Label-free single-cell antimicrobial susceptibility testing in droplets with concentration gradient generation

Abstract

Bacterial communities exhibit significant heterogeneity, resulting in the emergence of specialized phenotypes that can withstand antibiotic exposure. Unfortunately, the existence of subpopulations resistant to antibiotics often goes unnoticed during treatment initiation. Thus, it is crucial to consider the concept of single-cell antibiotic susceptibility testing (AST) to tackle bacterial infections. Nevertheless, its practical application in clinical settings is hindered by its inability to conduct AST efficiently across a wide range of antibiotics and concentrations. This study introduces a droplet-based microfluidic platform designed for rapid single-cell AST by creating an antibiotic concentration gradient. The advantage of a microfluidic platform is achieved by executing bacteria and antibiotic mixing, cell encapsulation, incubation, and enumeration of bacteria in a seamless workflow, facilitating susceptibility testing of each antibiotic. Firstly, we demonstrate the rapid determination of minimum inhibitory concentration (MIC) of several antibiotics with Gram-negative E. coli and Gram-positive S. aureus, which enables us to bypass the time-consuming bacteria cultivation, speeding up the AST in 3 h from 1 to 2 days of conventional methods. Additionally, we assess 10 clinical isolates including methicillin-resistant Staphylococcus aureus (MRSA) and multidrug-resistant Staphylococcus aureus (MDRSA) against clinically important antibiotics for analyzing the MIC, compared to the gold standard AST method from the United States Clinical and Laboratory Standards Institute (CLSI), which becomes available only after 48 h. Furthermore, by monitoring single cells within individual droplets, we have found a spectrum of resistance levels among genetically identical cells, revealing phenotypic heterogeneity within isogenic populations. This discovery not only advances clinical diagnostics and treatment strategies but also significantly contributes to the field of antibiotic stewardship, underlining the importance of our approach in addressing bacterial resistance.

---

Introduction

The excessive usage of antibiotics has led to the global emergence of antibiotic resistance, posing a substantial peril to human well-being.^1–3 In particular, multidrug-resistant (MDR) bacteria are pervasive in numerous healthcare environments, resulting in severe, difficult-to-treat infections and elevated mortality risks.^4,5 Rapidly identifying and measuring antibiotic resistance, coupled with effective antimicrobial stewardship, emerge as pivotal responsibilities in clinical microbiology considering the life-threatening conditions.⁶ These actions are crucial for refining the management of infectious diseases and curbing the usage of broad-spectrum medications. While significant progress has been made in rapid pathogen identification through molecular diagnostics,^7–9 the development of corresponding rapid AST technologies is still a work in progress.

The current gold-standard antibiotic susceptibility tests (ASTs) encompass both liquid medium-based broth microdilution (BMD) assay and solid medium-based techniques like the disk diffusion assay or E-test.^10,11 These assays evaluate the collective behavior of a diverse and heterogeneous cell population in the presence of antibiotics and provide a crucial parameter known as minimum inhibitory concentration (MIC), the lowest concentration of the antimicrobial drug that prevents the growth of the pathogen, to ascertain whether the bacterial isolates are susceptible or resistant to antibiotics.¹² If the determined MIC is equal to or less than a specified breakpoint, the bacterial isolate is considered susceptible to the antibiotic. The Clinical and Laboratory Standards Institute (CLSI) in the United States updates clinical breakpoints for various combinations of bacteria and antibiotics annually.¹³ The MIC value holds significant importance as a parameter for analyzing phenotypic resistance in bacteria. It not only helps in assessing the effectiveness of new antibiotics but also aids in monitoring the global status of drug resistance.

Unfortunately, they necessitate prolonged measurement time ranging from 2 days to 4 days. Specifically, the procedure begins with collecting patient samples, from which bacteria are isolated by streaking them on selective culture media and then allowing them to grow overnight. The pure culture of isolated colonies is then applied to conduct ASTs through disk diffusion or broth microdilution assays. Thus, the whole procedure of conventional AST technologies requires about 48 hours after sample collection. This prolonged timeline comes from measuring the absorbance or visually assessing the turbidity resulting from multiple rounds of bacterial division. In addition, this delay leads to a minimum one-day setback in commencing targeted therapy using narrow-spectrum agents. This delay contributes to elevated patient mortality rates and unfavorable clinical outcomes.¹⁴

In this golden time, to prevent worsening the patient's condition, the clinician often prescribes an empirical antibiotic with a broad spectrum of activity in large doses to ensure its efficacy on the target pathogen. Consequently, this temporal gap could potentially foster the emergence or perpetuation of strains exhibiting broader spectra of antimicrobial resistance (AMR). This scenario underscores the critical but unmet need for developing rapid AST technologies to identify antibiotic susceptibility from the initial stages of treatment.

Moreover, there is a debated concern that the turbidity measurement in a bacterial culture might yield false positive results due to alterations in bacterial morphology when exposed to antibiotics,^15–17 as the measured optical density depends not only on the number of cells but also the size and morphology. However, these systems perceive the increase of absorbance as the increase in cell density, assuming the bacterial growth and absorbance show a linear relationship. Also, false negative results take place upon the existence of heterogeneity in the cell population.¹⁸ These phenotypic features can lead to misclassification of antibiotic efficacy, resulting in the use of ineffective drugs or the exclusion of potentially useful ones. Thus, recent development of AST methods focuses on monitoring the proliferation of smaller subsets of bacteria and reducing detection times, which enables clinicians to promptly make more informed treatment decisions.

Several microfluidic-based AST approaches have recently emerged, offering distinct advantages such as rapid analysis, high throughput, and cost-effectiveness.^19–24 Among these methods, droplet microfluidics has been employed to encapsulate bacteria within droplets.^25–27 The capability to isolate a single bacterial cell in a number of droplets allows the identification of rare bacteria within the population and rapid diagnosis of hetero-resistance. Such precise confinement enables the detection of even the slightest changes in bacterial metabolism, genetic composition, morphology, or replication in response to antibiotic exposure. They have demonstrated the capability to detect MIC values within 12 hours, depending on the doubling time of the bacteria and the analytical assay employed. This represents a remarkable advancement in response time when compared to the 48 hours typically required by conventional phenotypic AST methods.

