A microfluidic platform for rapid, stress-induced antibiotic susceptibility testing of Staphylococcus aureus

Abstract

The emergence and spread of bacterial resistance to ever increasing classes of antibiotics intensifies the need for fast phenotype-based clinical tests for determining antibiotic susceptibility. Standard susceptibility testing relies on the passive observation of bacterial growth inhibition in the presence of antibiotics. In this paper, we present a novel microfluidic platform for antibiotic susceptibility testing basedon stress-activation of biosynthetic pathways that are the primary targets of antibiotics. We chose Staphylococcus aureus as a model system due to its clinical importance, and we selected bacterial cell wall biosynthesis as the primary target of both stress and antibiotic. Enzymatic and mechanical stresses were used to damage the bacterial cell wall, and a β-lactam antibiotic interfered with the repair process, resulting in rapid cell death of strains that harbor no resistance mechanism. In contrast, resistant bacteria remained viable under the assay conditions. Bacteria, covalently-bound to the bottom of the microfluidic channel, were subjected to mechanical shear stress created by flowing culture media through the microfluidic channel and to enzymatic stress with sub-inhibitory concentrations of the bactericidal agent lysostaphin. Bacterial cell death was monitored via fluorescence using the Sytox Green dead cell stain, and rates of killing were measured for the bacterial samples in the presence and absence of oxacillin. Using model susceptible (Sanger 476) and resistant (MW2) S. aureus strains, a metric was established to separate susceptible and resistant staphylococci based on normalized fluorescence values after 60 minutes of exposure to stress and antibiotic. Because this groundbreaking approach is not based on standard methodology, it circumvents the need for minimum inhibitory concentration (MIC) measurements and long wait times. We demonstrate the successful development of a rapid microfluidic-based and stress-activated antibiotic susceptibility test by correctly designating the phenotypes of 16 additional clinically relevant S. aureus strains in a blinded study. In addition to future clinical utility, this method has great potential for studying the effects of various stresses on bacteria and their antibiotic susceptibility.

Introduction

Methicillin-resistant Staphylococcus aureus (MRSA) is a major health concern associated with over 18,000 deaths in the United States in 2005.¹ Not only has resistance to methicillin among S. aureus isolates become markedly more common,² but also numerous S. aureus strains with reduced susceptibility to vancomycin and linezolid have been reported.^3,4 Because antibiotics cannot always control S. aureus and MRSA isolates are becoming increasingly prevalent in the community, additional control strategies are sorely needed.

Traditionally, antibiotic susceptibility is determined either by broth dilution or disk diffusion techniques.⁵ Both of these tests rely on evaluation of bacterial growth in the presence of various concentrations of antibiotic. From the result of disk diffusion tests, the antibiotic susceptibility of a bacterial strain is characterized as resistant, intermediate, or susceptible. The broth dilution method allows for determination of the minimal concentration of the antibiotic that prevents bacterial growth (minimal inhibitory concentration, MIC). The results of these tests can be read in 18–24 hours but, in practice, it often takes longer,⁶ particularly in cases of slow-growing bacteria. Semi-automated systems based on the partial automation of the dilution method can give results in the range of 4 to 23 hours.^5,7–10

Genotypic methods, such as PCR, provide fast (within 2 hours after primary culture) identification of strains carrying specific antibiotic resistance genes,¹¹ such as the mecA gene for MRSA strains.¹² These methods are useful for screening, but may fail in cases where newly acquired antibiotic resistance is independent of target gene expression. An ideal test would be phenotypic so as to be robust to genetic variations and applicable to emerging diseases, while also delivering antibiotic susceptibility information as rapidly as possible to inform the patient’s treatment regimen.

Developing antibiotic susceptibility testing at the microfluidic scale carries the advantage of effective sample usage, since it requires small sample quantities. At the same time microfluidic devices are easily scalable to perform multiple tests on multiple samples.^13–15 A number of microfluidic methods have been developed to improve antibiotic susceptibility testing.^16–20 Bacteria are grown in the full volume of the microfluidic channel,^17,18 inside nano and pico-liter droplets^19–22 or as single bacterial cells electrically localized to the bottom surface of the channel.¹⁶ Microbial growth in the presence and absence of antibiotics is monitored via optical density, bright field/phase contrast or fluorescence measurements, or pH sensitive dyes.²³ These localizations allow susceptibility measurements in under 2 hours for samples taken from in vitro cultures (Escherichia coli)¹⁸ and in under 7.5 hours for samples taken from clinical specimens (E. coli and S. aureus).¹⁹ The common feature of these methods is that they are passive with respect to detection of antibiotic resistance. That is, the methods require the user to wait for bacterial growth as the final read-out.

