1. Introduction

Antibiotic resistance poses a serious and growing public health threat that demands continual efforts to identify and synthesize agents with improved antimicrobial activities [1, 2]. In clinical settings, efficient, robust, and sensitive methods of diagnosing antimicrobial resistance and measuring bacterial susceptibility to candidate compounds are critical to both infection control and drug discovery. Traditionally, culture-based methods, such as broth dilutions, have served as the gold standards for antimicrobial susceptibility tests, especially when coupled with miniaturization and automation to achieve rapid processing [3]. Nevertheless, alternative methods that incorporate more recent advances in molecular and non-phenotypic techniques offer potential advantages of improved accuracy, lower cost, reduced sample size, and shorter test time [4–6]. Here we showcase optical tweezers as a possible tool for direct and accurate assessment of susceptibility at the single-cell level.

Optical tweezers [7] have found many biological, biomedical and atmospheric applications in the past decades [8–13]. For example, they have been used to manipulate and study whole cells from diverse branches of life and to examine the forces exerted by molecular motors inside these cells; moreover, they have been employed extensively as an effective tool to characterize bacterial chemotaxis and flagellar rotation, revealing new features of bacterial motility previously imperceptible via other approaches [14–18]. Optical manipulation has also been used previously to investigate the viability of single bacterial cells [19], but the full capability of the technique remains to be explored.

In this work, we employ optical tweezers to examine the lethal effects of varying concentrations of ethyl alcohol on an Escherichia coli (E. coli) cell. By tracking and carefully analyzing the dynamics of the trapped bacterium, the “killing time” for different ethanol concentrations can be extracted. Our optical trapping method, which combines Back-Focal-Plane (BFP) interferometry detection [20, 21] with auto-correlation function (ACF) statistics, yielded measurements of ethanol’s germicidal effects that agree well with previously published results obtained from culture-based techniques [22, 23]. Thus, our method offers effective and quantitative testing of antibacterial susceptibility at the single-cell level and possesses significant developmental potential for biomedical applications.

2. Experimental setup and method

Our home-built Optical Tweezers (OT) system is based on an inverted microscope (Axio Observer D1-Zeiss) which is schematically shown in Fig. 1. A continuous wave laser beam with TEM₀₀ transverse mode (Coherent, λ = 1064nm) is expanded (represented by L1 and L2 lenses in Fig. 1) before it is transmitted into the microscope through the side port. The expanded beam is coupled into the optical pathway of the microscope by a Dichroic Mirror (DM1) after which it is tightly focused into the sample chamber using an oil immersion objective lens (100X, Zeiss) with Numerical Aperture (NA) of 1.3. The sample chamber is an open-top glass bottom dish (MatTek) mounted on a piezo-electric stage (PI NanoXYZ Piezo Stage), providing positional control of the trap inside the chamber with nanometer resolution. To avoid hydrodynamic effects of the chamber wall [24], the trap is kept about 10µm away from the bottom of the dish. In order to exclude artifacts that might be caused by photodamage and heating [25–27], in addition to implementing other optimization techniques [28], the trapping laser power is kept at no more than 40mW at the sample site. The forward scattered light from the trapped bacterium as well as the direct (un-scattered) light is collected by a condenser lens (NA = 1.2). The collected lights are reflected off the optical pathway of the microscope by another Dichroic Mirror (DM2) and then focused on a Position-Sensitive Detector (PSD) situated at the plane conjugate to the BFP of the condenser. PSD signals are amplified by low-noise preamplifiers (Stanford Research Systems) before being transferred to the computer via an Analog-to-Digital (A/D) card (National Instruments). The PSD positional signals are acquired at a sampling rate of 10 kHz, using a custom-made LabVIEW program. In addition, the trapped bacterium is monitored with a CCD camera. In order to deliver the “drug” under study (ethyl alcohol) into the dish with precise control and minimum hydrodynamic force, a syringe pump is used with a tapered glass capillary (inner diameter of about 2μm, generated with a micropipette puller (Narishige)).