In addition, several studies have demonstrated the feasibility of accelerating AST procedures using a single-cell optical imaging setup to determine antibiotic susceptibility within 4 hours.²⁸ This system detects bacteria by imaging changes in their morphology after exposure to antibiotics, such as cell division, swelling, and the formation of filaments. The optical method is often combined with bacterial immobilization techniques, including agarose gel immobilization, dielectrophoresis-assisted trapping, confined microchannels, and nanoliter wells.^29–31 Notably, parallel microchannels can efficiently trap and image single bacteria, determining susceptibility in as little as 30 minutes.³¹ However, these methods demand a high-resolution optical setup and the continuous monitoring of multiple locations, rendering them complex, expensive, and challenging for clinical applications.

Other endeavors aim at expediting AST without the necessity of individually imaging cells while maintaining sensitivity by incorporating detection methods, such as detecting subtle changes in mass through the cantilever,³² fluorescence detection,³³ asynchronous magnetic bead rotation,³⁴ electrical resistance,³⁵ electrochemical phenotyping with redox reporter,³⁶ or impedance measurements.^37–39 In particular, microfluidic impedance cytometry, one of the pioneering techniques, can determine the result of AST within an hour using an actively dividing culture sample. Nonetheless, these approaches often involve complex readouts, rely on specific electrochemical or fluorescent labels and intricate detection systems, and have not yet been in a multiplexing format for different concentrations of antibiotics. Lastly, the integration of antibiotic gradients presents a significant advancement in multiplexing by simultaneously testing a range of antibiotic concentrations.^40–42 However, these techniques require a continuous flow of antibiotics to maintain the gradient profile, which limits the number of simultaneous tests achievable.

Previous microfluidic methods based on fluorescence analysis represent a substantial improvement upon the state of the art and simply detect bacterial growth by using fluorescent metabolism markers, such as resazurin.³³ However, these methods face limitations when analyzing a small subpopulation of bacterial cells in the presence of antibiotics. Metabolic activities before cell death can produce a positive fluorescent signal indistinguishable from that of drug-resistant cells,⁴³ and resazurin is not compatible with anaerobic bacteria because resazurin is a marker of the respiratory activity of a cell. Also, there is a problem of leakage of resazurin from droplets to surrounding oil, which could potentially reduce the accuracy of the assay. Consequently, resazurin may not be the optimal choice as a detection marker for assessing bacterial growth in clinical samples. Therefore, it is imperative to develop label-free techniques to overcome these challenges and provide more accurate and adaptable solutions for AST in microfluidics. Such methods would not only circumvent the limitations posed by fluorescent markers like resazurin but also offer more accurate and adaptable solutions for antimicrobial susceptibility testing (AST) in microfluidic platforms. By focusing on label-free approaches, researchers can enhance the reliability of AST methods, enabling the detection of antibiotic resistance with greater precision and facilitating the use of these technologies in clinical settings.

Herein, we present a droplet microfluidics AST platform that assembles the design for the formation of the concentration gradient of antibiotics and subsequent encapsulation of bacterial cells in a high-throughput manner, each concentration composed of >10 000 droplets, to enable rapid single-cell AST without the use of any fluorescent molecules or genetically modified bacteria expressing fluorescent proteins. Such simplicity of the droplet microfluidic platform is achieved by executing bacteria and antibiotic mixing, cell encapsulation, incubation, and enumeration of a small number of bacterial cells under different antibiotic conditions, facilitating the investigation of the susceptibility of each antibiotic. The droplet AST provides rapid, same-day MIC results of clinically important antibiotics, and the efficacy of our assay is subsequently confirmed by testing clinical isolates and directly comparing our results to those obtained in the clinic using conventional broth micro-dilution (BMD) assay.

This approach shows the multiplexing potential of the AST platform by generation of a concentration gradient of antibiotics, parallelization of each droplet pool, and simple enumeration of cells through automatic image analysis. We develop an algorithm for automatable analysis of results and rapid determination of susceptible/resistant strains. Moreover, a droplet microfluidics AST platform provides the result of the heterogeneity of antibiotic susceptibility at the single-cell level, showing distinct growth patterns within droplets where bacterial cells are exposed to identical antibiotic concentrations.

Hence, our platform tackles critical issues that earlier studies struggled with, such as yielding quantitative AST outcomes restricted by low throughput in multiple concentration tests within a single experiment and an accelerated and precise screening process that has the potential to enhance the speed of obtaining results in clinical settings, circumventing the need for traditional culture techniques. As a result, this could lead to a further study removing the preincubation of the sample by performing AST directly on bacteria harvested from clinical blood samples, which results in practical application for faster decision-making in administering appropriate antibiotics to patients.

Results and discussion

Workflow of microfluidic AST platform

The microfluidic AST platform outlines the sequential steps involved in assessing bacterial susceptibility to antibiotics using a highly multiplexed droplet-based system. Fig. 1 illustrates the schematic workflow of the droplet antibiotic susceptibility testing (AST) process. The entire process can be broken down into five key steps: (1) generation of the concentration gradient of antibiotics, which involves creating a concentration gradient of antibiotics by mixing with bacterial cell suspension.⁴⁴ This is a crucial step where antibiotics and bacterial cells are introduced into the microfluidic system, and a gradient of antibiotic concentrations is established through a simple dilution strategy. (2) Droplet generation and encapsulation of bacteria: once the concentration gradient is achieved, the next step involves the generation of individual droplets containing the bacteria and antibiotics. This is typically facilitated by microfluidic techniques that enable precise droplet formation. (3) Incubation of droplets: the generated droplets, each encapsulating bacterial cell at specific antibiotic concentrations, are then subjected to incubation at 37 °C. The bacteria within the droplets interact with the antibiotics, and their growth or inhibition of bacteria in response to the antibiotic is monitored. (4) Image processing and bacteria enumeration: during the incubation, time-lapsed images of phase contrast microscopy are captured to track the growth of bacterial cells within each droplet. These images are then processed using image analysis software to enumerate bacterial cells. (5) Determination of MIC: the growth patterns and behavior of bacterial cells within the droplets are analyzed to determine the MICs of antibiotics. This stage involves comparing growth rates across different antibiotic concentrations to identify the threshold concentration at which growth is impeded. Overall, this approach enables rapid and label-free determination of antibiotic efficacy against bacterial cells, providing valuable insights into the effectiveness of different antibiotics and the potential presence of antibiotic-resistant subpopulations within a bacterial culture.