In contrast, we propose to actively initiate biochemical pathways that are sensitive to antibiotics, and challenge the bacteria with those antibiotics, thereby creating a more rapid susceptibility measurement. If the bacteria are resistant to the antibiotic, they will be able to withstand the stressful conditions. However, if the bacteria are susceptible to the antibiotic, the combined stresses will lead to rapid cell death. The rate of cell death would then determine the phenotype of the bacteria (susceptible vs. resistant).

To develop this proposed method, we chose to initially target a well understood biochemical pathway: cell wall biosynthesis. Gram-positive bacteria, such as S. aureus, are surrounded by a network of glycan strands cross-linked by peptide bonds that form the basis of the bacteria cell wall, peptidoglycan.²⁴ The mechanism of action of β-lactam antibiotics such as penicillin and methicillin is inhibition of peptidoglycan biosynthesis.^24,25 Penicillin binding proteins (PBPs) are cell membrane bound enzymes, which catalyze the transpeptidation reaction that cross-links the peptidoglycan of the bacterial cell wall. The β-lactam antibiotics covalently bind to the PBP active-site serine and inactivate these enzymes, which are essential for bacterial growth.^26,27 Methicillin-resistance in staphylococci is most commonly provided by PBP2a encoded by the mecA gene.²⁸ In methicillin-resistant cells, PBP2a, with its low affinity for β-lactam antibiotics,^29–31 can substitute for the essential transpeptidation function of high-affinity PBPs at concentrations of antibiotic that are otherwise lethal.

We chose two direct methods of stressing the bacterial cell wall: a more general mechanical stress based on shear forces of moving liquid, and the Staphylococcus-specific enzymatic stress produced by lysostaphin. Shear forces can be applied to induce stress on bacteria and have been shown to cause responses in various bacterial model systems, i.e. Staphylococcus aureus,^32–37 Bacillus subtilis,^38,39 Microbacterium lacticum,⁴⁰ Bacillus thuringiensis,⁴¹ and E. coli.¹⁶ If the shear rates are high (greater than 2000 s⁻¹) or are exerted over an extended period of time (hours), shear forces will lead to cell death.^38–40 More moderate shear rates have been shown to cause down regulation of protein secretions⁴¹ or change bacterial adhesion to target cells and surfaces.¹⁶ We posit that applied mechanical stress induced by liquid flow in a microfluidic channel could trigger cell wall repair pathways.

A number of studies have been conducted on the effects of shear stress on S. aureus, focusing particularly on dynamic adherence mechanisms mediated by adhesins important for the initiation of infections.^32–37 Adhesion to host tissues is a critical initial step in the pathogenic process. S. aureus and coagulase-negative staphylococci are especially prominent pathogens in endovascular infections and cause endocarditis, thrombophlebitis, and vascular and heart valve prosthetic infections.⁴² Staphylococci must therefore have the capacity to tightly adhere to host tissue and resist shear stress imposed by the flow of fluids where shear rates can range between 40 and 2000 s⁻¹ for stable flow and even higher in turbulent flow regions and at vessel entrances and bifurcations.^16,42 Studies have demonstrated stable adhesion of S. aureus at shear rates as high as 1500–2000 s⁻¹ and have documented changes in adhesion mechanisms with increased shear stress.^{33,34,36,37,42} Thus, a body of literature has documented the ability of S. aureus to respond to mechanical stress by activating various biochemical pathways.

The enzymatic stress in our system is produced by lysostaphin’s ability to enzymatically cleave linkages in the bacterial cell wall,⁴³ thereby triggering cell wall repair pathways that are then inhibited by the antibiotic. The peptidoglycan cell wall of S. aureus consists of a backbone made up of alternating β-1,4-linked N-acetylglucosamine and N-acetyl muramic acid residues. Tetrapeptide chains consisting of L-alanine, D-glutamine, L-lysine, and D-alanine are linked to the muramic acid residues. These tetrapeptides are cross-linked by polyglycine cross bridges between the lysine residue of one chain and the D-alanyl residues of another chain. Lysostaphin is a zinc metalloenzyme which specifically cleaves the pentaglycine cross bridges found in staphylococcal peptidoglycan, weakening the cell wall.^44,45

We hypothesized that in the presence of mechanical stress (or in conjunction with enzymatic stress), an antibiotic would affect susceptible bacterial strains and lead to cell death, whereas resistant strains would be able to survive. To test our hypothesis, we built a multi-channel microfluidic device capable of interrogating multiple bacterial strains and multiple antibiotics at the same time. The microfluidic channels were created by placing a polydimethylsiloxane⁴⁶ (PDMS) layer on top of an epoxy-coated glass slide. In each channel, bacteria were covalently immobilized on the bottom of the microfluidic channel (Fig. 1). The shear stress was created by the movement of fluid (media and additives) through the microfluidic channel. The enzymatic stress was caused by adding small amounts of the bactericidal agent lysostaphin at concentrations significantly below its MIC.⁴⁷ Cell death was monitored by the uptake of the fluorescent dye Sytox Green.^48,49 We have applied our rapid microfluidic method to test the antibiotic susceptibility of 18 S. aureus strains, including predominant clones of clinically significant isolates, and we have shown that our method gives the same phenotypic designations as conventional methods in a fraction of the time. In the future, this phenotypic method could be expanded to other bacterial species and other antibiotics to enable ultra-rapid susceptibility testing of a wide array of infectious agents.

