Fountain Flow cytometry, a new technique for the rapid detection and enumeration of microorganisms in aqueous samples

Authors

  • Paul E. Johnson,

    Corresponding author
    1. Department of Physics and Astronomy, University of Wyoming, Laramie, Wyoming
    2. SoftRay Incorporated, Laramie, Wyoming
    3. Observatoire Océanologique, Laboratoire ARAGO, Université Pierre et Marie Curie-Paris6; Centre National de la Recherche Scientifique, CNRS, INSU, UMR7621; Banyuls-sur-Mer, France
    • Department of Physics and Astronomy, University of Wyoming, Laramie, WY 82070, USA
    Search for more papers by this author
  • Anthony J. Deromedi,

    1. Department of Physics and Astronomy, University of Wyoming, Laramie, Wyoming
    2. SoftRay Incorporated, Laramie, Wyoming
    Search for more papers by this author
  • Philippe Lebaron,

    1. Observatoire Océanologique, Laboratoire ARAGO, Université Pierre et Marie Curie-Paris6; Centre National de la Recherche Scientifique, CNRS, INSU, UMR7621; Banyuls-sur-Mer, France
    Search for more papers by this author
  • Philippe Catala,

    1. Observatoire Océanologique, Laboratoire ARAGO, Université Pierre et Marie Curie-Paris6; Centre National de la Recherche Scientifique, CNRS, INSU, UMR7621; Banyuls-sur-Mer, France
    Search for more papers by this author
  • Jennifer Cash

    1. Department of Physics, South Carolina State University, Orangeburg, South Carolina
    Search for more papers by this author

Abstract

Background:

Pathogenic microorganisms are known to cause widespread waterborne disease worldwide. There is an urgent need to develop a technique for the real-time detection of pathogens in environmental samples at low concentrations, <10 microorganisms/ml, in large sample volumes, ≥100 ml.

Methods:

A novel method, Fountain Flow™ cytometry, for the rapid and sensitive detection of individual microorganisms in aqueous samples is presented. Each sample is first incubated with a fluorescent label and then passed as a stream in front of a laser, which excites the label. The fluorescence is detected with a CCD imager as the sample flows toward the imager along its optical axis. The feasibility of Fountain Flow cytometry (FFC) is demonstrated by the detection of Escherichia coli labeled with ChemChrome CV6 and SYBR Gold in buffer and natural river water.

Results:

Detections of labeled E. coli were made in aqueous suspensions with an efficiency of 96% ± 14% down to a concentration ∼200 bacteria/ml.

Conclusions:

The feasibility of FFC is demonstrated by the detection of E. coli in buffer and natural river water. FFC should apply to the detection of a wide range of pathogenic microorganisms including amoebae. © 2006 International Society for Analytical Cytology

Pathogenic microorganisms are known to cause widespread waterborne disease worldwide, including developed countries. Waterborne diseases are thought to account for nearly 6,000 deaths per day, mostly in children (1). Although progress has been made to identify pathogens in our environment and to understand their association with diseases, much progress is yet to be made for their reliable and facile detection. Microbial pathogens represent a wide diversity of organisms ranging from viruses to protozoa. Microbial pathogens can be detected and enumerated by different methods, including destructive methods (mostly based on DNA analysis) and non-destructive methods (culture and different cytometry techniques combined with specific probes). In most applications, there is a need not only to detect and to enumerate the pathogen but also to determine if the detected cells are living or dead. Furthermore, the delay between the sample collection and the availability of the results should be as short as possible to reduce health risks. The need for rapid methods that provide timely results has been widely recognized among responsible health departments. Detecting pathogens, especially at low concentrations, is one of the most important challenges in modern microbiology.

Typical microbial detection methods to evaluate drinking water sources or recreational waters begin with a filtration or concentration step aimed at isolating small concentrations of pathogens, i.e., 1–10 infectious units, from large volumes of water (hundreds of milliliters to thousands of liters). The number of pathogens to be concentrated generally depends on the instrument used for the quantification. When detection is based on methods that enumerate individual cells, the most commonly used techniques are epifluoresence microscopy and flow cytometry. However, these techniques are not appropriate for the enumeration of rare events, <10 microorganisms/ml, as clearly shown by Lemarchand et al. in a comparison study (2). Microscopy is not suitable for surveying a concentrate for a low number of microorganisms since background interference would make this a tedious approach to pathogen monitoring. Similarly, flow cytometry is not appropriate to analyze large volumes and to detect 1–10 microorganisms/ml.

