Observing Growth and Division of Large Numbers of Individual Bacteria by Image Analysis

We describe a method that enabled us to observe large numbersof individual bacterial cells during a long period of cell growthand proliferation. We designed a flow chamber in which the cellsattached to a transparent solid surface. The flow chamber wasmounted on a microscope equipped with a digital camera. Theshear force of the flow removed the daughter cells, making itpossible to monitor the consecutive divisions of a single cell.In this way, kinetic parameters and their distributions, aswell as some physiological characteristics of the bacteria,could be analyzed based on more than 1,000 single-cell observations.The method which we developed enabled us to study the historyeffect on the distribution of the lag times of single cells.

INTRODUCTION

Top

Introduction

The dynamics of bacterial populations is now frequently studiedby observing and describing the behavior of single cells. Ata recent conference on predictive microbiology (Quimper, France,16 to 19 June 2003), several papers and a keynote lecture werededicated to the connection between single-cell observationsand bacterial population dynamics (13). The interest in thisarea is due to the recognition that probability-based methods(such as quantitative microbial risk assessment) have to takeinto account the distributions of kinetic parameters in a cellpopulation and that these distributions cannot be derived fromobservations made at the cell population level.

For a bacterial growth curve describing the variation of a cellpopulation with time, Baranyi and Pin (2) showed the mathematicalequivalence between the transition from the lag phase to theexponential phase (curvature) and the distribution of the firstdivision times of individual cells. However, as Baranyi (1)pointed out, mapping from the distribution to the curvatureis theoretically invertible; in practice, for numerical reasons,it is impossible to identify the distribution of individuallag times from the curves for log cell concentration versustime (called population growth curves below). Hence, this distributionprovides more fundamental information than the population growthcurves.

It is therefore desirable to observe single cells to study thelag time or, more generally, the early stages of growth. McKellarand Lu (10) emphasized that modeling the bacterial lag (whosecharacterization is imperative in food microbiology) is inherentlydifficult; new measurements of the distribution of the physiologicalstate of single cells in the population is necessary to improvepredictive models. So far, however, workers have described veryfew reliable methods for observing the division times of sufficientlylarge numbers of individual cells so that statistically robustdistributions can be identified from the observations.

Analysis of bacterial growth at the single-cell level can betraced back to the study of Kelly and Rahn (6), who suggesteda method for detecting the growth rate of individual cells byusing a piece of agar placed under a microscope. These authorsobserved a slight decrease in the individual division timeswith the number of generations. Powell (11) suggested a flowchamber approach in which a cellophane membrane separates thecells from a flowing medium. This method enabled him to changethe environment and to monitor the change in growth.

Newer techniques for single-cell studies include turbidimetry(9), in which a serially diluted culture is inoculated intoa multiwell plate with approximately one cell per well and theturbidity is measured with an automated plate reader. The detectionlevel is ca. 10⁶ to 10⁷ cells/well (depending on cell size,etc.), which means that the original bacterium has undergonesome 20 divisions. Another common technique is flow cytometry(12), in which a sample from a culture is measured to determineDNA content and size. This technique, which can be considereda technique which provides a snapshot of a culture, allows alarge number of cells to be analyzed, but a single cell is notobserved for a longer period. To solve the latter problem, Wakamotoet al. (14) used lithographic techniques to create micrometer-sizewells and optical tweezers to transfer a single bacterium fromone well to another. In this way, it was possible to observeconsecutive divisions of the same cell, although a very limitednumber of cells could be analyzed in each experiment.

Flow cells are widely used in biofilm and microcolony studies(4), but they are used less frequently in single-cell studies(3, 8, 11). The reason for this is that the flow is insufficientto remove the daughter cells, so the surface rapidly becomescrowded and either it is impossible to measure the individualgrowth rate or the growth rate is significantly influenced bythe surrounding cells. Because our aim was to measure individualcell growth, it was essential that the daughter cells were removed,a method described by Kjelleberg et al. (7).

A common approach in single-cell studies is to put a bacterialsuspension onto an agar surface and let the liquid evaporate,thereby attaching the cells. This simple method gives rise toa number of problems; the agar is not rigid enough, and it movesand looses water when it is exposed to heat from the sourceof illumination, making proper time course studies difficult.Rapidly growing bacteria tend to outgrow their neighbors, makingtemporal and spatial analysis difficult. These problems canbe solved by using a flow chamber in which the cells are attachedto a solid surface and exposed to a sufficiently fast flow tofeed the bacteria and remove any newborn cells without removingsignificant numbers of the original bacteria. By using dark-fieldillumination and a low magnification, a large number of bacteriacan be monitored in each experiment.