Fig. 1.

Cartoon showing the cross-sectional view of a single microfluidic channel. Bacterial cells (orange) are covalently bound to the bottom of the microfluidic channel. Under shear stress (τ) with or without enzymatic stress (green circles), susceptible cells die over time (lighter shade) in the presence of the antibiotic (red pentagons). The fluorescent dye (yellow diamonds) can only permeate the cell membrane when the bacteria die, producing fluorescence when it binds to DNA.

Materials and methods

In this section, we present details on the design and manufacturing of the microfluidic flow cell, and its integration with the microscopy-based image collection and liquid flow control. We also report on bacterial growth conditions, experimental protocols, image analyses, and data processing.

Microfluidic flow cell design and manufacturing

Channel geometry and shear stress estimation

Each device had four identical microfluidic channels (Fig. 2). The rectangular cross-sectional geometry of the microfluidic channel was chosen to simplify manufacturing and shear stress calculations. The cell culture media is similar in its liquid properties to water, which is a Newtonian incompressible fluid. Given the physical dimensions of the channel (Fig. 2), the flow inside the channel is laminar (Reynolds number of 11).⁵⁰ In this case, a simple analytical expression illustrates the relationship between fluid viscosity μ, volumetric flow rate Q, channel width w, channel height h and shear stress, τ, when the width of the channel is much greater than its height:⁵⁰

Fig. 2.

Schematic drawing of the microfluidic channel geometry (not to scale).

$$\tau =-\mu \frac{6Q}{{wh}^2}$$ (1)

According to Eq.1, shear stress will be directly proportional to the volumetric flow rate and inversely proportional to the width and square of the height of the channel. Because the height and width of our channel are comparable in size, this equation represents an upper bound of attainable shear stress. The maximum shear stress achievable for the given channel cross-section equaled 6.25 Pa for a bulk flow rate of 1 mL/min in water at 20°C, corresponding to shear rate value of 6000 s⁻¹. This rate equals or exceeds previously studied shear stresses on staphylococci.^{33,34,36,37,42}

PDMS layer fabrication

The inverse of the channel shape was cut in a stainless steel metal base using an ultra-precision milling machine (UPM-0005, Fraunhofer-IPT, Aachen, Germany). The top surface of the channel mold was finished to below 20 nm roughness using a fly cutting technique.⁵¹ Such a finish was necessary since larger features imprinted into the PDMS layer caused significant light scattering and hindered microscopic investigation of the samples.

A standard protocol was used for making channels in the PDMS.^46,52 Briefly, a 10:1 degassed mixture of PDMS and a curing agent (Sylgard^® 184 silicone elastomer kit, Dow Corning Corp., Midland, MI) were poured over the mold. The amount of PDMS was chosen such that channel openings defined by mold pins would stay open. The whole assembly was placed into the oven and cured at 37°C overnight. The cured PDMS layer was dissected along the edge of the mold and carefully pulled off the mold surface with a pair of forceps. The mold surface was cleaned with 70% ethanol and a Q-tip and re-used for multiple PDMS replicates. Each PDMS layer was used for several experiments until deterioration of the channel surface occurred.

Flow cell assembly and testing

The PDMS layer was placed on top of a SuperEpoxy glass slide (Fig. 3). The glass slide (SuperEpoxy2, Arrayit Corporation, Sunnyvale, CA) was pre-coated by the manufacturer with epoxy groups for non-specific protein binding. Amino, hydroxyl and thiol groups on bacterial surface proteins react with the epoxy groups and covalently bind the bacteria to the glass surface.

Fig. 3.

Microfluidic flow cell assembly (exploded view). The top metal plate provides connectivity with the outside fluidics. The top glass window provides an even sealing pressure on top of the PDMS-layer, while the bottom sealing plate provides sealing pressure to the bottom of the coated glass slide.

