Microfluidic Detection of Movements of Escherichia coli for Rapid Antibiotic Susceptibility Testing

Abstract

Various nanomechanical movements of bacteria provide a signature of bacterial viability. Most notably, bacterial movements have been observed to subside rapidly and dramatically when the bacteria are exposed to effective antibiotics. Thus, monitoring bacterial movements, if performed with high fidelity, could offer a path to various clinical microbiological applications, including antibiotic susceptibility tests. Here, we introduce a robust and ultrasensitive electrical transduction technique for detecting the nanomechanical movements of bacteria. The technique is based on measuring the electrical fluctuations in a microfluidic channel, which the bacteria populate. The swimming of planktonic bacteria and the random oscillations of surface-immobilized bacteria both cause small but detectable electrical fluctuations. We show that this technique provides enough sensitivity to detect even the slightest movements of a single cell; we also demonstrate an antibiotic susceptibility test in a biological matrix. Given that it lends itself to smooth integration with other microfluidic methods and devices, the technique can be developed into a functional antibiotic susceptibility test, in particular, for urinary tract infections.

INTRODUCTION

Almost all bacteria exhibit nanoscale mechanical movements. These movements are generated by a number of different mechanisms and forces: for instance, some motile bacteria rely on their flagella for self propulsion, while others glide smoothly over surfaces. Even non-motile bacteria and bacteria immobilized on surfaces move incessantly by randomly oscillating [1, 2]. Recent studies [1–4] suggest that nanomechanical movements of bacteria and their viability are strongly correlated. In these experiments, nanomechanical movements of different bacteria subsided, shortly after effective antibiotics were administered. These observations open up the possibility that the viability of bacteria could simply be assessed from their movements, provided that these movements can be detected with sufficient sensitivity.

The most common assessment of bacterial viability is based on monitoring the growth of bacteria. For instance, common clinical antibiotic susceptibility tests (ASTs) rely on bacterial outgrowth from patient samples either in broth cultures or on agar plates in the presence of selected antibiotics. The degree of proliferation is measured after the culture step (e.g., by an optical technique), directly informing antibiotic susceptibility. Most emerging ASTs rely on this principle of growth detection, but aim to shorten the culture step by using sensitive measurement techniques that allow observation of tiny growth rates. In recent experiments, small bacteria populations and even single cells have been cultured in microchannels [5–12], in microdroplets [13, 14], on microbeads [15–19], and on microcantilevers [20], and the proliferation rates have been measured using a variety of ingenious techniques [21, 22].

In contrast to growth-based assays, monitoring bacterial movements could provide rapid assessment of bacterial viability. To accomplish such a feat, one needs a transducer that converts bacterial movements into detectable signals. First and foremost, such a device must come with an ultrasensitive transduction mechanism so that one can detect the slightest movements of small bacteria populations and even single cells. In addition, this sensitivity must be achievable over a large dynamic range, because the movements of motile bacteria have larger amplitudes than those of non-motile and immobilized bacteria. For broad-based clinical application, complex detection steps such as microscopy or molecular labeling are best avoided. Furthermore, one can take advantage of all the currently-available technologies for filtering, capturing and processing bacteria from patient samples, if the transducer can be seamlessly integrated with standard microfluidics. Most recently, two notable transducers, one based on micromechanics and the other on nanooptics, have been applied to the detection of bacterial movements. In the first [1–3], bacteria were immobilized on the surface of a microcantilever. The nanomechanical fluctuations of the bacteria increased the microcantilever oscillation amplitude, which was detected using optical methods. As noted above, these random oscillations of the microcantilever could be correlated with bacterial viability by exposing the bacteria to different antibiotics [1–3]. In the second, a single bacterium was adhered on an antibody-coated gold surface and the cell movements were detected by looking at the plasmonic image contrast [4]. As in the microcantilever experiments, the observed movements subsided in effective antibiotics.