The ChemScan™ system (Chemunex, Paris, France) is a recently developed solid-phase cytometer that uses direct fluorescent labeling of microorganisms in combination with an automated detection and counting system. This system can be applied to the detection of specific microorganisms when combined with the use of taxonomic probes, such as fluorescent antibodies or fluorescent oligonucleotides. Furthermore, fluorescent antibodies can be used in combination with fluorescent viability dyes to detect and enumerate viable and specific cells without any fixation procedure. The main interest of this technology is that it allows detection of rare events against a background of untargeted microorganisms (2). It has been successfully applied to the detection of bacteria (3, 4, 5, 6), protozoa (7, 8) and, more recently to the detection of toxic algae (9). This technology is probably the most powerful technique for the detection of rare events at the cellular level. The method still requires the concentration of targeted microorganisms on a filtration membrane before detection and enumeration. The ChemScan technology remains expensive, and low-cost instruments that can be used for the detection of rare events should be welcomed.

Although filtration has little consequence for most prokaryotes, it represents a limitation when applied to the detection of protozoa since it requires prefixation of the sample to preserve the membranes of eukaryotic organisms. Therefore, it does not apply to the detection of living organisms and this is an important limitation for viability assessment (7).

This study presents a novel scheme for detection of pathogenic microorganisms in aqueous samples, economically and in real time, first introduced by Johnson et al. (10). This technology is based on laser-induced fluorescence of labeled cells and requires no filtration step. Similar to conventional flow cytometry, the suspended microorganisms are stained to facilitate detection. The resulting aqueous sample is passed as a stream in front of a laser or LED, which excites the fluorescent labels. Unlike conventional flow cytometry, the resulting fluorescence is measured with a CCD (in this study) or CMOS imager, as the measured sample flows toward the imager along the optical axis. Samples can be illuminated along the optical axis in an epifluoresence mode or at an angle to the optical axis. We call both schemes Fountain Flow™, or FF (11), and refer to both Fountain Flow™ cytometry and the Fountain Flow™ cytometer as FFC. The sample flows through a circular orifice, large enough to inhibit clogging yet small enough to enable imaging all of the target particles flowing through it. The large diameter of the flow (2 mm in this study) removes the need for a sheath flow. Imaging the flow of microorganisms traveling toward the CCD imager yields images of in-focus particles in the focal plane as well as out-of-focus particles. Particles are normally not spatially resolved (in this study the dimension of a pixel at the focal plane is 3.4 × 7.9 μm), but the images are used to make photometric measurements of target particles throughout the focal plane. The focal plane is chosen to minimize the spot size of microorganisms giving a dramatically higher signal-to-noise ratio than conventional techniques. Flow cytometry with flow along the optical axis was first presented in the literature by Dittrich and Goehde (12), who used a photomultiplier in a conventional flow cytometer. Although imaging flow cytometry with flow along the optical axis has previously been described in the literature (13), it does not incorporate epifluoresence illumination and detection, and is therefore of limited use with a translucent medium. FFC allows for detection of microorganisms in translucent media, such as turbid surface waters, blood, or beverages. In addition, FFC is amenable to detection of microorganisms in the presence of high photometric background, including unbound fluorescent dye.

In this article, we illustrate the use of this new technology that can be applied to a large variety of microorganisms from bacteria to protozoa. Although in this study, FFC is demonstrated for the first time on bacteria at concentrations ≥200 bacteria/ml, bacteria are detected and enumerated individually. The current system configuration has a flow velocity limited to about 2 ml/h, making detection of 1–10 bacteria/ml too time-consuming to be practical.

MATERIALS AND METHODS

Fountain Flow Cytometry

Figure 1 shows a prototype FFC implemented with a solid state laser. Illumination with an LED is also possible. Figures 2 and 3 give an overview of the basic principles of FF. The sample suspended in an aqueous solution flows up a tube toward imaging optics. The sample particles are then imaged by a CCD or CMOS camera and finally counted or measured photometrically. The imaging optics would include, with a single color instrument, a filter isolating the wavelengths of fluorescent emission. Normally the sample flows upward, vertically.

Figure 1.

A Fountain Flow detection system with a 475 nm solid-state laser illuminator.

Figure 2.

A schematic drawing of the aluminum flow block used with the device in Figure 1. Upper Panel: The sample enters the flow block through a Tygon tube connected to a stainless steel tube and exits through a stainless steel tube. Two 2-mm holes have been drilled into the aluminum flow block: an entrance hole and an exit hole. As the sample flows up the internal entrance hole, it passes through the focal plane of the CCD/CMOS camera. This hole is generally painted black to reduce scattered light. A Teflon tape gasket is sandwiched between the aluminum flow block and a circular BK7 window, and tightly held with a screw-on brass cap. The gasket is cut to form a channel through which the fluid is diverted into the exit hole. Lower Panel: A photograph of a working flow block with attached tubing. The block is mounted onto a black-anodized plate.

