclusion of replicated positive and negative controls, even when a larger number of isolates were being processed. An example of typical hybridization reactions is shown in Fig. 1. This is similar to macroarray analysis used to identify genomic clones containing specific DNA sequences (27). The inclusion of multiple positive and negative control organisms and replications of unknowns on the membrane increase the statistical power of the results (22) and reduce the number of false-positive and -negative results that are common to many MST methods (29).

• Open in new tab
• Download powerpoint

FIG. 1.

Colony hybridization of ³²P-labeled GB2/GE11 goose/duck-specific marker gene probes to E. coli isolates obtained from geese. E. coli strains Go66 and Hu132 isolated from geese and humans, respectively, served as positive and negative controls, respectively.

In the procedure developed here, both negative and positive marker strains were placed in each section of the membrane, and at least one marker (negative or positive) was placed in each subunit of a section. This allowed establishment of a local binary classification scheme (positively or negatively hybridizing colonies) on a section-by-section basis. Quantitative differences in pixel intensities between positively and negatively hybridizing colonies could be easily distinguished by image analysis (Fig. 2). The effectiveness of this classification scheme using controls placed in each subunit of the array gave average rates of correct classification of 90 to 97% and of 83 to 92% for the six negative and positive control strains, respectively. The relatively high average-rate-of-correct-classification values reported here suggest that this method will be effective in correctly classifying environmental isolates.

• Open in new tab
• Download powerpoint

FIG. 2.

Pixel intensities from a macroarray hybridization membrane containing five E. coli strains isolated from geese (•) and five strains from humans (○). The E. coli strains from geese and humans were replicated 192 times. The membrane was probed with ³²P-labeled GB2/GE11 goose/duck-specific marker gene probes. The solid line indicates a cutoff value for determining positive and negative signals.

Method detection limit.Spiked-sample experiments were done to determine the method detection limit. The E. coli strain Go66 was added to a concentrated environmental water sample to prepare six treatments of spiked samples, which were subsequently analyzed using the developed procedure. A total of 1,000 E. coli colonies were isolated from each sample in each treatment. The number of E. coli isolates that hybridized to the combined probes in each treatment is presented in Table 1. One-way analysis of variance indicated that the number of E. coli isolates reacting to the probes in the spiked samples was significantly different (P = 3.95 × 10⁻⁶) than that for the control. Tukey's HSD analysis was performed to compare the spiked samples with the nonspiked control samples, and significant differences (α ≤ 0.01) were observed only for spiked samples receiving 400 or 800 Go66 E. coli cells. This analysis indicated that the minimum detection limit for the automated screening procedure was 108 spiked E. coli cells for each 1,000 isolates examined (Table 1). Using our method, however, the method detection limit is directly related to the existing background level of indigenous E. coli from geese/ducks in water samples, which was 21 CFU per 100 ml, and the background total number of E. coli cells in the water sample. Thus, our detection limit will most likely improve when using sample water containing less E. coli from these animals. For example, a smaller number of colonies would need to be screened in the more pristine Lake Hartwell water, since when sampled, it had a background of only 72 E. coli CFU/100 ml, and only 0.28% of these reacted with our probes.

Influence of sample size on method linearity.To determine the minimum number of colonies that needed to be examined in order to obtain statistically significant detection results, 1,152 E. coli isolates were randomly picked from spiked samples and hybridized with the goose/duck-specific probes using the developed screening procedure. Since E. coli isolates were picked in a random order for each spiked sample, the first 100, 300, 500, 700, and 1,000 isolates could be used individually to determine the effects of sample size on method linearity. For this analysis, r values were determined by linear regression of hybridizing isolates versus the number of spiked Go66 E. coli cells (Table 2). Results of these analyses (Table 2) showed that method linearity increased with larger sample sizes. When 700 isolates were analyzed, satisfactory method linearity was achieved (r = 0.95), whereas the larger sample size (1,000 isolates) did not further improve method linearity. Moreover, ANOVA analyses, followed by significance testing using Tukey's HSD test, confirmed that at least 700 environmental isolates should be screened in order to obtain statistically (P < 0.01) meaningful results. It should be noted, however, that while this number may appear to be large, it represents only about 3% of the hybridization capacity of the membrane, and the robot system used can pick 700 isolates in about 13 min. Thus, the developed procedure allows the analysis of approximately 30 water samples (sites) per membrane analyzed.

Field applications.To evaluate the utility of the developed protocol under more realistic conditions, we performed host source-specific macroarray hybridizations on E. coli collected from water samples obtained from two urban Minnesota lakes where geese and ducks were suspected to contribute to fecal pollution. Of the 1,000 E. coli isolates obtained from each lake sample, 206 from Lake Superior and 276 from Lake Calhoun reacted positively with the goose/duck-specific probes (Table 3). Based on the use of equation 2, the relative contributions of fecal E. coli from geese/ducks were estimated to be 34% and 51% in Lake Superior (St. Louis Bay) and Lake Calhoun, respectively. These high numbers are not unexpected, since Canada geese and ducks frequent these lakes many times during the year, and control measures are currently in place to control their numbers in Lake Calhoun, an urban lake 4.5 miles from the city of Minneapolis. This is in agreement with results from previous studies showing that migratory waterfowl substantially increase fecal counts in freshwater and saltwater systems (12, 13) and that beaches and water in Lake Superior are frequently contaminated with fecal bacteria originating from geese (Satoshi Ishii, personal communication). In contrast, only 1 of 360 E. coli isolates (0.28%) obtained from Lake Hartwell in Clemson, SC, reacted with the probes. This is consistent with reports that this lake is relatively pristine and is not frequented by migratory goose populations.

Summary and conclusions.Our studies are the first to use a robotics-based approach to quantitatively determine sources and sinks of fecal E. coli in the environment. If implemented on a large scale, this method should allow for the rapid determination of the contribution of geese/ducks to fecal loading of beaches and waterways. As currently configured, the system allows for the high-throughput quantitative analysis of approximately 20,700 E. coli isolates in a relatively short time, with minimal human intervention. However, it is possible to array up to 36,864 colonies on a single growth/hybridization membrane using an alternate grid pattern, thus allowing analysis of even more water samples at once. The high-throughput analysis realized by this method is different from that obtained using host-specific PCR methods (4, 21, 32, 34) in that a large number of individual isolates can be analyzed for each sample. In contrast, the latter method analyzes each water sample using a single PCR, usually in a nonquantitative manner. Additionally, since we analyzed for E. coli, our results can be directly correlated with fecal indicator bacterial counts commonly used by state and federal agencies to assess contamination in impacted lakes, rivers, and beaches. This is not possible using PCR-based methods directed against bacteria that are not currently used as fecal indicators (4, 5). Although the availability of colony-picking robots has steadily increased in recent years as genomic research expands, access may still be limited for some researchers. The probe method described here can still be used in the absence of a colony-picking robot, by hand-picking colonies and using a multiprong replicator to transfer cells to membranes. However, the use of this manual method will result in reduced throughput and accuracy relative to those of the robotic system.

While in our studies we used hybridization probes that were specific for geese/ducks in the central Midwest (10), the method can be easily adapted for use with any marker gene system developed to examine bacterial populations impacting waterways and beaches. Development of marker gene probes specific for E. coli from fecal host sources other than geese/ducks is still needed. With the availability of an array of probes specific for E. coli originating from other major fecal host sources impacting watersheds, the number of simplifying assumptions made in this study can be considerably reduced, and the overall reliability and quantitative nature of the method can be improved in the future.