Here, we present a simple and unique approach to detect the movements of a small bacteria population (generally 1 – 50 cells), regardless of whether the cells are planktonic or surface-immobilized. In this approach, bacterial movements directly impact the electrical resistance of a microfluidic channel filled with an electrolyte solution (i.e., phosphate-buffered saline or urine). Figure 1 shows an illustration of the device and the transduction principle. At the center of the device is a small microfluidic channel (constriction) with an effective cross-section w × h ~ 1 μm × 1 μm and length l ~ 20–100 μm; the microchannel region is connected to mm-sized reservoirs on both ends (Figure 1a) via larger channels. The reservoirs are used for insertion of electrical probes and sample input. The voltage drop across the microchannel is monitored while a small but constant-amplitude ion current passes through the microchannel in a standard 4-probe measurement (Figure 1a) [23]. When bacteria are in the microchannel, their movements change the effective channel diameter and the electrical conductance, resulting in voltage fluctuations. Figure 1b illustrates the source of electrical fluctuations for planktonic motile bacteria: the conductance is modulated by bacteria entering and exiting the microchannel as well as bacteria swimming inside (Figure 1b). These voltage fluctuations are similar to those in a Coulter counter [24], but there are no imposed flows in our device during the detection. In other words, the source of the observed signal here is the inherent movements of live bacteria. The Brownian movements of dead bacteria do not significantly contribute to the signal. Figure 1c illustrates a complementary measurement approach for bacteria immobilized on the microchannel walls or captured inside the microchannel. Here, the random nanomechanical oscillations of the bacteria give rise to similar electrical fluctuations by modulating the microchannel conductance.

FIG. 1.

(a) Illustration of the microfluidic transducer chip and its circuit model. The PDMS chip is bonded onto a glass substrate. There are four mm-scale reservoirs connected to 100-μm-depth channels; at the center is the tiny microchannel. A constant-amplitude electric current is injected into reservoir 1 and flows through the microchannel to the electrical ground (reservoirs 3 and 4). The voltage drop between reservoirs 2 and 4 is measured. The contact resistances in the system do not alter the measured signals. The electrical impedance of the microchannel is dominated by its resistance at the low-frequencies used in this study; the effects of the parasitic capacitance are thus neglected. (b) When a planktonic bacterium passes through the microchannel, the electrical resistance changes. (c) Collective nanomechanical fluctuations of bacteria, trapped in the microchannel or adhered to the microchannel surface, also modulate the resistance.

RESULTS & DISCUSSION

Our experiments always start with a measurement of the background electrical fluctuations (background noise) with the microchannel filled with only the relevant buffer solution. As shown in Figure 1a, we monitor the voltage drop on the microchannel as a function of time under a constant current of ~ 10 nA. The measured voltage has a slowly-varying component $\overline{V}$, which drifts on the order of minutes, and a rapidly-fluctuating component δV (t). The gray trace in Figure 2a is the background voltage fluctuations δV (t) for a microchannel (inset) with linear dimensions w × h × l ≈ 1.5 × 2 × 20 μm³ filled with phosphate buffered saline (PBS). Here, the root-mean-square (rms) value of the background noise measured under a constant (7.1 nA) current is $\delta V_{\mathit{rms}}^0\approx 1.6\mu \mathrm{V}$, with the (mean) channel resistance $\overline{R}\approx 3.3\mathrm{M}\Omega$. (In other terms, $\delta V_{\mathit{rms}}^0$ is the standard deviation.) In all the following experiments, the values of the voltages and currents used are close to the above values, ensuring that the bacteria are not affected by the electric fields and current densities in the microchannel [25, 26]. Relevant parameters and all the details of our experiments can be found in Materials and Methods Section and Supplemental Material.

FIG. 2.

(a) Electrical fluctuations in a narrow microchannel with linear dimensions w × h × l ≈ 1.5 × 2 × 20 m³ and mean resistance $\overline{R}\approx 3.3\mathrm{M}\Omega$. The gray data trace is the background, while the red data trace shows the voltage fluctuations due to bacteria swimming from left to right. Inset shows an optical image in which three bacteria are about to enter the microchannel (see supplementary video S1). On the left side of the microchannel, there is a V-shaped connection to the larger channels (and the reservoirs). This connection guides the bacteria loaded from the left side reservoirs into the microchannel. (b) Similar data for a wider microchannel (w × h × l ≈ 7.5 × 2 × 100 μm³, $\overline{R}\approx 2.7\mathrm{M}\Omega$). In addition to the background (gray) and bacteria (red) data traces, the number of bacteria that are present inside the microchannel are shown (blue data points) as counted from optical images (inset). Both scale bars correspond to 6 μm. Note that, in both (a) and (b), the voltage fluctuations are displayed as positive (i.e., resistance increases). (c) Fractional resistance change observed in different devices, when a single bacterium passes though the microchannel. Each data point (and error bar) comes from the average of ~ 20 measurements; the x error bars reflect the uncertainty in the linear dimensions of a bacterium.