Figure 3.

Schematic diagram of the epifluorescent Fountain Flow cytometer used in this study. A sample of fluorescently tagged cells flows through the flow cell toward the CCD camera and fore-optics. The cells are illuminated in the focal plane by a laser. When the cell(s) pass through the CCD camera focal plane they are imaged by the CCD camera and lens assembly through the transparent flow cell window, using a filter that isolates the wavelength of fluorescence emission. The fluid in which the cells are suspended then passes by the window and out the flow cell drain tube (in the path indicated by black arrows).

In FF, the column of flowing sample is both viewed and illuminated end-on. In this study, a dichroic mirror is used to allow both the viewing path and illumination path to coincide with the flow axis, but in the opposite direction, similar to the optics in an epifluoresence microscope. The volume imaged is controlled by the depth of field of the objective. This is normally kept small to prevent the negative effects of foreshortening on the resulting digital images. The ideal situation occurs when a particle flowing up the flow tube is imaged onto the same pixel(s), while it is in focus. This minimizes streaking and the number of pixels over which the bacterial cell detection is made. In turn, instrument noise and the background contribution from unbound dye are reduced.

The cross-section of the flow tube that can be monitored by the imager is not dependent on the opacity of the sample fluid and can be very large. This permits extremely high flow rates and sample sizes (∼100 ml/min). This article describes implementation of FFC with a moderate power laser (a 28-mW argon ion gas laser) and a 2-mm diameter circular orifice. The laser beam is expanded with a double convex lens so that the beam over-fills the orifice. Particles near the walls of the tube flow more slowly and may appear in multiple images; this effect must be taken into account when calibrating the device.

Computer Automated Detection and Enumeration

Automated microorganism detection and enumeration can be summarized by the following two steps (14):

  • 1To perform the detection of microorganisms from FF digital imagery in real time our Biocount™ computer program is used. The flow stream is continuously monitored by a digital camera under computer control, typically at rates of 2–5 frames/s. The digital images of the flow stream are examined by a computer for candidate bacteria pixels based on their signal strength above the noise.
  • 2Next, the counting results are analyzed by the computer to remove repetitive counts of the same bacterium. Software compares the coordinates of bacteria found in successive images and determines which bright spots or streaks seen in successive images are probably the same bacterium (because their coordinates are within a predetermined distance).

Biocount enumerates fluorescently labeled microorganisms in sequences of CCD/CMOS images. The program is written in IDL (RSI, Boulder, CO). Biocount queries the user for three parameters (Table 1): (1) the threshold intensity for a single pixel to be considered as a candidate for a positive detection, (2) the minimum number of contiguous pixels in which a positive detection is made in order for a pixel group to be considered as a candidate for a microorganism detection, and (3) the overall photometric intensity of a candidate microorganism before it is counted as a confirmed detection. Note that there are two intensity thresholds above: items 1 (a single pixel threshold) and 3 (the threshold of detection for a group of pixels). Biocount then determines the standard deviation for the intensity of pixels in an image and identifies candidate bacteria pixels based on their signal strength above background noise. It then counts these candidates based on the number of detected pixels in a single contiguous group and the total integrated photometric intensity within a specified rectangle centered on the group. Biocount has been tested on a number of CCD images and it has been found that the number of bacteria identified by eye is equal to the number of bacteria counted by Biocount. Finally, Biocount removes multiple counts of the same bacterium as explained above.

Table 1. Criteria for the Detection of a Single Microbial Cell
 CriterionMinimum value of criterion for microorganism detection
  1. Each must be satisfied in order for a group of bright pixels to be classified as a single cell by Biocount.

1Number of standard deviations above the mean signal intensity for a pixel to be bright enough for group membership≥3
2Number of continuous candidate pixels to be considered a group≥3
3Photometric intensity of a group in order to be considered a microorganism≥10.0 ADUs

The photometric background of the CCD images is non-uniform, owing to the non-uniform illumination of the FF orifice. Biocount eliminates the background by differencing two sequential images, and setting negative pixels in the difference to zero (Fig. 4). When the previous image is subtracted from the current image, target particles detected in the previous image are seen as groups of negative pixels. Setting negative pixels to zero means that only target particles (bacteria) from the current image and noise remain in the difference image. Since target particles move laterally from one image to the next, pixels comprising a particle in one frame are not normally subtracted from pixels comprising the same particle in a successive frame, except at high particle concentrations.

Figure 4.