 

Fig. 1 Schematic of the experimental setup. An expanded 1064-nm laser beam is coupled into the microscope from the side port with a Dichroic Mirror (DM1). The expanded beam is tightly focused into the sample dish mounted on piezoelectric transducer (PZT).The forward scattered light from the trapped object is first collected by a condenser lens and then sent to a Position-Sensitive Detector (PSD) by reflection from another Dichroic Mirror (DM2). Low-noise preamplifiers and A/D card are used for signal conditioning and data acquisition. A microsyringe pump is used for delivering alcohol into the dish. The insert on the left illustrates a zoomed view of a single bacterium trapped by the optical tweezers. Insets on the right show typical dark-field images of an E. coli bacterium trapped by an optical tweezers. (a) When the bacterium is trapped inside the sample chamber, i.e., in bulk, the bacterium aligns itself along the laser beam axis, such that only a transverse projection is seen; (b) when it is gently pushed against the chamber wall, the cell changes orientation, showing clearly the entire rod-shaped body.

Download Full Size | PDF

Generally speaking, bacteria inside the sample chamber can swim, tumble, and move anywhere in the aqueous environment. A bacterium in the trap, however, tends to align itself along the optical axis [29, 30] and is typically driven by two kinds of activities: Brownian motion and flagella-mediated propulsion. Since the latter requires cellular energy, the movement of a dead or metabolically inactive E. coli cell in an optical trap can be attributed solely to passive Brownian motion. High temporal resolution of optical tweezers allows for accurate interpretation of these bacterial movements. The positional signal recorded from a trapped bacterium does not seem to have reasonable information; however, statistical analysis of the signal can deliver valuable information, such as rotation frequency of the flagella and the cell body [17]. Importantly, such analysis can determine whether the trapped bacterium is dead or alive, enabling accurate assessment of the effects of antibacterial agents. It should be noted that the trajectories of a trapped bacterium (PSD signals) are associated with several factors, including trap stiffness, cell size, medium viscosity, and dynamic strength of the bacterium [16–18]. Different cell sizes can change the corner frequency of PSD and the trap stiffness as well, but such factors should not affect the final results obtained from our autocorrelation analysis described below, because they are based on data obtained from the same single bacterium. Also, the presence of more than one bacterium in the trap canintroduce a complicating noise into the PSD signal. To prevent this, a trapped bacterium is moved close to the chamber wall and then gently pressed against the chamber wall, where it changes its orientation. A typical dark-field image of a trapped bacterium under such consideration is shown in Fig. 1. With this simple procedure we can identify whether there is only one bacterium or multiple bacteria in the trap. For all experiments described below, we took great care to ensure that all data acquired are from single cells in the trap.

In this work, our objective is to trap a single bacterium with optical tweezers and track its positional changes while introducing an antibacterial agent into the medium, thus allowing us to observe the consequences of bacterial exposure to the chemical agent. To verify the feasibility of our proposed technique, we used different concentrations of ethyl alcohol as the antibacterial agent and obtained the positional time series of a trapped E. coli cell after its exposure to alcohol. E. coli strain MG1655 was grown in Luria–Bertani (LB) medium at 37°C overnight and then diluted 10 times into the same medium at 25°C one hour before experiments. The cultures were further diluted 100 times (to less than 10⁵ cells per mL) before addition to the sample chamber to avoid bacterial accumulation at the trap during measurements. Since high concentrations of alcohol may destroy the cell body and cause morphological changes that affect the positional signals, the typical concentrations of ethyl alcohol used in our experiments were 20%, 22.5%, and 25% by total volume (i.e., the volume of the solution inside the chamber plus the volume of added alcohol).

3. Experimental results and analysis

In a typical experiment, alcohol is injected gently at a position far away from the trapped bacterium before data acquisition starts. Figure 2 shows the positional time series of a representative bacterium in aqueous solution after exposure to 25% ethanol. The acquired data is divided into 30sec sequences. Figure 2(a) shows three representative time sequences at different stages of the acquisition. For each sequence the power spectral density and the Auto Correlation Function (ACF) are calculated; the results are shown in Figs. 2(b) and 2(c), respectively. Although there is no clear difference in the positional time series [Fig. 2(a)] and its associated power spectrum [Fig. 2(b)], their ACF graphs are clearly distinguishable, both from their initial value and their decreasing rate (the slope of the graph) [Fig. 2(c)].The inset of Fig. 2(c) shows that despite long-time differences, the graphs have a very similar short-time behavior. In a few milliseconds, the ACF of each graph exponentially drops to a slowly varying value. The time-constant of this exponential function, which is inversely related to the corner frequency of the relevant power spectrum, is approximately equal for all three graphs.