The red data trace in Figure 2a shows the observed voltage fluctuations, after a PBS solution containing ~ 10⁷ colony forming units/mL (CFU/mL) motile Gram negative Escherichia coli (E. coli) is introduced into the reservoirs of the device and the device is allowed to equilibrate. Each voltage spike in Figure 2a is caused by the instantaneous resistance increase δR(t) due to the passage of a single bacterium or several bacteria through the microchannel. The timescale of each spike is ~ 1 s and roughly corresponds to the time it takes for the bacteria to swim from one end of the microchannel to the other, as observed with an inverted microscope during the measurement (Figure 2a inset). Unlike a Coulter counter [24], no external forces are used to push the bacteria through the microchannel; instead, the motile free-floating bacteria propel themselves through the channel using their flagella, indicating that the observed signal directly corresponds to inherent bacterial movements.

Figure 2b shows the voltage fluctuations due to the swimming of several E. coli inside a relatively large microchannel (w × h × l ~ 7.5 × 2 × 100 μm³). As above, the gray data trace is the background, and $\delta V_{\mathit{rms}}^0\approx 2\mu \mathrm{V}$. The red data trace is due to the movements of bacteria obtained after filling the reservoirs of the device with a bacteria solution containing ~ 10⁶ CFU/mL. Both ends of the microchannel have large V-shaped inlets to guide the planktonic bacteria into the microchannel. The blue data points in the figure are obtained from simultaneous microscope images and show the number of cells inside the microchannel (e.g., inset shows an instant with six cells). The size and design of this device allows many bacteria to swim in and out of the microchannel at the same time, resulting in large resistance and voltage fluctuations. For the 75 s duration shown in Figure 2b, the rms value (standard deviation) of the voltage fluctuations with bacteria is $\delta V_{\mathit{rms}}^b\approx 20\mu \mathrm{V}$. In both Figure 2a and b, the voltage fluctuations δV^b(t) are arbitrarily displayed as positive (i.e., resistance increases) by taking the empty microchannel as the reference state. This is not the case in what follows.

Figure 2c shows how the observed resistance fluctuations in our devices depend upon the linear dimensions of the microchannels and a bacterium. Assuming that the mean resistance of the channel is $\overline{R}=\rho \frac{l}{A}$, where ρ is the resistivity of the solution filling the microchannel and l and A are respectively the length and the cross-sectional area of the microchannel, an expression can be found for the resistance change when a bacterium is in the the microchannel due to simple blockage:

$$\overline{R}+\delta R=\rho \left[\frac{l-l_B}{A}+\frac{l_B}{A-A_B}\right].$$ (1)

Here, l_B and A_B are the length and cross-sectional area of the bacterium, through which the electrical resistance is assumed large. This leads to $\frac{\delta R}{R}\approx \frac{l_B}{l}\left[\frac{1}{\frac{A}{A_B}-1}\right]$. To test this expression, we measure $\overline{R}$ for devices with different dimensions (20 μm < l < 200 μm; 1.5 μm < w < 7.5 μm; h = 2μm); we also determine δR for each device when a single E. coli passes through the microchannel. Figure 2c shows $\delta R/\overline{R}$ as a function of the dimensionless comparison scale, $\frac{l_B}{l}\left[\frac{1}{\frac{A}{A_B}-1}\right]$, for each device. The average length and cross-sectional area of a bacterium, assuming a cylindrical shape, are taken as l_B ≈ 2 ± 0.25 μm and A_B ≈ 0.785 ± 0.157 μm² [27]. The dashed line shows y = x, indicating that this first-pass analysis can relate the observed resistance changes to linear dimensions. However, as we show below, there are more subtle factors at play when the device transduces bacterial movements into electrical signals: the amplitude of the cell movement is strongly reflected in the measured electrical signal — in addition to the linear dimensions of the cell.

To evaluate the sensitivity of the device, we measure the voltage fluctuations while observing the movements of a single E. coli bacterium inside the microchannel with a microscope. E. coli are motile and are known to move either by swimming or tumbling, depending on the context [27, 28]. We present data in which a bacterium stays in the microchannel for ~ 60 s and performs these different movements. Figure 3a shows optical snapshots of the bacterium inside the channel (See the supplementary movie S2); the corresponding voltage fluctuations during this 60 s duration are shown in Figure 3b, starting with the background fluctuations. The bacterium swims and tumbles in the microchannel several times in the first ~ 13 s. This corresponds to the time interval between 60 s ≤ t ≤ 73 s in Figure 3b. Each of these tumbles give rise to noise spikes, marked by the arrows in Figure 3b. These tumbling movements result in ~ 90° and ~ 180° rotations, which generate spikes with slightly different amplitudes and durations (~100–200 ms). The amplitude of the spikes probably depend upon the dynamics of the tumble and how the bacterium blocks the microchannel during the tumble. The bacterium continues to swim in the same direction after the tumbles and eventually reaches the left end of the channel at t ≈ 73 s. It sticks to the microchannel wall there (the last optical image in Figure 3a) and randomly oscillates on the wall for the next ~ 45 s (75 s ≤ t ≤ 120 s). An analysis of the voltage fluctuations shows that even these slightest oscillatory movements are detectable in the electrical signal. The rms voltage fluctuations (standard deviations) for the background is $\delta V_{\mathit{rms}}^0\approx 1.87\mu \mathrm{V}$. For 60 s ≤ t ≤ 75 s and 75 s ≤ t ≤ 120 s, during the tumbling movements and random oscillations, $\delta V_{\mathit{rms}}^t\approx 2.87\mu \mathrm{V}$ and $\delta V_{\mathit{rms}}^{\mathit{osc}}\approx 2.27\mu \mathrm{V}$, respectively.