Panels A and B show two successive images of amoebae (bright spots), Naegleria lovaniensis, and background from a Fountain Flow cytometer. The bright background in panels A and B is from unattached dye in the FFC and corresponds to the 2-mm FFC orifice. The amoebae were labeled with CV6 and imaged with a 100 ms exposure time. Rings are out of focus N. lovaniensis. Panel C is the difference between the two images (C = B − A), which shows the bright point sources while virtually eliminating background.

Illumination Beam Profile

Figure 5 shows digital images taken through the FFC, comparing the laser beam profile with the orifice diameter. The laser beam was expanded with a double convex lens in order to overfill the 2-mm orifice. The Gaussian beam profile is nearly 60% of peak intensity at the edge of the orifice.

Figure 5.

Images of the illuminated 2-mm flow cell aperture made with the fountain flow camera. Panel A: Aperture and flow channel illuminated in white light. Panel B: Laser beam illuminating a white card placed onto the aperture. Panel C: 20% contours of the illuminating laser (from image in the upper right) placed over the image of the flow aperture and channel (upper left).

Signal-to-Noise Ratio

The signal-to-noise ratio of detection is dependent on the sources present, the fluorescent intensity of the microorganism being detected, and the number of pixels over which the signal is spread. Noise sources include: photon counting statistics from signal and background in each pixel, variations in background owing to unbound dye, dark current noise in the CCD, and the noise associated with reading out each pixel (readout noise). While photon counting statistics in measuring the signal from a source are not dependent on the number of pixels over which the source signal is spread, the noise contribution from background, dark current, and readout noise do depend on the number of pixels over which the source is detected. In particular, if the source contributes intensity to n adjacent pixels, the intensity can be measured by summing these n pixels. The dark current noise, readout noise, and noise contribution from background counting statistics all increase with the square root of n. It is therefore highly desirable, and a major feature of the FFC, that microorganisms are imaged onto a small number of pixels as they pass through the focal plane. In addition, to reduce the size of stored images, images were binned, or summed, so that each 2 × 2 group of four pixels in the original image was stored as a single pixel representing the sum of the original pixels. This occurred after data acquisition but before any data analysis. In this article all measurements and dimensions refer to the pixel format in the original image format (prebinned).

Figure 6 shows the result of photometric measurements made on four sets of images of Naegleria lovaniensis labeled with CV6. These brightly fluorescing amoebae are used to illustrate the photometry of microorganisms as they move through the FFC focal plane. Photometry was performed by software on image differences, by drawing an imaginary 40 × 20 pixel box centered on each amoeba and summing the digital counts (in analog to digital converter units, or ADUs) within each box. The 40 × 20 box is roughly square, as each pixel has a 1:2 aspect ratio in its linear dimension. The peak signal was between 200 and 3700 ADUs. The method of Fowler et al. (15) was used to calibrate the gain of the CCD camera, which gave a conversion factor of 95 electrons (detected photons) per ADU. Because detected photons are converted into electrons in the detector, it is common for photons detected in an electronic imager to be measured in “electrons,” and the associated noise in units of “electrons rms” (where rms is root mean square). The manufacturer's specification for readout noise, found at their website (http://electrim.com/EDC1000DataSheet.html), is 35 electrons rms/pixel or ∼0.3 ADUs.

Figure 6.

Photometric intensity within a 40 × 20 pixel box (background subtracted) vs. frame number for four randomly selected N. lovaniensis labeled with CV6 as shown in Figure 4. The frame numbers have been adjusted so that the photometric peak for each amoeba occurs in frame number 4. This figure shows the progression of photometric measurements as amoebae pass through the focal plane. The photometric intensity for each microorganism varies according to the intensity of illumination, the amount of label attached, and the size of each amoeba. At the CCD gain setting used in this study, 1.0 ADU corresponds to 95 detected photons.

Threshold for Detection

For this study, the detection criteria were based on the signal (from the difference image) of a group of three or more contiguous (2 × 2 binned) pixels, each with signal >3 standard deviations (determined for a single CCD column) above the noise. The digital counts (in ADUs) from these three or more pixels were summed, and must exceed the detection criterion of 10.0 ADUs, equivalent to 950 detected photons. Counting statistics of 950 electrons is equal to the square root of 950 electrons rms, or 31 electrons rms. In addition, the detection threshold of three standard deviations for an individual (2 × 2 binned) pixel is about 3.2 ADUs (owing to system noise). Background noise is proportional to the square root of the number of pixels that contribute to the detection (at least three pixels as given by Criterion 2 in Table 1). If three pixels in a single contiguous pixel group have a signal-to-noise of 3.0 or greater, then the signal from each pixel (binned) is >9.6 ADUs. This means that criteria 2 and 3 in Table 1 are somewhat redundant: three pixels will almost always achieve a signal of >10.0 ADUs (equivalent to 950 detected photons). A better signal-to-noise ratio can be achieved by using a CCD or CMOS imager with a smaller system noise. (In more recent work we use a CMOS camera with a system noise ∼60 times smaller at a 200 ms exposure time.)