Fig. 1 Scheme of the microfluidic AST process workflow. The process consists of five main parts: (1) generation of the concentration gradient of antibiotics, (2) encapsulation of bacteria in droplets, (3) incubation of droplets, (4) image processing and enumeration of bacteria within a droplet, and (5) determination of MICs.

Generation of a concentration gradient in droplets

This study aims to develop a microfluidic device capable of generating droplets with varying concentrations of antibiotics, enabling comprehensive AST within a single device. Fig. 2 illustrates the methodology for creating a specific antibiotic concentration gradient within microfluidic channels and visually demonstrates the methodology employed to establish a concentration gradient for subsequent experimentation or analysis.

Fig. 2 Generation of the concentration gradient. (A) Simplified illustration of the microfluidic network composed of eight sets of flow-focusing generators connected to three microchannels. (B) A microscopic image of two aqueous phases to visualize the flow with different volume proportions under fluorescence microscopy. (C) Quantitative analysis of the expected concentration in numerical simulation (red circle; dotted line) and the normalized concentration derived from average fluorescence intensities measured within microfluidic channels (black square; solid line). (D) A photograph of the final assembled microfluidic device.

The schematic in Fig. 2A depicts the setup for gradient generation, where the flow through channels of differing widths facilitates the formation of an antibiotic gradient. The microfluidic device integrates eight parallel flow-focusing droplet generators, each assigned for the distinct concentration of antibiotics. In these generators, two different fluids converge and flow side by side in a smooth, orderly manner, thanks to channels of different widths. The dilution ratio for the two solutions is predetermined by the widths of the two channels—the antibiotic solution channel and the cell solution channel.

To verify the accuracy of the gradients generated, we perform both simulation and a flow visualization experiment (Fig. 2B and S1†). Our proposed gradient generator is operated by adjusting the volumetric flow ratio of the two solutions to achieve the predetermined dilution.⁴⁵Table 1 presents specific details regarding the width of the channel, volumetric flow rates, and expected concentrations. This calculation is derived from the Hagen–Poiseuille equation, which describes the relationship among the pressure drop, fluidic resistance, and volumetric flow rate. The equation below illustrates this relationship, enabling the calculation of expected concentrations based on the ratio of channel widths:

where C_i represents the expected concentration from channel i; C₁ and C₂ represent the concentrations of two stock solutions 1 and 2, respectively. W_i,1 and W_i,2 denote the widths of the bifurcated channels for the antibiotic solution (1) and cell solution (2), respectively. Q_i,1 and Q_i,2 represent the normalized volumetric flow rates of laminar flows of stock solutions 1 and 2, respectively.

Table 1 Evaluation of expected concentration based on different ratios of channel widths

Channel index (i)	1	2	3	4	5	6	7	8
Width of the channel of antibiotics (Wi,1, μm)	50	70	90	110	130	150	170	190
Width of the channel of cell suspension (Wi,2, μm)	190	170	150	130	110	90	70	50
Volumetric flow rate of antibiotics (Qi,1)	2.25	3.15	3.98	4.75	5.47	6.14	6.76	7.34
Volumetric flow rate of cell suspension (Qi,2)	7.34	6.76	6.14	5.47	4.75	3.98	3.15	2.25
Expected concentration (Ci)	23.5	31.8	39.4	46.5	53.5	60.6	68.2	76.5

---

Numerical simulations are employed to validate the feasibility of our microfluidic gradient generator, offering the guidance for subsequent experiments (ESI† and Fig. S1). The numerical simulation primarily focuses on characterizing flow patterns and solution exchanges within the microfluidic device, utilizing the incompressible Navier–Stokes equation. The simulation results demonstrating the dilution of two stock solutions show a gradual increase in chemical concentrations from left to right, corresponding to the relative widths of the channels. Visualization through a color gradient from blue to red represents the changing concentrations. This simulation confirms the efficiency and accuracy of our approach in achieving the desired concentration gradients of antibiotics within the microfluidic device.

To confirm the dilution ratio across the channels and validate the establishment of the concentration gradient within the microfluidic device, a buffer solution (water) and a fluorescein solution are used. The two liquids are introduced at a junction and subsequently mixed by eventually forming droplets with expected concentrations. When the volume proportions of the solution with the specific concentration are introduced from different widths of a channel in laminar flow, the final concentrations are continuously changed, respectively. Therefore, the volume proportions of two flows pre-arranged in a microchannel produce a concentration gradient on demand. To achieve a consistent concentration gradient, an ordered laminar flow within a microfluidic network is created using eight sets of fixed inlets (Fig. 2). These sets of merged channels are connected to an immiscible oil fluid channel for subsequent droplet generation. The microfluidic design incorporates a cross-junction, where the main microchannel intersects with two aqueous fluids and two perpendicular oil fluid channels. In this setup, the dispersed phase fluid within the main channel interacts with the continuous phase fluid from the side, undergoing a hydrodynamic flow-focusing effect that generates symmetrical shear forces. This process leads to the production of microdroplets downstream in the main channel. Fig. 2B shows overlapped images of microchannels under eight sets of laminar flows, one with green flow denoting the mass fraction of antibiotic fluid and the other representing the mass fraction of cell fluid. The width of the fluorescein solution gradually expands from channel 1 to channel 8 (Movie S1†). Average fluorescent intensities measured in the cross-sectional area of the channels also increase with the width of the fluorescein solution.