FIG. 3.

(a) Optical images of a single E. coli bacterium swimming inside a microchannel (see supplementary video S2). Bacterium enters the microchannel from the right side, and swims to the left, tumbling several times and eventually sticking to the right end of the microchannel. The size of the microchannel is w × h × l ≈ 7.5 × 2 ×100 μm³ with a mean resistance of $\overline{R}\approx 2.7\mathrm{M}\Omega$. Scale bar is 13 μm. (b) The corresponding voltage fluctuations: background (black) and with bacterium (red). The arrows show the electrical signatures of the tumbling movement. Inset shows the rms values (standard deviations) for the background and the different movements of the bacterium.

Having shown that even the nanomechanical oscillations of a single bacterium can generate detectable electrical signals, we attempt to capture a small population of bacteria in the microchannel region and perform measurements on these cells. Our eventual goal, as will be shown below, is to detect the susceptibility of this small population to administered antibiotics. To achieve this goal, we either jam the bacteria or immobilize them by surface functionalization in the microchannel, and then measure the electrical signals. In these experiments, we use untreated microchannels or microchannels coated with poly–D–lysine (PDL) [29, 30]. To capture the bacteria, we flow the bacteria solution through the microchannel by establishing a pressure gradient, i.e., by applying positive pressures to reservoirs 1 and 2 in Figure 1a. This flow can be treated as a granular flow in which the grains are the bacteria. After a few bacteria clog the microchannel, 35 ± 5 more bacteria pile up in the V-shaped region, as shown in the inset of Figure 4b. Once the bacteria are captured, they do not swim away but tend to continue their random oscillations on the surface (see supplementary video S3). In Figure 4a, we present the voltage fluctuations before the bacteria are introduced (black) and after the capture (red). The rms values are $\delta V_{\mathit{rms}}^0\approx 1.56\mu \mathrm{V}$ and $\delta V_{\mathit{rms}}^b\approx 4.39\mu \mathrm{V}$, respectively. Once the channel is clogged, the mean resistance value $\overline{R}$ also increases by ~ 40 %. The inset shows that the background electrical noise is Gaussian as expected. Moreover, the fluctuations due to bacteria are Gaussian, presumably because they come from a number of almost independent events (i.e., bacteria). We estimate that the bacteria in both the V-shaped region and the microchannel modulate the electrical resistance by their movements, but the largest contribution comes from the few bacteria captured inside the microchannel. We note that we have obtained very similar results in microchannels coated with an adhesion-promoting layer, such as poly-D lysine [11, 12].

FIG. 4.

(a) Time-dependent electrical fluctuations (red trace) due to nanomechanical movements of ~ 35 cells (E. coli) captured in the microchannel shown in the rightmost inset (w × h × l ≈ 1.5 × 2 × 20 μm³; $\overline{R}\approx 4.5\mathrm{M}\Omega$; scale bar is 6 μm; see supplementary video S3). Black trace is the background registered before the bacteria are introduced. The inset is the normalized probability density functions (PDF) of the data; the dotted line is a Gaussian. The x-axis is in units of standard deviation (each data set normalized with its standard deviation). (b) Power spectral density (PSD) of the noisy signals in (a) showing the frequency content.

A frequency domain analysis can provide more insight into these noise-like voltage signals. Figure 4b shows the Power Spectral Density (PSD) of the fluctuations in Figure 4a obtained numerically from the time-domain signal. The band-pass filter in the electronics only passes the voltage fluctuations in the frequency range 0.1 Hz < f < 16 Hz. Most of the noise power coming from the bacterial movements is at low frequencies in the range 0.1 Hz ≲ f ≲ 10 Hz; the noise power appears to scale as ~ 1/f² for 1 Hz < f < 10 Hz but saturates at very low frequencies. These observations are consistent with previous studies [1, 4].