System Configuration

Measurements made for this study were taken from an FFC with the specifications given in Table 2. The data acquisition system consisted of a 400 mHz PC (Hewlett Packard 6640C). Data acquisition software was written in IDL. The Electrim 1000L CCD (Electrim Corporation, Princeton, NJ) camera (with custom optics) used in this study has an ISA card interface to the PC bus. Owing to the relatively slow clock speed of the data acquisition computer, images were not obtained continuously; there was a 320 ms dead time between each 200 ms exposure.

Table 2. Detection Specifications for Fountain Flow Cytometer Using Electrim 1000L CCD Camera (with a Texas Instruments TC 245 CCD), Braintree BSP-99M Syringe Pump, and a Spectra-Physics 163-A1202 Argon Laser
ParameterValue
CCD sensitive area6.24 mm × 4.78 mm
CCD format753 × 242
CCD pixel size8.5 × 19.75 μm
CCD camera system noise150 electrons rms
CCD camera gain95 electrons/ADU
Diameter of circular orifice2.0 mm
N.A. of collection optics0.26
Image magnification2.5×
Field of view of CCD2.5 mm × 1.9 mm
CCD camera depth of field in the flow tube200 μm
FWHM of argon ion laser beam at flow tube1.4 mm
Argon ion laser power at flow cell19.5 mW
Fraction of time (duty cycle) that CCD is imaging the flow tube using a 400 mHz PC38%

A Braintree Labs (Braintree, MA), BSP-99M syringe pump was used to flow samples through the system at a constant pump rate of 2.1 ml/h from a 3-ml syringe. The pump speed was calibrated both by weighing the output of the pump over a fixed time interval and by visually measuring the rate of motion of the syringe plunger. Approximately 20–50 ml of distilled water was used to cleanse the system between samples and was injected rapidly by hand.

The lenses selected for the objective and camera are stock 30-mm diameter coated achromats (Edmund Industrial Optics, Barrington, NJ). The objective has a focal length of 50 mm, while the camera lens has a focal length of 125 mm. These lenses provide a satisfactory magnification (2.5×) while allowing 60% of a 30° beam through the optical system. The lenses were selected for minimum spot size using ray tracing software. The approximate spot diameter is 25 μm, filling three CCD pixels. This is optimum for unbinned images, as it spreads the signal from a spatially unresolved microorganism over three or more pixels. This offers a three-fold redundancy that allows one to differentiate spurious signal from a single pixel and signal from a spatially unresolved microorganism spread over at least three contiguous pixels. Practically, we find that blurring owing to motion of microorganisms through the focal plane increases the number of pixels over which a microorganism is detected, allowing us to require a minimum of three contiguous pixels even in 2 × 2 binned images. The image sensor in the Electrim Camera body is an uncooled Texas Instruments TC 245 CCD with a peak quantum efficiency of ∼70% at 750 nm. Illumination of the FFC was performed with a 28-mW Spectra-Physics (Mountain View, CA) 163-A1202 argon gas laser, emitting at 488 nm. The laser power incident on the orifice window was periodically measured at ∼19.5 mW. The FFC used a 505DRLP dichroic mirror (Omega Optical, Brattleboro, VT) for epiillumination and an OG515 yellow longpass filter (Edmund Industrial Optics, Barrington, NJ) to isolate emission from the fluorescent labels. The inner surface of the mounting tube was covered with black light-absorbing paper (Edmund Industrial Optics, Barrington, NJ) and the inner surface of the orifice was painted with flat black paint to absorb scattered light.

The length of the illuminated flow was 2 cm, which caused cells to be exposed to laser illumination for ∼100 s at the 2.1 ml/h flow rate used here. The path length could easily be reduced to ∼2 mm. In addition, as we discuss below, more recent implementations of the FFC have flow velocities that are two orders of magnitude larger, which reduce the cell exposure time proportionately. At the current flow rate it takes about 1 s to replace the volume that spans the optical field depth. Since the exposure time plus the time interval between exposures is 520 ms, each cell is exposed within the in-focus volume for at least one full camera exposure.

Sample Preparation

Escherichia coli K12 samples (ATCC No: 25404) were purchased from US Biological (Swampscott, MA). The bacteria were maintained on trypitc soy agar and were cultured in tryptic soy broth in a shaking incubator at 37°C for 16–24 h before use. Extra stocks were kept at −80°C in 25% glycerol.