The standard sealing procedure of oxygen plasma treatment⁴⁶ to bond PDMS to glass could not be used, as it would destroy the coating on the glass surface. Instead, the PDMS-glass slide combination was inserted into a metal frame with a precisely engineered pocket and clamped closed to ensure even pressure and a seal against the glass surface. The metal frame provided an interface to the macrofluidic part of the setup via tapped through ports matched to the PDMS layer channel openings. The assembled flow cell was connected to a KDS-230 syringe pump (KD Scientific Inc., Holliston, MA) on the input side and to waste collection on the output side. Various thicknesses of the PDMS were fabricated using 0.02 g increments of pre-vacuumed PDMS-curing agent mixture weight to find a thickness that ensured leak-free functioning of the device at the target shear stress values. Leak-free functionality was achieved for bulk flow rates of up to 1 mL/min with a PDMS pre-cured weight of 4.47 g resulting in a PDMS layer 2.47 mm thick.

Microscopy and system integration

The assembled flow cell was placed on a computer driven XY stage (H107AENN, Prior Scientific Inc., Rockland, MA) mounted onto an IX-70 epi-fluorescence inverted microscope (Olympus Inc., USA). The XYZ position of the sample was set manually or through computer commands via the RS232 port of the controller (H3-XYZ2, Prior Scientific Inc., Rockland, MA). The long working distance 60× microscope objective was used for phase contrast and fluorescence imaging (LCPlanFl, Olympus Inc., USA).

Both the excitation and emission light passed through a filter cube (U-MNIB, 470–490 nm excitation, >510 nm emission, Olympus Inc., USA). Ninety percent of the field of view (250 × 250 μm) was collected onto a monochrome 8-bit CCD camera (Retiga-4000R, QImaging Surrey, BC, Canada), which was connected to a PC USB port. Image acquisition was controlled by in-house custom software.

Other than the initial manual focusing, the data collection process was automated. At the outset, the user defined the number of cycles and the time between individual cycles. Each data collection cycle consisted of the acquisition of one phase contrast and one fluorescence image for each of the four channels. The microscope objective was automatically relocated to the initial position at the beginning of each new cycle.

Fluid flow rate and volume were automatically controlled with the custom software. Luer-lock syringes (McMaster-Carr, Robbinsville, NJ) were connected to the channel inputs by 0.03 inch inner diameter fluorinated ethylene-propylene (FEP) tubing (IDEX Health & Science LLC, Oak Harbor, WA). The output waste was collected into four separate containers. The collected volumes were measured at the end of the experiment to detect any leakage or cross-contamination.

Bacterial strains and cultivation conditions

This study included clinically significant strains of S. aureus that had acquired resistance to methicillin or were methicillin-susceptible (refer to Table 1 for strain list, MICs and references). S. aureus strains ST22 and ST80 were provided by Dr. Frank Hanses, Dr. Binh Diep provided S. aureus SF8300, and Dr. Kristina Hultén provided S. aureus TCH959. Other MRSA strains were obtained through the Network on Antimicrobial Resistance in S. aureus (NARSA) program supported under NIAID, National Institutes of Health Contract HHSN272200700055C. Frozen stock cultures of all strains were maintained at −80°C. Prior to the microfluidic experiments, the frozen cultures were streaked onto tryptic soy agar plates, incubated overnight at 37°C, and then stored at 4°C. Staphylococci were cultivated overnight in Mueller-Hinton broth with 2% sodium chloride and seeded 1:1000 into fresh medium the day of the experiment. The bacteria were cultivated at 37°C for 3 h on an orbital shaker at 250 rpm before harvesting the cells by centrifugation at 1625 × g for 2 min at 20°C. According to an experimentally established growth curve, harvested bacteria were in the logarithmic phase of growth at concentrations of 10⁷–10⁸ cfu/mL. The staphylococci were suspended in fresh medium to a concentration of 1–5 ×10⁸ cfu/mL for susceptibility testing, as determined by quantitative plate counts.

Table 1.

Comparison of the microfluidic test results with clinically determined phenotypes. Lines separate prototype strains from the blinded study strains, and MSSA from MRSA within the blinded study.

Strain name	Microfluidic Test	Clinical Designation	Oxacillin MIC (μg/mL)	Reference
Sanger 476a	S	MSSA	0.5b	53,57
MW2a	R	MRSA	>16b	58–60
MN8	S	MSSA	0.12b	61
NCTC8325	S	MSSA	0.25b	62
Newman	S	MSSA	0.06	63,64
PS80	S	MSSA	0.5b	65
TCH959 - USA300	S	MSSA	1	66
COL	R	MRSA	>16b	67
LAC - USA300	R	MRSA	32	68
NRS382 – USA100	R	MRSA	>16b	59
NRS482 - USA300	R	MRSA	4b	69
NRS699 - USA100	R	MRSA	>16b	70
Sanger 252	R	MRSA	>16b	53,71
SF8300 - USA300	R	MRSA	--c	72
ST22 #30	R	MRSA	--c	73
ST80 #13	R	MRSA	--c	73
ST80 #16	R	MRSA	--c	73
ST80 #17	R	MRSA	--c	73

^a-Prototype strains,

^b-Data from NARSA,

^c-Determined by disk diffusion.