Capturing and/or immobilizing the bacteria in the microchannel region (e.g., by a pressure-driven flow and/or surface functionalization) increases the effective local density of the bacteria by orders of magnitude in a time frame of minutes. For instance, the volume of the region shown in the inset of Figure 4a is approximately ~ 5 × 10⁻⁹ ml and the number of bacteria in that region is approximately 35 ± 5. This results in a local bacterial density of ~ 10¹⁰ CFU/ml. This may be an advantage for future tests by reducing the time for cell culture and sample purification steps.

Finally, we show how this approach and device can be used for antibiotic susceptibility testing in a biological matrix. Two strains of E. coli are used in these experiments: one is sensitive to nalidixic acid (Figure 5a; ATCC 25922) and the other is resistant to the antibiotic (Figure 5b; ATCC 700609); these two strains are tested in two separate but identical chips. In both Figure 5a and b, the black data traces show the background noise taken with just the matrix, which is a mixture of human urine and tryptic soy broth or nutrient broth. After the background measurements, ~ 50 cells of each strain are captured in the two chips as described above (see Figure 4). The fluctuations before the antibiotics are administered (Before-Abx) in Figure 5a and b are taken in the mixture of urine and broth. Then, the measurement is repeated, as a function of time, in the same mixture but with nalidixic acid (50 μg/mL) added at above the minimum inhibitory concentration [31]. Shown in Figure 5a and b are representative one-minute data traces as a function of time. Figure 5c and d show the rms values of all the measured one-minute-long fluctuations for these two strains as a function of time.

FIG. 5.

Antibiotic susceptibility testing with two different E. coli strains in human urine. Both experiments are performed at the same concentration of nalidixic acid (50 μg=ml) in two identical microfluidic chips (w × h × l ≈ 1.5 × 2 × 20 μm³). (a) The fluctuations of the susceptible strain (E. coli ATCC 25922) decay as a function of time. The black data trace is the background, and the red traces are collected right before and after the administration of the antibiotic. Here, $\overline{R}\approx 2.9\mathrm{M}\Omega$ with bacteria. (b) Similar fluctuations of the resistant bacteria (E. coli ATCC 700609). Here, $\overline{R}\approx 4.6\mathrm{M}\Omega$ with bacteria. (c)–(d) The standard deviations of the fluctuations of the two strains as a function of time.

The fluctuations for the susceptible strain decay monotonically [Figure 5a; also see supplementary videos S4 (before antibiotics) and S5 (105 minutes after antibiotics)]. The movements of the resistant strain, on the other hand, do not show significant changes as a function of time [Figure 5b; supplementary videos S6 (before antibiotics) and S7 (105 minutes after antibiotics)]; $\delta V_{\mathit{rms}}^r\approx 8.36\mu \mathrm{V}$ remains constant to within 14 %. To confirm the results, the measurements on both chips are repeated after 24 hours. The small discrepancy between the standard deviations of the background fluctuations and the fluctuations of the susceptible strain after 24 hrs is mostly due to the change in $\overline{R}$ by ≈ 20% (due to drifts, evaporation, dead cells breaking down, etc.), which changes the noise power coupled to the electronics. The magnitude of the voltage fluctuations before antibiotics are different between the two strains (Figure 5a and b). This is not only due to the difference in the number of bacteria captured in the two experiments; the way the bacteria are trapped and wiggle inside the microchannel (e.g., sideways or out of plane) also impacts the measured signal. We have also observed the resistant strain to be somewhat less motile compared to the susceptible strain. Finally, the bacteria population captured in the microchannel mostly remain in the microchannel during the course of the experiment. It is also important to note that the resistant population captured in the microchannel proliferates over time (the number of cells increases), but the signal stays constant. This is because most of the signal comes from the narrower regions of the microchannel and the number of cells in these regions do not change significantly.