Label Selection

For this project, we chose to illustrate the efficacy of detecting viable bacteria at relatively low concentrations using the ChemChrome V6 (CV6) dye (Chemunex) and SYBR Gold (Molecular Probes/Invitrogen, Eugene, OR). CV6 is activated by esterase activity, which converts CV6 to the fluorescent fluorescein molecule. Fluorescein is retained by cells with intact membranes (viable cells) and lost by cells with damaged membranes. Fluorescein is pH-sensitive, with its greatest emission and absorption occurring between a pH of 8.0 and 9.0 (16). SYBR Gold is a sensitive DNA/RNA dye commonly used for gel electrophoresis (17). The SYBR Gold stain has absorption peaks at 300 and 490 nm and an emission peak at 537 nm (16).

Measurements of Cells Stained with CV6

We made two sets of comparisons of sample concentrations using CV6, between FFC and Whipple grid on two dates using two different solutes: Chemunex ChemSol (Chemunex) (a pH 7.0 buffer recommended by the manufacturer) and PBS with 0.1% Tween-80 (pH 7.4). The staining protocol used for CV6 in this study is that described in Parthuisot et al. (18). Cells were prepared by pelleting 1 ml of an overnight culture of E. coli K12 at 1300g for 5 min. The resulting pellet was washed with 1 ml of buffer (ChemSol or PBS), centrifuged a second time, and suspended in 1 ml of buffer. In a separate tube, 100 μl of the washed E. coli were transferred into 900 μl of buffer. Then 10 μl of CV6 was added to this sample. The resulting mixture was incubated at 30°C for 30 min in the dark. This sample was then diluted by 1/10,000 and 5 ml was filtered onto a polycarbonate 0.4 μm filter (GE Osmonics, Minnetonka, MN) for Whipple grid enumeration. Between incubation and measurement, samples were kept in the dark and on ice to maintain the fluorescent intensity of the CV6.

Measurements of Cells Stained with SYBR Gold

We made two sets of comparisons of sample concentrations using SYBR Gold (Invitrogen, Eugene, OR) between FFC and Whipple grid counts on two dates using two different solutes: a sterile saline solution (0.9% NaCl) (pH 6.5) and water from the Tech River (near Perpignan, France) sampled on November, 2004 (pH 8.4, with NaCl added to achieve a 0.9% salinity). Tech River water was passed through a 20-micron filter to eliminate detritus larger than bacteria.

SYBR Gold was used to stain E. coli K12 in a tris acetate-EDTA (Sigma, St. Louis, MO) buffer. This buffer was 40 mM tris acetate, pH ∼ 8.3, with 1 mM EDTA. Tris-acetate-EDTA (Tris) was used as it does not cause dye aggregates when combined with the SYBR Gold. Cells to be stained with SYBR Gold were prepared by pelleting 1 ml of an overnight culture of E. coli K12 at 1300g for 5 min. The resulting pellet was washed with 1 ml of Tris, centrifuged a second time, and suspended in 1 ml of Tris. The resulting sample was diluted 1/100 in Tris (giving a final sample volume of 1 ml). Subsequently, 40 μl of 200× SYBR Gold was added to the sample, giving a final concentration of 8×. The sample was then incubated for at least 1 h at room temperature. Then the sample was washed with Tris and resuspended in the same. The resulting sample was diluted 1/10 in 0.9% saline solution or river water and counted both with Whipple grid and FFC measurements.

Whipple Grid Enumeration

For comparison of cell concentration with an independent method, we used Whipple grid counting of samples filtered onto a 0.4 micron polycarbonate filter (GE Osmonics, Minnetonka, MN). The Whipple grid was calibrated with a micrometer-driven translation stage. The original overnight cultures (∼2–3 × 108 bacteria/ml) of the E. coli K12 were diluted. Then sub-samples (at ∼2–3 × 104 bacteria/ml) of the resulting solution were counted with a Whipple grid reticle mounted on the eyepiece (20×, for a total magnification of 200×) of an Olympus BH-2 epifluoresence microscope to obtain the total number of cells in the sample. Additional serial dilutions of the sample (1/10–1/100) were then enumerated with the FFC. Whipple grid counts were made on 20–90 random fields of a single sample. The number of fields was chosen to yield at least 300 counts to produce a high degree of accuracy in the measured concentration of bacteria. The fields were divided into three groups, and three sets of means and standard errors were produced for each sample.

We expect insignificant growth in the samples between FFC and Whipple grid measurements because samples contained no nutrients, and the samples were kept refrigerated at 3°C or on ice (in the case of CV6 samples to be measured by FFC) after incubation with stain and prior to measurement. The time difference between FFC measurements and Whipple grid measurements of samples was less than 2 h.