Our results are presented as a ratio of final fluorescence intensity to initial fluorescence intensity. The quantitative evaluation aligns consistently with the simulation result (Fig. 2C). The microfluidic device design precisely generates a well-defined concentration profile. Linear regression analysis demonstrates a high correlation coefficient (R² = 0.999), indicating the accuracy and precision in establishing our concentration gradient. To further estimate the concentration gradient of antibiotics in a droplet, fluorescence images of droplets are collected using fluorescein in place of antibiotics in the source channel. Considering the similar molecular weights between the general antibiotics tested in this study (ranging from Mw 385.82 to Mw 575.68) and fluorescein molecule (Mw 376.28), we infer similar diffusion characteristics, leading to the analogous formation of the concentration gradient at the flow-focusing junction, as depicted in Fig. 2.

The fluorescence intensity profiles of collected droplets in each channel indicate changes in intensity from channel 1 to 8, similar to the results shown in Fig. 2 and S2.† The concentration profiles can be easily derived from these intensity profiles due to their excellent linear relationship. Moreover, ensuring stable flows within the microfluidic channels is crucial to maintaining a consistent profile throughout the experiment. The concentration profile of fluorescein serves as a calibration curve for determining antibiotic concentrations across the channels. These calibrations reveal that the consistency in generating concentration gradients varied by less than 3%.

Determination of MIC by droplet AST

For the proof of concept, the microfluidic AST platform demonstrates its precision in determining the minimum inhibitory concentrations (MICs) for the reference of the strain of Gram-negative E. coli and Gram-positive S. aureus against eight clinically significant antibiotics, as listed in Table 4. The response of these bacteria to the selected antibiotics is monitored by observing the increase in bacterial count within a droplet. We have defined the MIC as the concentration where the fold increase in the average bacterial count does not exceed 2.0 throughout the experiment, ensuring that bacterial growth is effectively inhibited while maintaining the consistency of the assay. Simultaneously, we measure the MIC using the broth microdilution method (BMD) as a standard AST assay to comprehensively compare our system's performance. This parallel assessment confirms the feasibility of our droplet-based AST method's outcomes. It not only allows for quantifying antibiotic sensitivity but also lays a robust foundation for evaluating the performance of our microfluidic AST platform.

Susceptibility of Gram-positive S. aureus to gentamicin

Firstly, we examine the Gram-positive S. aureus NCTC 8325-4 for gentamicin susceptibility in a droplet (Fig. 3). Gentamicin is a potent aminoglycoside antibiotic recognized for its effectiveness against a broad spectrum of bacteria, including S. aureus. It operates by binding to the 30S subunit of bacterial ribosomes, leading to the disruption of protein synthesis and ultimately bacterial death.⁴⁶ We select a range of gentamicin concentrations (ranging from 0.17 to 0.61 μg mL⁻¹) for the AST of S. aureus.

Fig. 3 Gentamicin susceptibility test of Gram-positive S. aureus. (A) Time-lapse images of bacteria proliferation. Scale bar: 50 μm. (B) Bacterial growth curves with gentamicin (0–0.61 μg mL⁻¹). The red dotted line indicates that the average number of cells per droplet is 2. Data obtained from n ≥1000 droplets and represented as average number ± standard error of the mean (SEM). (C) Result of broth microdilution (BMD) AST in a 96-well plate to assess antibiotic efficacy.

As shown in Fig. 3, the growth of S. aureus in response to different concentrations of gentamicin is examined through direct cell counting within individual droplets. As the gentamicin concentration increases, growth is gradually inhibited, resulting in an extended doubling time ranging from 65.3 to 425.5 minutes (Table 2 and Fig. 3B). In contrast, in the absence of antibiotics, S. aureus shows a doubling time of 33.2 minutes and maintains its natural spherical morphology within the droplets (Table 2 and Fig. S4†). This doubling time is similar to previous findings from other research groups.⁴⁷

Table 2 Doubling time of S. aureus under different concentrations of gentamicin

Concentration (μg mL^−1)	Control	0.17	0.23	0.29	0.35	0.41	0.48	0.54	0.61
Doubling time (min)	33.2	65.3	94.2	131.9	168.5	194.5	198.1	300.4	425.5

---

The lower concentrations below 0.29 μg mL⁻¹ do not inhibit growth significantly while higher concentrations (ranging from 0.29 to 0.61 μg mL⁻¹) demonstrate significant growth inhibition in droplets (Fig. 3A and B). Notably, we also observe the evidence of bacterial replication even at lower concentrations (<0.29 μg mL⁻¹) after 1 and 1.5 hours of incubation time (Fig. 3B). Based on these observations, the MIC of gentamicin for S. aureus in droplets is determined to be 0.29 μg mL⁻¹. The response of the strain in the standard BMD assay indicates that the lower concentrations of gentamicin (below 0.25 μg mL⁻¹) do not significantly inhibit growth (Fig. 3C). Our detected MIC (0.29 μg mL⁻¹) for S. aureus is similar with the range of MIC values (≥0.25 μg mL⁻¹) obtained from BMD assay and reported in the literature.⁴⁸ This alignment confirms the reliability and consistency of our droplet-based AST approach.

Susceptibility of Gram-negative E. coli to tetracycline

We next investigate the performance of AST using Gram-negative E. coli O157:H7 in response to the addition of tetracycline because E. coli O157:H7 serves as an example of Gram-negative and tetracycline-susceptible bacteria. In our study, we evaluate this strain within the range of 0.42 to 1.52 μg mL⁻¹ tetracycline (Fig. 4 and Table 3). The average doubling time for E. coli O157:H7 within the droplets without antibiotics is determined to be 42.7 minutes, aligning well with the published range of doubling time.⁴⁹

Fig. 4 AST of Gram-negative E. coli versus tetracycline. (A) Bacterial growth curves with tetracycline (0–1.52 μg mL⁻¹) in droplets. The red dotted line indicates that the average number of cells per droplet is 2. Data obtained from n ≥1000 droplets and represented as average number ± standard error of the mean (SEM). (B) The result of broth microdilution (BMD) AST in a 96-well plate to assess tetracycline efficacy against E. coli.