We have also measured the antibiotic response of the susceptible E. coli strain (ATCC 25922) to different nalidixic acid concentrations. We use the same experimental procedure and devices as above in Figure 5. Here, the solution filling the device is a mixture of PBS and tryptic soy broth. Figure 6 shows the standard deviations of voltage fluctuations as a function of time at different antibiotic concentrations. The antibiotic concentrations are all at or above the minimum inhibitory concentration [31]. At the two highest antibiotic concentrations (50 and 6.25 μg/mL), the measured standard deviations decrease rapidly after the antibiotics are administered; the standard deviations at lower concentrations decrease at a slightly longer time scale. This may be attributable to the fact that nalidixic acid is bacteriostatic at lower concentrations and bactericidal at higher concentrations [32]. We have not observed a consistent monotonic decay (e.g., an exponential decay) of standard deviation values (i.e., cell viability) with increasing antibiotic concentration at fixed times [33]. We also note that the response of E. coli (ATCC 25922) to nalidixic acid depends upon the solution [34]: standard deviations in PBS (Fig. 6 yellow data) appear to decay faster than those in urine (Fig. 5c) at 50 μg/mL. More experiments are needed to fully understand the correlation between viability and antibiotic concentrations, including the role of the solution.

FIG. 6.

Time-dependent standard deviations of the susceptible E. coli strain (ATCC 25922) at various antibiotic concentrations. The solution in the device is a mixture of PBS and tryptic soy broth Here, w × h × l ≈ 1.5 × 2 × 20 μm³ and $\overline{R}\approx 3.8\mathrm{M}\Omega$ with bacteria.

CONCLUSIONS

This work describes a novel microfluidic approach for transducing nanomechanical movements of bacteria into electrical signals. Given that the viability of bacteria is directly correlated with their movements, the approach may eventually be useful for rapid antibiotic susceptibility testing. The results in Figure 5 obtained are particularly promising: E. coli is the most common pathogen for urinary tract infections [35, 36], and the device could be developed into a clinical test that works directly with unprocessed urine samples with low bacteria concentrations (i.e., ≲ 10⁵ CFU/ml). Because the required urine volume is small (< 0.5 ml) and the electrical measurement is straightforward, the test can be multiplexed to provide the susceptibility of the pathogen to multiple antibiotics in parallel. The device may also be integrated with “front end” microfluidics in order to perform some degree of sample processing. One could thus estimate the susceptibility of bacteria to 4 – 5 major classes of antibiotics from unprocessed urine samples in ~ 1 hr at the point of care. Finally, the approach could be developed for other bodily fluids, infectious exudates from wounds or even washings of suspect medical devices (e.g., catheters) in order to rapidly detect bacterial presence and antibiogram.

Sensitive novel probes and devices for detecting bacterial movements can also provide biological scientists with an important research tool. For instance, the study of bacterial movements in confined spaces and nanochannels have so far relied on high-resolution microscopy [37–39]. Measurement techniques such as ours could help uncover some of the subtle movements that may be essential for bacterial survival in such environments.

Finally, the method introduced here could be further developed to work with non-motile bacteria and with bacterial biofilms. Because non-motile bacteria (e.g., Staphylococcus aureus) have been reported to move with significantly smaller amplitudes [2], observing the movement signals here would be a more challenging task. Likewise, bacteria inside a biofilm are expected to move very subtly, if at all. In summary, our unique and novel approach complements the rich variety of impedance measurements [40–42] in microbiology.

MATERIALS & METHODS

Materials

Lyophilized pellets of Escherichia coli (ATCC 25922, and ATCC 700609) were purchased from ATCC. Tryptic Soy Broth (TSB), and Tryptic Soy Agar (TSA) were purchased from Sigma Aldrich; Nutrient Broth (NB) and Nutrient Agar (NA) were bought from Beckton Dickinson. Highly-purified glycerol was purchased from MP Biomedicals. Phosphate buffer saline (1X PBS) (without Calcium, Magnesium) was purchased from Lonza Walkersville Inc. Nalidixic Acid Sodium Salt was purchased from Alfa Aesar. Human urine (male, viral negative) was purchased from BioreclamationIVT.

Preparation and Growth of Bacteria

Lyophilized bacteria were re-solubilized with 1 mL of broth (TSB for ATCC 25922, and NB for ATCC 700609); the solution was aspirated from the vial and mixed with 5 mL of designated broth for each bacteria in a 15 mL centrifuge tube. The tube was kept in an incubator-shaker at 30°C and 120 rpm for 24 hours. After 24 hours, the turbid bacterial solution was centrifuged for 7 minutes at 2500 rpm, and the bacteria pellet was re-suspended in a mixture of 2.5 mL designated broth, 2 mL 1X PBS and 0.5 mL highly purified glycerol. This final bacterial solution was collected in small aliquots of 150 μL and frozen at −80°C.