FFC Enumeration

Samples were vortexed for 30 s immediately before FFC detection. No sample measurements were made more than 20 min after vortexing to minimize the effects of sedimentation of bacteria within the syringe. No measurements of CV6-labelled cells were made more than 30 min after the end of the CV6 incubation. For each sample enumerated with the FFC, the sample was counted in multiples of 100 frames (each frame consisting of a 200 ms exposure) with each set of 100 frames representing a sub-sample. A mean and standard deviation was produced from the ensemble of the 100-frame sub-samples representing a single sample.

RESULTS AND DISCUSSION

Measurements of FFC Noise

At short integration times (≤10 ms) with no incident illumination, system noise is 1.6 ADUs, or 150 electrons rms. At integration times up to 800 ms the system noise is the same, indicating that dark current noise and CCD readout noise are insignificant compared with noise in the electronics. In addition, different images used in this study indicate that for both SYBR Gold and CV6, the background noise attributable to unattached dye is insignificant. Therefore, integration-time independent system noise can be used as the sole source of system noise for the camera in this study. Noise in photometric measurements is dominated by system noise up to 250 ADUs. Above this limit counting statistics from the measured microorganism dominates the noise in photometric measurements. In addition, systematic photometric errors are produced by non-uniform illumination and transmission effects within the instrument.

Measurements of E. coli Stained with CV6

Figure 7 shows data comparing Whipple grid counts and FFC from samples of E. coli labeled with CV6 and inoculated into buffer. Measurements were done with both ChemSol and PBS buffer. Each set of FFC measurements consists of an integral number (3–10) of 100 camera frames (images), each taken with a 200 ms exposure time and a 320 ms inactive or “dead” time between exposures. With a pump rate of 0.036 ml/min (1.0 ml in 28 min), each 100 frames of data corresponds to 0.031 ml of fluid measured. At this pump speed, each bacterium is measured in multiple frames so that the 320 ms dead time does not result in incomplete counting. However, care must be taken to account for multiple detections of the same bacterium. Biocount accomplishes this by tracking bacteria in consecutive frames. Objects detected in multiple frames that moved less than 20 pixels in both the x- and y-direction between successive frames are counted as a single bacterium. The standard deviation of FFC counts was computed from n sub-samples taken from n × 100 frames, each sub-sample consisting of 100 frames. Counting statistics were computed by taking the square root of the number of mean counts/100 frames. Whipple grid (WG) counts were made from a number of randomly chosen fields of view (frames). A standard deviation for the total number of WG counts is computed using the counts taken from each frame. The WG counts were used to compute the bacterial concentration of the stock solution and the resulting samples diluted from that solution. Figure 8 shows the corresponding histogram of intensity for a single sample comprising the data shown in Figure 7. Ideally, there should be a linear relation, with slope of 1.0, between WG and FFC counts if the FFC counting efficiency is 100%. The relationship gives a counting efficiency of 113% ± 14% with the ChemSol buffer and 99% ± 14% with PBS.

Figure 7.

Comparison of Whipple grid measurements of CV6 labeled E. coli K12 samples enumerated on October 18, 2004 and October 21, 2004. Black diamonds denote data taken on the earlier date, diluted with ChemSol buffer. White triangles represent data from the later date, diluted with PBS buffer. A solid line with a slope of 1.0, intersecting the origin, is drawn through the data for comparison. The best fit line through the data points is given by y = 1.10x − 11.6 with an R2 of 0.955. Error bars are displayed for 1σ, as determined from counting statistics. Each data point represents four to ten individual measurements of subsamples.

Figure 8.

Histogram of intensity (in ADUs) for CV6 labeled E. coli in ChemSol buffer. Intensity measurements are the sum of all of the pixel intensities within a 20 × 10 pixel rectangle in the difference image. Because many microorganisms were detected in multiple images as they traveled through the focal plane, the greatest intensity was used. Eight data points, not shown, have intensities greater than 500 ADUs.

To test the hypothesis that FFC enumeration errors were consistent with Poisson (counting) statistics, a reduced χ2 test was performed on each data point used in Figure 7, representing the number of cells counted in n subsamples. The reduced χ2 is given by χ2 = σ2/m (n −1) where σ is the standard deviation measured from the n subsamples comprising a single sample and m is the mean number of cells counted by FFC in each subsample. The maximum value of reduced χ2 for the seven data points for the CV6 measurements was 0.64. This corresponds to a probability >80% (for each point) that the measured variance can be attributed to counting statistics.