Table 3 Doubling time of E. coli O157:H7 under different concentrations of tetracycline

Concentration (μg mL^−1)	Control	0.42	0.57	0.71	0.87	1.02	1.19	1.35	1.52
Doubling time (min)	42.7	49.0	70.5	107.3	143.5	276.4	751.8	16 872.5	143 374

---

Similar to the experiment with Gram-positive S. aureus, tetracycline concentrations showing an average cell count of 2 or fewer start from 1.02 μg ml⁻¹ (Fig. 4A). Based on these assessments, we estimate the MIC for E. coli O157:H7 to be 1.02 μg mL⁻¹. The standard BMD assay confirms the results obtained from the droplets and provides comparable results of E. coli O157:H7 growth inhibition at 1 μg mL⁻¹ tetracycline (Fig. 4B). In addition, the doubling time is extended from 42.7 to 276 minutes at tetracycline concentrations of 1.02 μg mL⁻¹ within the droplets (Table 3). Therefore, the results of droplet-based AST analysis and the conventional BMD assay are consistent for Gram-negative E. coli as well, proving that the droplet-based AST analysis method proposed in this study is effectively applicable to most bacteria.

By utilizing this droplet AST platform, the process of MIC determination is significantly expedited, taking around 180 minutes, while the BMD assay requires about 18 to 24 hours to yield results. This rapid prediction capability holds the potential to significantly reduce turnaround times in scaled-up screening applications. It is worth noting that the rapid response time depends on the pathogen type and its doubling time as well as the specific antibiotic being tested. Generally, 1 to 2 cycles of proliferation are sufficient to distinguish distinct growth patterns in the presence of antibiotics. Bacteria with shorter doubling times may exhibit faster detection in the droplets. Overall, our findings suggest that the droplet platform is applicable for evaluating both Gram-positive and Gram-negative strains. This versatility positions it for potential future applications in preclinical and clinical settings.

To verify the applicability of our droplet AST method across a range of antibiotics, we have tested it against both Gram-positive and Gram-negative bacteria with various antibiotics that have different mechanisms of action. A comprehensive comparison of MIC values measured for a range of antibiotics targeting cell wall synthesis, protein synthesis, DNA synthesis, and mRNA synthesis in E. coli and S. aureus is presented in Table 4. The MIC values obtained from the droplet-based AST method are compared with those from the conventional BMD assay, illustrating the efficacy and accuracy of the droplet approach across different bacterial targets and antibiotic mechanisms of action. Interestingly, the results obtained from the droplet AST are in close agreement with those from the BMD assay. For E. coli O157:H7, the MIC values measured using the droplet AST method show no significant discrepancy when compared to those obtained with the BMD method, indicating that our droplet-based approach aligns well with the sensitivity of the conventional BMD method. Similarly, for S. aureus NCTC 8325-4, the droplet AST method tends to report identical MICs. Thus, we confirm the droplet-based microfluidic AST system's potential for precise determination of the antibiotic concentrations needed to inhibit bacterial growth. The results confirm the potential of the droplet-based method as a rapid and reliable alternative to the BMD assay for determining antibiotic susceptibility. The droplet microfluidic AST system thus provides a fast, reliable analysis of antibiotic susceptibility and can be precisely applied in the future as a diagnostic approach to minimize usage of broad-spectrum antimicrobials in large doses.

Table 4 Comparative analysis of MIC determination using droplet-based AST and standard BMD assay

	Target	Antibiotics	Droplet	BMD
E. coli O157:H7	Cell wall synthesis	Ampicillin	3.57	4
		Ceftazidime	2.54	2
		Ceftriaxone	0.17	0.125
	Protein synthesis	Gentamicin	0.53	0.5
		Tetracycline	1.02	1
		Kanamycin	2.18	2
	DNA synthesis	Ciprofloxacin	0.03	0.015
		Levofloxacin	0.36	0.25
S. aureus NCTC 8325-4	Cell wall synthesis	Oxacillin	0.35	0.25
		Cefepime	1.57	2
		Vancomycin	0.65	0.5
	Protein synthesis	Gentamicin	0.29	0.25
		Tetracycline	0.43	0.25
		Erythromycin	0.60	0.25
	DNA synthesis	Ciprofloxacin	1.01	0.25
	mRNA synthesis	Rifampin	0.05	0.06

---

Image analysis for the enumeration of bacteria

As our approach is designed to be a universal platform capable of testing any species of bacteria against any type of antibiotic, practical validations for clinical samples against several different antibiotics have to be confirmed. While previous microfluidic AST approaches rely on fluorescence detection of bacteria using fluorescent markers or viability indicators, we propose a relatively inexpensive, rapid, and widely applicable phase contrast microscopy-based technique, which solves significant conventional limitations by implementing a label-free approach.

Fig. 5 shows our simple workflow for counting bacterial cell numbers within droplets using phase contrast images to eliminate the need for additional labeling processes. Firstly, after collecting droplets from the microfluidic device, we carefully place them into a specially designed grid to arrange them in a monolayer within a predetermined area. Subsequently, we capture time-lapse phase contrast images of droplets using a 20× magnification objective lens and an automated x–y stage (Fig. 5A). In this study, we capture images of each of the approximately 1000 droplets every 30 minutes.