Our flow cell design made it possible to study some visiblephysiological characteristics of individual cells, as well asthe distribution of cells in the population, since many cellscould be observed. The characteristics studied included thesize of the cells before and after division, the variation insuccessive generation times of the same cell, the spatial andtemporal distributions of the cell size, and the generationtime within the population. By changing the culture conditions,such as temperature, pH, medium components, vitamins, antibiotics,toxins, etc., the effects of these factors on the visible physiologicalstate of the cells could be investigated.

MATERIALS AND METHODS

Top

Materials and Methods

Inoculum preparation. (i) Escherichia coli.
One colony of E. coli strain MRE C600 was picked from an agarplate and inoculated into a flask containing 100 ml of Luria-Bertanimedium (10 g of tryptone per liter, 5 g of yeast extract perliter, 10 g of NaCl per liter; pH 7.2) with 0.2% glucose. Theflask was shaken for 18 h at 37°C until the culture reachedthe stationary phase. This culture was used to inoculate theflow chamber.

(ii) Listeria innocua.
L. innocua strain Li73/99 growing on a stock tryptone soya agar(Oxiod CM131) slope was subcultured in tryptone soya broth (OxoidCM129) and incubated for 24 h at 30°C. This culture wasdiluted 1:1,000 in peptone salt dilution fluid (1 g of BactoPeptone [Difco Laboratories] per liter, 8.5 g of NaCl per liter),and 10 µl was inoculated into 9 ml of TSYGB (30 g of tryptonesoya broth per liter, 3 g of yeast extract per liter, 10 g ofglucose per liter). The tubes were incubated for 48 h at 22°C(stationary phase).

Flow system.
The flow system consisted of a feed flask (Schott, Mainz, Germany),autoclavable tubing, a peristaltic pump (P4; Belach Bioteknik,Stockholm, Sweden), a bubble trap manufactured in house, theflow chamber, and a waste flask. The setup is shown in Fig.1.

View larger version (112K):
[in this window]
[in a new window]
FIG. 1. Setup of the flow system. a, feed flask; b, peristaltic pump; c, bubble trap; d, microscope equipped with a charge-coupled device camera; e, flow chamber; f, waste flask.

Flow chamber.
The flow chamber consisted of five main parts, as shown in Fig.2. The device was tightly screwed together with eight steelscrews. The cross section was 10 by 0.1 mm, and using a mediumvolumetric rate of 0.7 ml/min resulted in an average linearvelocity of approximately 1.1 cm/s.

View larger version (18K):
[in this window]
[in a new window]
FIG. 2. Flow chamber. (a) Top block of aluminum with inlet and outlet pipes (diameter, 1.2 mm) with two O rings; (b) polycarbonate slide with two holes (diameter, 1.2 mm); (c) polymer spacer; (d) microscope slide; (e) bottom block of aluminum.

Slide preparation.
The microscope slides used were either glass slides that weremade hydrophobic or polystyrene slides.

(i) DDS coating.
An ordinary glass slide (Knittel Gläser, Braunschweig,Germany) was made hydrophobic by dip coating it in 2% dimethyldichlorosilane(DDS) dissolved in 1,1,1-trichloromethane (LKB, Bromma, Sweden).The coated slide was allowed to air dry in a laminar flow hoodbefore it was inserted into a flow chamber.

(ii) Sterile polystyrene slides.
Sterile polystyrene cell culture slides (16004; Nalge Nunc International,Naperville, Ill.) were used as delivered.

Flow chamber and tubing preparation.
The tubing and flow chamber were autoclaved separately and assembledin a laminar flow cabinet. The setup was connected to the peristalticpump, and the system was flushed (0.7 ml/min) with sterile phosphate-bufferedsaline (PBS) (0.2 g of KCl per liter, 0.2 g of KH₂PO₄ per liter,1.15 g of Na₂PO₄ per liter, 8 g of NaCl per liter; pH 7.3).