Measurements of E. coli Stained with SYBR Gold

Figure 9 shows data comparing Whipple grid counts from samples of E. coli labeled with SYBR Gold and inoculated into a 0.9% saline buffer as well as water from the Tech River filtered through a 20-micron paper filter (Whatman, Maidstone, UK) to eliminate detritus. Each of the FFC measurements consists of an integral number of 100 camera frames (images), each taken with a 200 ms exposure time and a 320 ms dead time between images. With a pump rate of 0.036 ml/min (1.0 ml in 28 min), each 100 frames of data corresponds to 0.031 ml of fluid measured. As above, each bacterium is measured in multiple frames, so that the 320 ms dead time does not result in incomplete counting. Figure 10 shows the intensity histogram for a single sample comprising the data shown in Figure 9. The relationship between WG and FFC counts gives a counting efficiency of 79% ± 14% with the saline buffer and 92% ± 14% with Tech River water. A null sample (not inoculated with bacteria) was measured and is included in Figure 9.

Figure 9.

Comparison of Whipple grid measurements of SYBR Gold labeled E. coli K12 samples enumerated on March 26, 2005 and March 29, 2005. Black diamonds denote data taken on the earlier date, diluted with a 0.9% solution of NaCl in distilled water. White triangles represent data from the later date, diluted with water from the Tech River (France) passed through a 20-μm filter to eliminate detritus. A solid line with a slope of 1.0, intersecting the origin, is drawn through the data for comparison. The best fit line through the data points is given by y = 0.90x − 18.6 with an R2 of 0.963. Error bars are displayed for 1σ, as determined from counting statistics. Note that two points lie very close to the origin. Each data point represents three to five individual measurements of subsamples.

Figure 10.

Histogram of intensity (in ADUs) for SYBR Gold labeled E. coli in a 0.9% saline solution with distilled water. Intensity measurements are the sum of all of the pixel intensities within a 20 × 10 pixel rectangle in the difference image. As many microorganisms were detected in multiple images as they travel through the focal plane, the greatest intensity was used. Seven data points, not shown, have intensities greater than 500 ADUs.

The maximum value of reduced χ2 for nine out of the ten data points for the SYBR Gold measurements (Fig. 9) was 1.44. This corresponds to a probability >40% that the measured variance for each individual sample can be attributed to counting statistics. The tenth data point had a reduced χ2 of 2.5, corresponding to a probability of 0.04. The results are consistent with the hypothesis that variance in the enumeration data is attributable to counting statistics.

Sensitivity of Results to Intensity Threshold for Detection

Figures 8 and 10 show histograms of photometric intensity within a 20 × 10 pixel box centered on the detected microorganisms. If a microorganism is seen in more than one difference image, the greatest intensity is used.

A bimodal distribution of intensities is evident in Figures 8 and 10. In addition, detections are made at intensities as low as 16.0 ADUs, as in Figure 8. This spread in intensities in detected bacteria is due to non-uniform illumination of microorganisms, cell clustering, the perspective from which the rod-shaped E. coli are viewed, and the amount of stain retained by each cell. Chains of E. coli are commonly seen in our samples when viewed with the microscope, so multiple peaks are expected.

Since some of the cells enumerated in our study are from the low intensity end of the intensity histogram, particularly in Figure 8, an obvious concern is the number of low intensity cells that are not detected/counted. Nearly all detections based on the intensity of pixels exceeding three standard deviations were at 30.0 ADUs or greater. Therefore, tripling the intensity threshold from 10.0 to 30.0 ADUS would have less than a 5% decrease in the counts of the samples shown in Figures 8 and 10.

FUTURE WORK

Although the data shown here validate the viability of the FFC for counting bacteria within a limited range of matrices, flow rates, and range of fluorescent intensity, much more work is needed to explore the full range of viability of the technique. More recent data taken by us from an FFC using a CMOS imager, an LED illuminator, an 8-mm circular aperture, and flow rates from 8 ml/min (for bacteria) to 100 ml/min (for amoebae) show that this technique is viable for enumerating amoebae and bacteria in large samples of untreated surface water.

Another challenge is to select a fluorescent label that avoids interference from background fluorescence, especially chlorophyll A and B, in natural waters. Using LEDs as an illumination device is a natural extension of the current work, as very high intensity LEDs with a variety of wavelengths are now readily available. Other obvious improvements to the current system include implementation of a faster computer and a CMOS imager, which will virtually eliminate dead time. (CMOS cameras allow the chip readout to occur line by line while the device is exposing.) Furthermore, in a two-color adaptation of FFC, two fluorescent signals could be combined to simultaneously address the specific detection of the targeted pathogen and its viability using a physiological probe.

Acknowledgements

Paul E. Johnson is grateful for a sabbatical fellowship from the Université Pierre et Marie Curie (Paris 6).

Ancillary