Fig. 5 The enumeration of bacteria within droplets. The segmentation workflow and enumeration of cells is performed using ImageJ software. (A) First, phase-contrast images are loaded, and (B) the background is separated from the cells by inverting the intensity profile through the “invert” function in image analysis software. An inverting algorithm is applied to create a binary mask and we obtain black and white inverted image. (C) This mask separates the background from the foreground and is then used to find the center of individual cells by measuring each pixel's Euclidean distance to the background. The pixels with the highest Euclidean distance within their surroundings are considered as peaks, which are used to define individual cells. The scale bar indicates 50 μm. (D) Comparison chart for the correlation between phase contrast image-based counted cells and fluorescence-based counted cells (n = 3825). Each point on the plot represents the median phase contrast count for corresponding fluorescence levels, with error bars extending to show variability within the data set.

To analyze the individual droplets, we first define the sub-regions that each droplet occupies within a frame as the region of interest (ROI). The images are separated into the feature of interest (droplets) and the background using a modified IsoData thresholding algorithm and converted to a binary image based on the threshold setting. ROI segmentation is performed by scanning the image and finding the edge of the droplets, for which the information of ROI is recorded based on the x–y coordinates of the images. The droplets that contain bacterial cells are then sorted out from the other empty droplets that lack any cells and selected as the target to be analyzed.

In these images, bacterial cells appear darker than the surrounding droplet background (Fig. 5A). To segment single bacterial cells within pre-defined ROIs of droplets in phase contrast images, the bacterial cells are highlighted by inverting the intensity profile through the “invert” function in ImageJ (Fig. 5B). The local maximum intensity is determined in each droplet, and individual cells are segmented using the Euclidean distance map, assuming that each maximum belongs to a single cell. Finally, the outcome displays the number of maxima counted in every individual droplet (Fig. 5C). This label-free method clearly enumerates the change in cell numbers residing within an individual droplet. The proposed cell counting method is validated by comparing the results with a fluorescence count of cells (Fig. 5D). The average number of cells counted within droplets is consistent among these methods, with an error of less than 5% for each time point. The approach developed has broad applicability and utility for researchers with standard laboratory imaging equipment.

This approach offers the distinct advantage of allowing clear differentiation between bacterial cells and the background medium within droplets. This differentiation is based on the refractive indices of these components. Processing of the resulting images produces an image that can be easily enumerated using conventional open-source software (ImageJ). The number of cells can be used to quantify the susceptibility profiles of antibiotics. Thus, this approach allows for the analysis of bacterial cell numbers without any labeling processes and holds potential for real-world clinical settings. We overcome the limits in using fluorescently tagged biomolecules or staining, which might impact cell viability and limit clinical applicability due to high costs and time-consuming design.

Furthermore, we develop an automated image analysis framework which is optimized for the quantification of bacterial cells within the droplets in phase contrast images, allowing the users to extract information on replication features. Manual assignment of image processing steps involved in bacterial enumeration is time-consuming, laborious, and prone to human error when a large number of data sets or images are taken. A customized algorithm by using the ImageJ-associated plugins has been established for the entire, repetitive processes.

Determination of MIC for clinical isolates

Subsequently, we examined the clinical applicability of our microfluidic droplet AST by testing the susceptibility of clinical isolates from blood samples which were submitted to the Kangdong Sacred Heart Hospital (KSHH). The clinical isolates consist of both methicillin-resistant and multidrug-resistant S. aureus. We have carried out AST with five antibiotics such as oxacillin, gentamicin, ciprofloxacin, linezolid, and vancomycin.

The MICs of five CLSI-recommended antibiotics are determined through automatic image analysis (Table 5 and Fig. S5†). In this experiment, there are specific isolates exhibiting normal bacterial growth within droplets even under high antibiotic doses (exceeding 64 μg mL⁻¹), which are evaluated to be resistant to specific antibiotics. Interestingly, the MIC of methicillin-resistant S. aureus (MRSA; HL-SA-16278, HL-SA-18807, HL-SA-18888, HL-SA-20835, and HL-SA-21008) and multidrug-resistant S. aureus (MDRSA; HL-SA-17064, HL-SA-18840, and HL-SA-18883) cannot be defined within the concentration range of oxacillin because these strains exhibit normal growth profile even at extremely high concentration of oxacillin (64 μg mL⁻¹), indicating a resistant phenotype (Table 5). Further susceptibility testing is carried out for gentamicin, where methicillin-resistant strains (MRSA) are considered susceptible, while all MDRSA strains show a resistant phenotype (Table 5). By comparing the results of both strain groups, we successfully distinguish the resistant phenotype from the susceptible strains.

Table 5 The antibiotic susceptibility of clinical isolates

Type	Bacterial strain	Droplet AST MIC (μg mL^−1)
		Oxacillin	Ciprofloxacin	Gentamicin	Linezolid	Vancomycin
a In this study, microbial strains that demonstrate a doubling or more in population size across all tested antibiotic concentrations are defined as “resistant”. For these resistant strains, determining the minimum inhibitory concentration (MIC) is challenging due to their sustained growth despite antibiotic exposure. The MIC is operationally defined as the lowest concentration of an antibiotic at which there is no more than a twofold increase in the microbial population within the droplets. This definition allows for a quantifiable assessment of antibiotic resistance levels.
Methicillin-resistant S. aureus	HL-SA-16278	67.55	1.35	0.53	2.27	1.53
	HL-SA-18380	8.11	16.27	0.72	2.43	1.07
	HL-SA-18807	Resistant^a	107.5	0.71	1.02	1.31
	HL-SA-18888	78.4	Resistant	0.87	1.35	1.07
	HL-SA-20835	Resistant	39.2	1.02	1.02	1.07
	HL-SA-21008	Resistant	0.64	0.64	1.19	1.79
Multidrug-resistant S. aureus	HL-SA-17064	Resistant	Resistant	Resistant	2.03	1.31
	HL-SA-17078	8.11	19.11	65.34	3.40	1.07
	HL-SA-18840	Resistant	68.2	Resistant	1.69	1.53
	HL-SA-18883	Resistant	Resistant	107.5	1.19	1.31