S. aureus USA400 strains Sanger 476 (MSSA) and MW2 (MRSA) were used to develop the experimental protocols, to set experimental parameters, and to develop the image-processing algorithm. These isolates are related (both multilocus sequence type 1), but they differ in their susceptibility to oxacillin.⁵³ The other 16 strains were tested for susceptibility in a blinded fashion such that the identity of the isolates were unknown by the investigator at the time of microfluidic testing.

Experimental protocol

Approximately 150 μL of each bacterial suspension was introduced into each microfluidic channel. The flow cell was incubated at 37°C for 45 min to allow for bacterial settling and subsequent attachment to the coated glass slide surface. This timing was determined experimentally with the prototype strains by varying settling times and initial bacterial concentrations to maximize the number of attached bacteria. Bacterial oxygenation can be an important factor in microculturing.¹⁸ During the bacterial attachment phase, fresh oxygen can only enter through the inlet ports (Fig. 2) and passively diffuse along the channel, since the rest of the channel is covered by the glass window and body of the flow cell (Fig. 3). However, in the experiment, the oxygen is delivered continuously with the flow of the fresh media. To study the effect of oxygenation change during the attachment phase, we incubated an epoxy slide with only the PDMS on top of the glass slide for the duration of the attachment phase. After incubation, the PDMS-epoxy slide combination was inserted back into the flow cell for the experiments. However, these attempts did not improve our test results, and so bacterial attachment was performed with a fully assembled device as described above.

In each experiment, two S. aureus strains were tested in parallel. Pairs of channels were inoculated with the same strain and tested with or without antibiotic. The fluid used in the control channel comprised Mueller-Hinton + 2% NaCl broth, enzymatic stress (lysostaphin), and a fluorescent viability stain. The experimental channel used the same media but with antibiotic added. The enzymatic stress agent, lysostaphin (Cell Sciences, Canton, MA), was added to all channels at the concentration of 0.7 ng/mL. This lysostaphin concentration was almost 100-fold lower than the S. aureus MIC value of 60 ng/mL, determined for the prototype strains using standard microdilution broth methods. The antibiotic, oxacillin (Sigma-Aldrich Inc., Saint Louis, MO), was tested at 10, 50, and 250 μg/mL. For comparison, the MIC values for Sanger 476 and MW2 are 0.5 and 16 μg/mL, respectively. The fluorescent stain, Sytox Green (Molecular Probes, Invitrogen Corporation, Carlsbad, CA), was added to all media at 0.5 μM as per the manufacturer’s instructions. The selection of the dye was based on its ability to selectively stain only damaged cells, its non-toxicity to live cells, and its low background fluorescence, which allowed for its direct addition to the cell media without additional steps.^48,49

For each experiment, four syringes, loaded with 60 mL of the appropriate solutions, were attached to a syringe pump. The media were pumped through the flow cell at a rate ranging from 0.3 to 1 mL/min for one hour. The lower boundary was set because the inherent fluid resistance inside the flow cell and attached tubing needed to be overcome, otherwise bubbles developed during testing. On the other hand, flow rates higher than 1 mL/min caused leakage between the PDMS-layer and the glass slide. Since temperature can also be an important factor affecting bacterial metabolic activity, the Sanger 476 prototype strain was tested at a higher temperature. For this experiment, two filled syringes were kept at room temperature, while the medium in two additional syringes was preheated to 37°C. Because we saw no significant changes in our experimental outcomes, subsequent experiments were performed with media at room temperature (~22°C).

At the beginning of the experiment, any non-attached bacteria were washed away by flowing media through the channel, leaving only a single layer of attached bacteria on the bottom of the microfluidic channel. The number of bacteria per microscope field of view ranged from 1 × 10³ to 2 × 10⁴ depending on bacterial strain and experimental conditions. Because the only bacteria remaining were covalently attached to the glass slide, the microscope was focused on the channel bottom for the remainder of the experiment. Phase contrast and fluorescence images were acquired with 10 ms and 800 ms exposure times, respectively. The first image cycle occurred 1 min after initiation of the flow and was repeated every 2 min thereafter.

Additionally, we performed no-flow control experiments in which bacteria were allowed to settle onto epoxy-coated glass slides placed inside four-well rectangular NUNC^™ culture dishes (Electron Microscopy Sciences, Fort Washington, PA). Each well held one slide mimicking a single channel of the microfluidic flow cell. Following a 45-minute settling period, the slides were gently washed to remove the non-adherent bacteria and media were added to the different wells (control media contained lysostaphin and Sytox Green while experimental media contained lysostaphin, Sytox Green, and oxacillin). The cell death was monitored microscopically over one hour interval, as in the flow experiments.