Inoculum.
By using a 1-ml syringe and a 0.8-mm-diameter needle, 100 µlof culture was injected into the bubble trap through a rubbermembrane. The culture was pumped into the flow chamber, andthe bacteria were allowed to settle by switching off the pump.When there was sufficient adhesion (after 15 min), the pumpwas restarted, and all unattached bacteria were flushed away.

Effect of growth environment.
The effect of salt on the growth of E. coli was studied in threeexperiments with media (10 g of tryptone per liter, 5 g of yeastextract per liter) supplemented with 1, 2, and 4% NaCl.

Effect of sublethal heat shock on L. innocua.
The L. innocua culture was centrifuged at 3,100 x g at 22°Cfor 10 min, the supernatant was discarded, and the pellet wasresuspended in 0.5 ml of TSYGB. Tubes containing TSYGB (10 ml)were preheated to 52°C in a water bath. A cell suspension(100 µl) was injected into each tube, and the tube washeated for a predetermined time (1, 2, or 5 min) and then placein ice water to cool rapidly. In order to concentrate the cells,the heated culture was transferred to a centrifuge tube andcentrifuged at 3,100 x g at 22°C for 10 min. The supernatantwas discarded, and the pellet was resuspended in 0.5 ml of TSYGB.The suspension (100 µl) was used to inoculate the flowchamber through the bubble trap.

Growth.
Growth was initiated when the feed flask was changed from PBS(no growth) to growth medium (Luria-Bertani medium containingglucose or TSYGB). The pump speed was set at 0.7 ml/min, whichwas sufficient to feed the bacteria and remove newborn daughtercells without removing significant numbers of attached mothercells (Fig. 3a). Assuming that there was laminar flow in theflow cell and plug flow in the rest of the system, we calculatedthat the PBS was washed out within 1 to 2 min.

View larger version (25K):
[in this window]
[in a new window]
FIG. 3. Series of images obtained over time (a) and resultant graph (b). Taken at 5-min intervals, the images are close-up photos (magnification, x500) of a single E. coli cell dividing during the exponential phase. More than 1,000 such images and graphs were generated in an experiment and used to create the distributions shown in Fig. 4.

Microscopy and image analysis.
The flow chamber was mounted on the stage of a dark-field microscope(Zeiss Standard 25; Zeiss, Oberkochen, Germany) equipped witheither a x4 objective (Leiz, Wetzlar, Germany) or a x10 objective(Zeiss). A high-resolution (1040 by 1392 pixels) charge-coupleddevice camera (CoolSnap Pro cf; Roper Scientific, Trenton, N.J.)was mounted on the microscope, and the digital signal was transferredto a PCI interface card. An image analysis program (Image ProPlus; Media Cybernetics, Silver Spring, Md.) controlled thecamera to grab an image every 1 to 5 min depending on the requiredfrequency. The program could semiautomatically improve the imagequality by using built-in functions to remove background noiseand enhance object edges. Image Pro Plus was able to automaticallyrecognize clusters of brighter pixels as cells and to calculatethe area of every cell and designate every cell with x and ycoordinates. An in-house Visual Basic program transferred thesize (in pixels) and the x and y coordinates to a spreadsheet,where the data were analyzed further. The images were grabbedat a low magnification (x200) by using dark-field illumination.The area of the computer image was 1,657 by 1,238 µm,as measured with an objective micrometer scale (Zeiss). Thisarea was large enough to include as many as 1,500 to 3,000 bacteria.Although the low magnification resulted in very few pixels (20to 30 pixels) per cell, the resolution was sufficient for ourpurposes since we were only interested in observing the divisiontimes of the cells. Note that, as Lawrence et al. (8) pointedout, it is impractical to try to estimate the size of a cell(in micrometers) based on dark-field images, since the sizeis inflated three- to sevenfold.

Data analysis.
From the Excel output of Image Pro Plus identification of individualcells was unreliable since the order of the cells could changebetween images. Therefore, to ensure the correct correspondence,another in-house program was written in which the geometricdistance between the cells in consecutive images was utilized.The program, written in Visual Basic for Microsoft Excel, madeit possible to relate the fate of any cell in the image to onerow of the spreadsheet created. The resulting graphs (Fig. 3b)were evaluated by using another Visual Basic program that helpedrecognition of the division times.

All the Visual Basic programs described here are available uponrequest.