 

Fig. 2 (a) Positional time series (trajectory) of a trapped bacterium in the lateral direction for 300 s (bottom trace, gray) after introducing 25% ethyl alcohol. Three different sections, at intervals of 0-30, 90-120 and 180-210 s, were expanded and represented in black, red and blue, respectively. (b) Calculated power spectrum and (c) Auto-Correlation Function (ACF) corresponding to the three sections in (a). (d) Normalized ACF (ξ) calculated for each 30s time series obtained at different times after adding the alcohol. The points are from calculation of measured data, while the red line is the sigmoidal dose-response-function fit.

Download Full Size | PDF

The first time sequence (black curve) represents the behavior of a living, non-damaged bacterium, while the third sequence (blue curve) mimics pure random Brownian motion, suggesting that at this time the bacterium loses its flagellar activity completely (the behavior of a dead bacterium). Both of the above mentioned cases were checked in separate control experiments. Considering that the ACF of a living bacterium has the largest area under the graph, we define the normalized ACF, ζ, as the area under the ACF graph divided by that of the living bacterium. Figure 2(d) shows the evolution of ζ with time. As time goes by, ζ decreases from its maximum plateau to the minimum baseline (corresponding to the time window of the lethal effect, as illustrated by the aqua-shaded region). We define the time it takes for the normalized ACF to reach half maximum as the “killing time.” Apparently this parameter can be used to quantitatively characterize the bacterium’s lethal response to alcohol exposure. In other words, the normalized ACF degrades with time, which would not happen for a living bacterium. The lethal response is better characterized by fitting our data to a sigmoidal dose-response function (red solid line in Fig. 2(d)), with the effective “killing time” of about 130 s (corresponding to the time at half maximum, marked by dotted line).

We expected that lower concentrations of alcohol would lead to a longer killing time. To validate this quantitatively, other concentrations (22.5% and 20%) were also tested. In addition, a control experiment was performed in which no alcohol was added (0%), to eliminate the possibility of lethal effects due to heating and bacterial photodamage [25–27]. Representative experimental results are summarized in Fig. 3, which shows that the measured killing time values become larger when the alcohol concentration is lowered. For autocorrelation analysis and comparison, it is crucial to avoid any other unwanted disturbance in the time signal, such as bacterial tumbling or the trapping of multiple cells. Thus, we repeated the experiments with different, randomly chosen bacteria from ten independent samples at each alcohol concentration. The average killing times for 25%, 22.5%, and 20% ethyl alcohol, as shown at half maximum of ζ by red dotted lines in Fig. 3, are 130 ± 20s, 230 ± 35 s, and 550 ± 50 s, respectively. (The standard deviations are extracted from measurements of ten independent samples.) Our results demonstrate clearly that higher concentrations of alcohol lead to faster killing and that such killing is not due to photodamage, as the bacterium remains alive for the whole data acquisition period (about 550 s) when no alcohol is added. In order to confirm the reliability of the current technique, we have also performed a series of experiments with another E. coli strain (KS272) and obtained similar results of alcohol concentration dependence.

 

Fig. 3 Normalized ACF (ξ) vs. time for different alcohol concentrations. The killing times for 25%, 22.5%, and 20% concentrations are determined to be 130 s, 230 s, and 550 s, respectively, and these are representative killing times of a single cell. For each concentration, measurements were done for at least 10 cells, yielding similar results with standard deviations. When no alcohol (0%) is added, the normalized ACF (ξ) does not change appreciably with time, indicating that the lethal effect is not due to photodamage.

Download Full Size | PDF

4. Conclusion

We have employed an optical tweezers to study the toxic effect of an antimicrobial agent on single bacterial cells. We demonstrated that the “killing time” of an E. coli cell coming into contact with ethanol in aqueous solution can be calculated from statistical analysis of its positional time signals, which allows distinction of bacterial flagellar motion from random Brownian motion. This technique may be further optimized to develop an effective tool for studying bacterium-drug interactions.