Imaging data analysis and cell death percentage calculation

The percentage of dead bacteria due to the presence of antibiotic was used to determine the susceptibility of a specific bacterial strain to oxacillin. Phase contrast and fluorescent images were analyzed together to determine the percentage of dead cells in a specific channel (Fig. 4). Bacterial cell counting was performed with routines written in the Matlab-based open source image analysis software CellProfiler⁵⁴ (The Broad Institute, Cambridge, MA). The fluorescent cell counts (N_f) were normalized to the total cell counts in phase contrast (N_p) for comparison between different time points (since a small number of cells are washed away over the course of an experiment). To determine the percentage of cell death due to antibiotic only, it was necessary to normalize the data to both (1) the cell death present at the outset of the experiment (t=1min) and (2) the cell death caused by mechanical and enzymatic stress (control channel). The normalized bacterial cell death percentage (DP_norm) at different time points was determined using Eq.2:

$${DP}_{\mathit{norm}}=[{\left(\frac{N_{f,t=T}}{N_{p,t=T}}-\frac{N_{f,t=1\mathit{min}}}{N_{p,t=1\mathit{min}}}\right)}_{\mathit{antibiotic}}-{\left(\frac{N_{f,t=T}}{N_{p,t=T}}-\frac{N_{f,t=1\mathit{min}}}{N_{p,t=1\mathit{min}}}\right)}_{\mathit{control}}]\times 100$$ (2)

Fig. 4.

Analyzed phase contrast and fluorescent fields at 60× magnification at T=59 min for channels with or without (control) antibiotic. Automated bacterial counts in phase contrast (red dots) and fluorescence (white speckles) images are shown in the top left corner of each panel.

The first term in this equation represents bacterial death in the channel containing both stress and antibiotic. The second term represents the bacterial cell death in the control channel due to the same stress conditions but in the absence of the antibiotic. The difference between the two terms represents bacterial cell death in response to antibiotic. The results of our experiments were plotted as the percentage of cell death due to antibiotic over time (DP_norm; see Fig. 5). This representation highlights the difference in the magnitude as well as in the dynamics of susceptible and resistant strain responses to antibiotic treatment.

Fig. 5.

Normalized cell death percentage in the presence of 50 μg/mL oxacillin as a function of time for Sanger 476 (oxacillin MIC: 0.5 μg/mL, MSSA, blue) and MW2 (oxacillin MIC: 16 μg/mL, MRSA, red). Data are shown for mechanical stress only (Mech, circles) and for combined mechanical and enzymatic stress (Mech/Enz, triangles). The DP_{norm_max} for each experiment is highlighted with a black square.

Results

Sanger 476 and MW2 prototype strain testing

A time course of normalized cell death (DP_norm) was calculated from measurements taken for both prototype strains according to Eq. 2. The susceptible strain Sanger 476 showed higher cell death percentages than the resistant strain, and the differences were significant after ~25 min (Fig. 5). Sanger 476 also showed a steady increase in fluorescence over the time course of the experiment, which corresponded to an increase in cell death due to its susceptibility to the antibiotic. In contrast, the resistant strain MW2 showed a relatively low fluorescence signal overall and a minimal increase in fluorescence over time.

The shear flow of antibiotic-containing media was sufficient to induce cell death in the susceptible strain that was greater than the resistant strain (Fig. 5, blue vs. red circles). The addition of enzymatic stress increased the rate of cell death for the susceptible strain (Fig. 5, blue vs. red triangles), without changing the rate of killing for the resistant strain. Experiments run in the absence of shear flow failed to give DP_norm values greater than 0.5% for either S. aureus Sanger 476 or MW2, indicating that applied stress was critical for rapid distinction between oxacillin susceptibility and resistance.

Due to the steady increase in DP_norm values for susceptible strains over time, the greatest difference between the susceptible and resistant strains was achieved at the maximum experimental time (60 min). The cell death percentage value at the maximum time (DP_{norm_max}) was used to establish criteria for the blinded strain susceptibility testing (see black squares in Fig. 5).

Establishing susceptibility criteria

We compared oxacillin concentrations of 10 and 50 μg/mL for distinguishing oxacillin susceptibility of our prototype S. aureus strains. DP_{norm_max} values for Sanger 476 and MW2 were measured repeatedly at both antibiotic concentrations under standard conditions (flow rate of 1 mL/min, lysostaphin concentration of 0.7 ng/mL, and experimental solutions at 22°C). The data are presented on a log-scale to encompass the full dynamic range of DP_{norm_max} values (Fig. 6). The mean of the Sanger 476 data points is similar for the two oxacillin concentrations tested: 4.29±3.33% (50 μg/mL) and 4.35±5.02% (10 μg/mL), but the higher concentration of antibiotic (50 μg/mL) gave clearer results for the susceptible strain without significantly changing the results obtained with the resistant strain. When we tested even higher concentrations of oxacillin (250 μg/mL), the results were similar to those obtained with 50 μg/mL (not shown).