RESULTS AND DISCUSSION

Top

Results and Discussion

General observations.
We had extensive problems with bubbles forming in the tubingand in the flow chamber itself. To overcome these difficulties,we added a bubble trap and decreased the downstream diameterof the tube with a clamp. The latter modification made it possibleto maintain the pressure inside the flow chamber, which preventedbubble formation. Routine sterility measurements were takento avoid contamination for several days. In order to minimizemedium consumption, we decreased the thickness of the polymerspacer from 0.25 to 0.1 mm, and this had no effect on the laminarflow. The tightness of the chamber was significantly improvedwhen we changed the polymer spacer from polyethylene film toTeflon. Commercial polystyrene slides proved to be the mostreliable material of the materials tested (glass, DDS-coatedglass, and polystyrene slides). So far, we have proven thatthe method works for the gram-positive bacterium L. innocuaand the gram-negative bacterium E. coli. Preliminary tests withBacillus cereus and Saccharomyces cerevisiae have shown thatit can also be applied to other species. Polar attachment ofthe cells would have made the data analysis more complicated,but high-magnification studies showed that all bacteria werein a supine position due to the shear force of the flow.

Distribution of the doubling times of individual cells.
We acquired data for the doubling time of an individual cell,which was defined as the time that it took for the pixel sizeof a cell to double. Figure 4 shows how the salt concentrationof the medium affected the distribution of individual doublingtimes for E. coli cells. Both the mean and the variance of thedoubling time increased with higher salt concentrations.

View larger version (16K):
[in this window]
[in a new window]
FIG. 4. Distributions of first doubling times of E. coli in different salt concentrations (1, 2, and 4% NaCl). Each culture was inoculated from the stationary phase and then grown at the ambient temperature in medium containing 10 g of tryptone per liter, 5 g of yeast extract per liter, 2 g of glucose per liter, and NaCl.

The pixel size is a two-dimensional projection of the cell volume,so it is not linearly proportional to the cell mass. However,this simply means that it is not possible to calculate the specificgrowth rate of the population from the growth rates of singlecells, measured in pixels. The lack of linear proportionalitydoes not affect the distribution of the individual divisiontimes, so it can be a suitable (stochastic) variable to studythe effect of history and growth conditions on the distribution.

Effect of sublethal heat shock on L. innocua.
As expected, sublethal heat shocks were shown to cause an increasein the time to the first division of the individual cells (referredto as individual lag time below). However, by using our technique,we were able to show that not only the mean of the individuallag times but also the variance increased with the durationof the heat shock (Fig. 5). Furthermore, the data suggest thatthe relationship between the duration of the heat shock andboth the mean individual lag time and its variance is closeto linear.

View larger version (13K):
[in this window]
[in a new window]
FIG. 5. Effect of sublethal heat shock on L. innocua. Both the mean (•) and the variance ( ${circ}$ ) of the individual lag times increased with the duration of the heat shock.

Conclusions.
The method described here can provide quantitative data suitablefor statistical analysis of bacterial growth. The use of lowmagnification increases the area in each field of view, therebyproviding a large number of observations in the same experiment.The flow of the medium not only feeds the cells but also removesnewborn daughter cells. This allows the study of slowly growingcells which would otherwise be outgrown by cells that are growingfaster. It is also possible to study the generation times ofsingle cells. Métris et al. (10a) described the firstmodel for the growth and division of single cells that takesinto account the distribution of generation times, and withour method, we were able to support that model.

Measurement of the distribution times of single cells has recentlybecome a crucial question in the field of predictive microbiology(13). However, so far, only simulation studies have been performed(10) or observations of a maximum 100 to 200 cells have beenreported (5). Our method makes it possible to observe the distributiontimes of more than 1,000 cells and can decrease the error ofparameter estimation of mathematical models significantly. Thiscould contribute to more accurate understanding, modeling, andpredicting of the bacterial responses to environmental changes.

ACKNOWLEDGMENTS

This paper was prepared with funding from the EU Programme Qualityof Life and Management of Living Resources (project QLK1-CT-2001-01145)(BACANOVA).

We thank Susie George for checking the manuscript and Lars Petterssonfor manufacturing the flow chamber.

FOOTNOTES

* Corresponding author. Mailing address: Center for Surface Biotechnology, University of Uppsala, Husargatan 3, P.O. Box 577, S-751 23 Uppsala, Sweden. Phone: 018 471 00 00. Fax: 018 55 50 16. E-mail: andras.ballagi{at}ytbioteknik.uu.se.

REFERENCES

Top