For the experiments, E. coli was cultured overnight at 37°C on an agar plate (TSA for ATCC 25922; NA for ATCC 700609). A single colony was harvested and grown in broth (TSB for ATCC 25922; Nutrient Broth for ATCC 700609) overnight in a shaker at 160 rpm at 37°C. The next day, a small amount of bacteria solution was taken from the tube and put into a fresh TSB medium; the solution was subsequently grown in a shaker at 160 rpm at 37°C. The solution was used when its optical density at 600 nm (OD₆₀₀) reached ~ 0.4 – 0.8. For the experiments, the bacterial solution was diluted to the desired concentration (~10⁵ – 10⁷ CFU/mL). In order to verify the density of our bacterial solutions, we periodically measured Colony Forming Units (CFU) from OD₆₀₀ by preparing serial dilutions and plating on agar plates.

Antimicrobial Preparations

The antibiotic stock solution was prepared in accordance with the supplier’s recommendations. The stock solution was kept at −20°C and, when needed, was thawed at room temperature. It was then diluted to the desired concentration in DI water.

Device Fabrication

The process for manufacturing our microfluidic device started with the fabrication of the master mold. The mold was prepared using two steps of conventional photolithography. Briefly, a 3-inch Silicon wafer was first immersed in Piranha solution (H₂SO₄:H₂O₂ = 3:1) for 20 minutes to clean its surface. Then, the wafer was baked on a hot plate for 5 minutes at 200°C to remove the water from its surface. SU-8 2 was applied to the wafer by spin coating it at 2000 rpm for 30 seconds. This resulted in a film of thickness 2 μm. After a soft-bake step (1 min at 100°C), the wafer was exposed in a mask aligner (Karl Suss MA6) in soft-contact mode. This exposure to UV light (for 4 seconds at 275 W of power) defined the microchannel region of the device with the small features (central region in Figure 1a in main text). After post-exposure bakes, the photoresist was developed in MicroChem SU-8 developer for 25 seconds. Pressurized gas was used to remove the solvent from the surface. This completed the fabrication of the microchannel on the mold. In the second part, macrochannels of 100 μm thickness were fabricated. To do this, the wafer was spin coated with SU-8 50 at 1000 rpm for 60 seconds. The wafer was then soft-baked at 100^◦C for 30 minutes. Next, the wafer was put into the mask aligner (Karl Suss MA6), and the mask for the macrochannels was aligned with custom marks on the wafer. The wafer was exposed to UV light for 80 seconds at 275 W. Two steps of post exposure bake were performed, as above. The wafer was then developed in MicroChem SU-8 developer for 10 minutes and rinsed with fresh developer.

The microfluidic chips were obtained by curing Polydimethylsiloxane (PDMS) on the mold. PDMS prepolymer (Dow Corning Sylgard 184) was mixed thoroughly with catalyst at a 10:1 ratio by weight in a disposable container for about 5 minutes. Then the mixture was put into a vacuum desiccator for 30 minutes to remove all the bubbles. The bubble free PDMS mixture was then slowly poured over the master. After 1 hour in a 90°C oven, the PDMS hardened. Next, the PDMS structure was carefully peeled off from the master. The PDMS was cut into individual chips with complete devices on each, and reservoir holes were punched on the macrochannels. Before bonding, the PDMS chips were cleaned by pressurized air and adhesive tape. Immediately before an experiment, the PDMS chip was activated in a plasma asher and then bonded to a clean glass slide.

Capturing the Bacteria in the Microchannel

In these experiments, first background measurements were performed using buffer electrolyte solution. Then, the background solution was aspirated, and the V-shaped side of the chip was filled with a bacteria solution of density ~ 10⁷ CFU/mL. A pressure gradient was applied by simply filling the V-shaped side of the chip with 5 μL more solution. It took a few minutes for the bacteria to become captured in the microchannel. The bacteria were tracked with the microscope while they propelled towards the microchannel. The excess bacteria were removed from the reservoirs by aspirating the bacterial solution and filling the chip with the same amount of the background solution in all reservoirs.

Antibiotic Susceptibility Experiments

In the antibiotic susceptibility experiments, we used a mixture of human urine and TSB (80% urine, 20% TSB, v/v) for susceptible E. coli strain (ATCC 25922); human urine and NB (50% urine, 50% NB, v/v) for resistant E. coli strain (ATCC 700609). After obtaining background measurement, the solution was aspirated and the V-shaped side of the chip was filled with bacteria solution in the same human urine-broth mixture (80% human urine, 10% TSB, 10% bacteria solution v/v for the susceptible strain; 50% human urine, 40% NB, 10% bacteria solution v/v for the resistant strain). After bacteria were captured in the microchannel, the excess bacteria were removed from the reservoirs and the chip was filled with the urine-broth mixture. Voltage fluctuations were recorded, then the solution was aspirated from the chip, and the same solution but with 50 μg/mL nalidixic acid added (80% human urine and 20% TSB for the susceptible strain; 50% human urine and 50% NB for resistant strain) was re-introduced. The voltage fluctuations over time were recorded to quantify the effects of the antibiotic on the E. coli strains. All the experiments were conducted at room temperature.