Fig. 6.

Distribution of DP_{norm_max} values for Sanger 476 (open, blue) and MW2 (solid, red) in the presence of 0.7 ng/mL lysostaphin and either 10 μg/mL (circles) or 50 μg/mL (triangles) oxacillin. Each data point is the result of a single experiment (i.e. 34 replicates of the Sanger 476 at 10 μg/mL) and is slightly offset on the x-axis for clarity. Note: data are plotted on a log-scale. Arrows and letters indicate susceptibility zones with 0.5% and 1% line borders; S=susceptible; R=resistant. Zone of indeterminate susceptibility is highlighted in gray.

Comparing DP_{norm_max} distributions of the MW2 and Sanger 476 strains, the following empirical criteria were established to separate susceptible and resistant strains: DP_{norm_max} <0.5% (strain is resistant, R), 0.5% < DP_{norm_max} < 1% (indeterminate strain susceptibility), DP_{norm_max} >1% (susceptible, S) (Fig. 6). According to these criteria, the higher concentration of oxacillin (50 μg/mL) gave a clearer separation between the susceptible and resistant strains than the 10 μg/mL concentration. The susceptibility determination of Sanger 476 gave 11 correct identifications out of 11 measurements when 50 μg/mL oxacillin was used as compared to 29 out of 34 at the lower concentration of antibiotic. The resistance phenotype of MW2 was correctly determined in 9 out of 11 tests at 50 μg/mL oxacillin versus 5 out of 8 tests at 10 μg/mL.

To assess the effect of the bacterial growth phase on our experimental system, we harvested strains Sanger 476 and MW2 from overnight cultures (stationary phase of growth), and the staphylococci were attached to epoxy slides at a concentration of >5×10³ bacteria per microscope observation area. To achieve sufficiently high numbers of attached bacteria, the flow cell incubation time was increased to one hour. There was no cell death registered for either the susceptible or resistant prototype strain at 10 or 50 μg/mL of oxacillin during a 1-h experiment, nor was there cell death recorded in the antibiotic-free control channel (data not shown). Because beta-lactam antibiotics interfere with peptidoglycan biosynthesis, which occurs primarily in rapidly dividing bacterial cells,⁵⁵ this result was anticipated for a 1-hour assay. Based on these results, all experiments for the blinded study were conducted with bacteria in the logarithmic growth phase.

Blinded susceptibility study of clinically relevant S. aureus strains

For the blinded strain susceptibility testing, the target accuracy value was set at 95%. The 81% accuracy (9 out of 11) of MW2 susceptibility determination was assumed as a base error rate. To improve the expected value to the target accuracy, we required repetition of strain measurements until three results with the same outcome (resistant or susceptible) were achieved. From these results, the strain was scored as either resistant or susceptible. Using these criteria, the estimated probability of a correct susceptibility assignment was >96% using a negative binomial probability distribution.⁵⁶

Our established susceptibility criteria were tested in a blinded study of 16 representative S. aureus strains that included well-characterized laboratory strains (NCTC8325, Newman, and PS80), as well as various hospital-associated clinical isolates (USA100 [US] and Sanger 252 [Europe]) and community-acquired strains (USA300 [US] and ST22 and ST80 [Europe]). The experimental parameters were fixed at the values pre-determined for the prototype strains. All strains were subjected to 50 μg/mL concentrations of oxacillin (with antibiotic-free controls). Each strain was tested in at least three different experiments to guarantee 95% accuracy of the outcome as determined by the prototype strain testing above. The diagnostic results of the microfluidic testing are summarized in Table 1 and are compared to the clinical designation of the strains.

All of the S. aureus strains in the blinded study were classified correctly. Each of the individual susceptibility tests repeated for the same strain had the same outcome: either susceptible or resistant. Hence, no more than three measurements were needed per strain in the blinded study. As shown in Fig. 7, a clear separation exists between DP_{norm_max} values for MSSA and MRSA strains. Susceptibilities of all of the tested strains were determined with 100% sensitivity and 100% specificity. The specificity and sensitivity values were calculated based on the number of strains tested, and the outcome of the microfluidic test as compared to the previously reported clinically determined susceptibility profile.

Fig. 7.

Average DP_{norm_max} values measured during the blinded study for various strains in the presence of mechanical and enzymatic stress at 50 μg/mL oxacillin. Susceptible strains (blue triangles) and resistant strains (red circles) are separated into groups. Each strain was measured three times, with the exception of Newman (N=6). The error bars show a single standard deviation. Note: data are plotted on a log-scale.