---

Ciprofloxacin, a fluoroquinolone antibiotic, is used for the treatment of MRSA infections, especially oral infections.⁵⁰ Unfortunately, since the introduction of ciprofloxacin to treat MRSA strains, these bacteria rapidly become resistant to ciprofloxacin. The prevalence of ciprofloxacin-resistant methicillin-resistant S. aureus (CR-MRSA) strains is increasing, and studies showed that a high number of MRSA strains are resistant to ciprofloxacin.⁵¹ Our investigation also confirms that two MRSA and three MDRSA strains exhibit a resistant phenotype, suggesting that the use of ciprofloxacin may lead to the failure of conventional antibiotic therapy against MRSA or MDRSA infections. Moreover, several previous reports reveal that biofilm formation is one of the major factors.⁵²

Lastly, the susceptibility of linezolid and vancomycin, considered susceptible for both methicillin-resistant and multidrug-resistant strains, is determined, yielding the MIC values (Table 5). In the case of linezolid, the first oxazolidinone class of antibiotic approved by the Food and Drug Administration (FDA) in 2000 for the treatment of MRSA infections, has a broad range of activity against antibiotic-resistant Gram-positive bacteria. Since then, significant clinical studies have been conducted to evaluate the effectiveness of linezolid against serious infections, enhancing our understanding of its capabilities. Current treatment options for MRSA or MDRSA infections are limited. Linezolid has proven to be a valuable antibiotic against this common and dangerous pathogen. This is evident in the inclusion of linezolid in multiple clinical practice guidelines. Furthermore, our results from the droplet AST method on linezolid for clinical strains are valuable for clinicians, providing reassurance regarding the drug's effectiveness (Table 5). In addition, vancomycin is one of the oldest antibiotics and exerts its bactericidal action by interrupting proper cell wall synthesis in susceptible bacteria. It has been in clinical use for more than 60 years for the treatment of drug-resistant Gram-positive infections. However, there are significant concerns due to decreasing susceptibility to vancomycin among S. aureus (VRSA). Interestingly, our experimental result shows the high susceptibility of clinical strains, which aligns well with conventional guidelines. S. aureus remains highly susceptible to vancomycin, and vancomycin remains the initial antibiotic of choice for the treatment of patients with MRSA bacteremia and endocarditis.

In total, we evaluate 10 clinical samples for MIC using both the droplet AST and the standard BMD assay (Tables 5 and S1†). We observed an overall accuracy of 94.5% in determining the MIC between the two methods. Further, we calculated the matching rate for each bacterial strain as the number of samples that matched (m) divided by the total number of bacteria samples of that type (n). The clinical isolates show a high matching rate of 95%. Thus, our analysis does not find a statistically significant difference in the probability of achieving an accurate MIC using the droplet AST method in comparison to the BMD.

The current approach abstains from utilizing any fluorescence molecules or genetically modified bacteria. Instead, it demonstrates one-pot simplicity and applicability to concurrently assess several antibiotics within a label-free assay. Rapid image analysis combined with readily available open-source image software can accurately enumerate individual cells even with extensive experimental data. This streamlined approach not only reduces analytical time and costs but also enhances throughput.

Heterogeneity in bacterial growth profiles in droplets

Furthermore, we investigated the feasibility of the expansion of our approach to enhance the capabilities of the single-cell AST platform beyond simply determining how quickly bacteria become resistant to antibiotics in droplets. It allows for the examination of different bacterial subgroups within each droplet in the presence or absence of antibiotics by differential growth profiles (Fig. 6A). To summarize bacterial growth profiles throughout the experiment, we have classified three distinct patterns: active growth for an increase of 4 or more bacteria, slow growth for an increase of 1 to 4 bacteria, and inhibited growth for an increase of 1 bacterium or less.

Fig. 6 Phenotypic heterogeneity of bacterial growth in a droplet (A). Each experimental graph shows the percentage of droplets with active growth (≥4 generations, black), slow growth (1–4 generations, dark grey), and inhibited growth (≤1 generation, white gray) of S. aureus NCTC 8325-4 over 3 h (B). Phenotypic heterogeneity of clinical isolates identified with MRSA (C) and MDRSA (D). Data obtained from over 3000 droplets for each condition.

For example, in the study of S. aureus NCTC 8325-4, a clear pattern emerges in response to the exposure of several antibiotics (Fig. 6B). In the absence of antibiotics, droplet-based AST reveals active growth in 100% of the cases. However, the response of S. aureus to varying oxacillin concentrations demonstrates different growth dynamics. At lower concentrations of oxacillin MIC (≤0.29 μg mL⁻¹), a considerable number of cells continue to show active growth, suggesting tolerance to these sub-MIC levels of the antibiotic. As the concentration of oxacillin increases, the percentage of actively growing cells declines sharply. Notably, once the concentration exceeds the MIC (≥0.35 μg mL⁻¹), the proportion of cells with active growth drops to approximately 10%, while the profiles of inhibited and slow growth become more pronounced. Even at higher oxacillin concentrations (ranging from 0.48 to 0.61 μg mL⁻¹), a small subset of droplets (below 3%) exhibits active growth. Concurrently, the incidence of slow growth in droplets varies from 3% to 45%, increasing with the oxacillin concentration. This trend is particularly significant at the MIC of oxacillin (0.35 μg mL⁻¹), where about 10% of the droplets still display active growth, confirming the heterogeneity in antibiotic susceptibility among S. aureus populations. This phenomenon, referred to as phenotypic heterogeneity, wherein cells within an isogenic culture exhibit different responses to the same antibiotic concentration, aligns with previous studies in bacterial research.⁵³ The methodological approach of confining a small number of bacterial cells (1 to 4) within individual droplets, combined with dynamic monitoring, is essential in discerning these distinct bacterial growth profiles. This approach offers an insight into the understanding of bacterial behavior under antibiotic stress and highlights the heterogeneity of bacterial populations to antibiotic treatment.