During the experiment, the microchannel region was constantly monitored in an inverted optical microscope. We confirm that the cells captured/immobilized in the microchannel mostly remained in the microchannel. It is important to note that the bacteria did not escape the microchannel even during the removal and addition of the various solutions, which created transient flows.

We kept all our devices from antibiotic susceptibility experiments overnight to observe whether or not bacteria proliferated in them. In effective antibiotics, no proliferation was observed; the resistant strain, on the other hand, proliferated, consistent with our expectations.

Image Collection

An inverted microscope (Olympus IX 81) with a 40X Olympus objective (numerical aperture 0.6) was used to track the movements and fluctuations of bacteria inside the microchannels. Video images were obtained by a high speed camera (Hamamatsu Orca-Flash4 C11440). The images were analyzed by either Olyvia, CellSens or ImageJ.

Electrical Measurements

The circuit diagram and an illustration of the device are shown in Figure 1a. The devices were filled with the desired solutions (e.g., PBS, human urine). Ag/AgCl electrodes were inserted into the reservoirs (Figure 1a) for generating the ion current and for sensing the voltage drop. A lock-in amplifier (Stanford Research SR-830) was used for the measurement. The lock-in oscillator output (at a frequency of f_c = ω_c/2π ≈ 10 Hz and an amplitude of 1 V) was connected to a 100 MΩ resistor to create a current source with a magnitude of I ≈ 10 nA. This current was passed through the microchannel to ground via reservoirs 1 and 3 in Figure 1a. The second pair of Ag/AgCl electrodes in reservoirs 2 and 4 were connected to the lock-in input (in parallel with the 10 MΩ input resistance) and sensed the voltage drop across the microchannel. Since this was a 4-wire configuration, contact resistances, resistances of the macrochannel regions and other parasitic resistances did not strongly affect the measured voltages. In the experiments, the electric field and the current density in the microchannel remained in the range 0.2 – 3 × 10³ V/m and 6 × 10² – 3 × 10³ A/m², respectively.

The resistance of the channel can be expressed as $R(t)=\overline{R}+\delta R(t)$, where $\overline{R}$ is the slowly-varying component of the resistance and δR(t) comes from the bacterial movements in our experiments. The resistance fluctuations can be expanded in terms of Fourier frequencies as

$$\delta R(t)\approx {\displaystyle \sum _k\delta R_k\cos \left({\omega }_{k}t+{\phi }_k\right)},$$ (2)

where δR_k and φ_k can be treated as random variables. The lock-in amplifier, after mixing the input voltage down with the oscillator frequency at f_c, measured the voltage $\overline{V}+\delta V(t)\approx I\overline{R}+I\delta R(t)$, with the voltage $\overline{R}I$ at d.c. and the voltage fluctuations ∝ δR(t)I at frequencies within 0 ≲ f ≲ B. Here, B is the bandwidth set by the time constant of the lock-in amplifier, and in our experiments, B ≈ 16 Hz. The lock-in output was band-pass filtered using a voltage preamplifier (Stanford Research SR-560) between 0.1 and 30 Hz in order to remove the d.c. signal, the low-frequency drifts, and other high-frequency noise.

In order to validate that the fluctuatiosn are due to the movements of bacteria, we have additional performed experiments with microbeads (diameter ≲ 1 μm). We have observed that, even at high concentrations of ~ 10⁷ microbeads/mL, it is extremely unlikely for a microbead to diffuse into the microchannel (or diffuse out if it is initially inside) when there are no external forces or flows. For instance, at ~ 10⁷ microbeads/mL, no microbeads diffused into the microchannel over periods of ~ 20 minutes. Furthermore, particles that are undergoing Brownian motion inside the microchannel do not contribute to excess electrical fluctuations because they are in thermal equilibrium.

Data Collection

A data acquisition board (DAQ NI 6221 USB BNC) and Labview software were used to collect the voltages from the output of the lock-in amplifier and the voltage preamplifier. The results were analyzed with MATLAB and Origin. The power spectral densities were computed by calculating the auto-correlation function for the noise voltage and by Fourier transforming, performed in both MATLAB and Origin. A band block filter was applied between 10 and 10.4 Hz to remove camera noise.