In our study, all of DP_{norm_max} values of the susceptible strains were above the pre-assigned susceptibility threshold (DP_{norm_max} > 1%), and all those of the resistant strains were below the pre-assigned resistance threshold (DP_{norm_max} < 0.5%), allowing a robust designation of strain susceptibility. While the standard deviation of the DP_{norm_max} values for the resistant strains did not exceed 0.15%, the susceptible strain DP_{norm_max} values varied markedly when exposed to 50 μg/ml oxacillin. For example, the mean and standard deviations of DP_{norm_max} values for multiple assays with Sanger 476 was 4.29±3.33% and for the panel of susceptible strains was 4.45±2.45%. The similarity in these averages suggests that MSSA respond similarly to 50 μg/ml oxacillin, and the sizable deviation may be caused by variability in unmonitored experimental conditions. Sampling of five different microscope fields within the same microfluidic channel for Sanger 476 gave DP_{norm_max} values of 3.25±0.64%, whereas sampling across replicate channels (three channels measured under the same parameters and repeated in two different experiments) led to higher variability (4.45±2.48%). The variability is likely due to a combination of factors including variability from channel to channel (variation in applied shear stress), which might be overcome in the future by replacing the compliant PDMS layer with a less compliant hard plastic device.

Discussion & Conclusion

Our results clearly demonstrate the ability of our rapid microfluidic method to distinguish MRSA from MSSA strains. Time-dependent cell death monitoring gave a quantitative criterion, allowing an objective susceptibility designation to be made without the need for MIC assays. In contrast to passive growth-based methods, the application of mechanical and enzymatic stresses in our method pro-actively accelerates cell killing. Because of this, our microfluidic platform may facilitate testing of antibiotic resistance in slow growing microbes, possibly fulfilling an unmet clinical need.⁷⁴

Because the reported method measures the number of dying bacteria over time, it is well suited for testing bactericidal antibiotics that actively promote death. However, many bacteriostatic antibiotics become bactericidal at concentrations above their MIC,⁷⁵ and so it may be possible to find appropriate conditions for testing resistance to these antibiotics as well.

In the present study, we investigated some of the parameters that may be responsible for a large spread of cell death values for the susceptible strains. Bacteria exhibit population heterogeneity under standard culture conditions and in response to environmental changes.^76–79 The outcome of the microfluidic testing was not influenced by changes in two major factors affecting bacterial growth in bulk culture: temperature and oxygenation. Both temperature and oxygenation differ between the attachment and flow phases of the experiment in our standard protocol. To increase the metabolic activity of the bacteria, we varied both at the point in the experiment where the greatest effect could be achieved. As such, we increased temperature during the flow phase and increased oxygenation during the attachment phase. However, changes to these parameters did not significantly affect our experimental outcomes. Hence, it appears that our method is not sensitive to changes in temperature and oxygenation within the tested range, thus allowing flexibility in the protocol conditions.

Further clinical applicability of the reported method will depend on its validation with multiple bacterial species. Our applied enzymatic stress (lysostaphin) is specific to staphylococci, whereas the applied mechanical stress is much more general. The study of bacterial responses to shear stress in the context of our method is currently uncharted ground. The most straightforward way to determine the effect of mechanical stress would be to increase its dynamic range via changes to the microfluidic channel design. Future work will include making microfluidic channels with a smaller cross-section (Eq. 1), while optimizing channel length and height outside of the observation area to minimize liquid flow resistance.

Our platform is well suited for further study of the effects of various stresses (i.e., osmotic, ionic, pH) on the speed and accuracy of antibiotic susceptibility testing. Moreover, recent studies indicate that antibiotics with different primary targets in bacterial cells activate the same downstream bacterial response pathways.^80,81 Therefore, our method may be applicable for the rapid identification of susceptibility to antibiotics that target biosynthetic pathways unrelated to cell wall biosynthesis. As such, we have the potential to develop a fast clinical method applicable to testing a broad spectrum of bacteria and antibiotics. Additionally, we believe that our antibiotic testing method may lead to the discovery of new biology within these systems.

In summary, we have described the development of a microfluidic platform and demonstrated a new stress-based approach for phenotypic antibiotic susceptibility testing. A key step in the method’s implementation was immobilization of bacteria on the bottom surface of the microfluidic channel. As such, bacteria can be subjected to a variety of stresses and simultaneously imaged under a microscope. We have chosen mechanical and enzymatic stresses that directly affect the bacterial cell wall and have paired them with an antibiotic that inhibits cell wall biosynthesis. Using a pair of prototype MSSA and MRSA strains, a unified protocol was developed for antibiotic susceptibility testing. Using this methodology, the oxacillin susceptibilities of 16 additional S. aureus isolates were correctly designated in a blinded study. Diagnostic results can be obtained within 60 minutes after introduction of the antibiotic to attached log-phase bacteria.