In addition to oxacillin, our investigation extends to the bacterial responses against different antibiotics including ciprofloxacin, gentamicin, and vancomycin (Fig. 6B). Our findings provide a multifaceted bacterial adaptability, revealing distinct patterns of growth in response to different antimicrobial agents. Notably, we observe a significant decline in the proportion of bacteria exhibiting active growth at concentrations exceeding the MICs for these antibiotics. Interestingly, despite the reduction in active growth, a considerable portion of the bacterial population continues to exhibit slow growth. This persistence suggests the potential for these subpopulations to contribute to the development of antibiotic resistance. The sustained slow growth in the presence of antibiotic concentrations above the MIC underscores the complexity of bacterial adaptation and emphasizes the critical importance of understanding these dynamics in the context of antibiotic resistance.

Further proving the applicability of our method, we subsequently examined the heterogeneity of clinical isolates by performing single-cell AST on both methicillin-resistant S. aureus (MRSA) and multidrug-resistant S. aureus (MDRSA) against five antibiotics such as oxacillin, gentamicin, ciprofloxacin, linezolid and vancomycin (Fig. 6C and D). Our method can classify that substantial proportions of droplets exhibit active, slow, and inhibited growth at the MIC levels of antibiotics tested on clinical samples. For instance, when clinical strains of MRSA and MDRSA are subjected to oxacillin treatment, we find that resistant cells determined by our method show a significant portion of the active cells at the whole concentration range of antibiotics. In contrast, the susceptible cells such as HL-SA-16278, 18380, 18888 (MRSA), and 17078 (MDRSA) interestingly exhibit a small portion of active growing cells from 0% to 11% above MIC concentration (Fig. 6C and D).

In the case of ciprofloxacin, resistant cells (HL-SA-18888, HL-SA-17064, and HL-SA-18883) show a small proportion of active growth ranging from 20% to 50% above the MIC level, while susceptible cells include a small portion of active growing cells from 1% to 9%. This observation suggests a significant subpopulation with compromised susceptibility and the cells, often characterized by active or slow growth in the presence of antibiotics, can contribute to the reduced effectiveness of drug treatment.

Likewise, exposure to gentamicin shows that although MIC analysis from BMD assay (Table S1†) indicates resistance in every MDRSA strain, droplet AST reveals the precise MIC of HL-SA-17078 and 18883 strains, also containing a significant portion of active growth from 30% to 47% just below the MIC. Although other susceptible strains determined by droplet AST show a similar MIC obtained from BMD assay, our approach indicates that there is a smaller portion of active growth above the MIC.

For linezolid and vancomycin, despite every clinical strain being classified as susceptible as shown in Fig. 6 and Tables 5 and S1,† a small portion of samples exhibit active growth under linezolid treatment, whereas vancomycin treatment eliminates most of the active growing cells (below 5%) above the MIC. This suggests that a surviving subpopulation within these strains exhibits compromised susceptibility to the antibiotic, raising concerns for treatment efficacy. These findings underscore the importance of recognizing distinct subpopulations within bacterial cultures, which may exacerbate antimicrobial resistance and potential treatment failures.

These findings provide evidence of phenotypic heterogeneity among inhibited, slow, and actively growing bacteria. Phenotypic heterogeneity has previously been observed in bacteria, where cells from an isogenic culture do not respond uniformly to the same drug dosage.⁵⁴ The encapsulation of bacterial cells in droplets, coupled with the monitoring of cell numbers, enables the verification of distinctive growth profiles within both isogenic and polymicrobial communities. Thus, our method proves that substantial proportions of all clinical samples, including both methicillin-resistant S. aureus (MRSA) and multidrug-resistant S. aureus (MDRSA), exhibit active and slow growth above the MIC concentrations of antibiotics tested. However, the conventional BMD assay fails to provide this detailed information, highlighting a significant discrepancy between the two methods. It remains unclear whether these differences in drug susceptibility across strains are intrinsic or acquired properties, and the underlying mechanisms have yet to be elucidated. However, it is noteworthy that droplet AST allows for the detection of a small proportion of resistant bacteria, unlike standard AST methods. Also, it underestimates the presence of heterogeneity at both the lower (active) and higher (inhibited) boundaries of growth, suggesting that the observed distinct subpopulations emphasize the potential risk of antimicrobial resistance, which could lead to treatment failure.

In the future, this approach will be extended to isolates from biological fluids to determine resistance in bacterial subpopulations, particularly in polymicrobial environments, and the integration of single-cell technologies will facilitate the development of the most rapid AST method.^55,56 The turnaround times reported for the result of AST are approaching the necessary time frame for preventing the empirical prescription of broad-spectrum antibiotics. The rapid single-cell AST technologies excel in promptly determining whether a bacterium is susceptible or resistant to a specific antibiotic.

Conclusions

In these days, the emergence of drug-resistant pathogens necessitates the advancement of rapid antibiotic susceptibility testing (AST) methods. Nonetheless, traditional AST methods are widely implemented in current clinical microbiology laboratories and often involve blood culture to separate the target pathogen selectively and bacterial culture to amplify cell density for detection, which demands 2 to 5 days. Despite the rapid evolution of pathogen identification techniques through molecular diagnostics, the capability to promptly assess bacterial susceptibility remains an unresolved requirement in clinical microbiology.

This study presents a rapid droplet AST that determines MIC values against bacteria. This is accomplished by integrating a mechanism to create a concentration gradient of antibiotics and simultaneously segmenting them into droplets within a single device. From a technical perspective, cultivating obligate aerobes in droplets surrounded by an immiscible oil phase has challenges due to limited and uneven oxygen supply, which is critical for cell metabolism and growth during incubation. A previous study proposed an effective solution using continuous carrier oil recirculation to enhance oxygen transfer to the droplets, maintaining stable oxygen concentrations during incubation in a dynamic droplet incubation (DDI) system.⁵⁷ Nevertheless, this strategy holds the potential to curtail assay time, consuming a minute sample and reagent within a picoliter-sized droplet that can be employed for the early recognition of bacterial proliferation. Moreover, it allows for the direct assessment of antimicrobial susceptibility from clinical samples at a single-cell level.

Our approach is built upon three fundamental concepts. Firstly, it enables the creation of droplets composed of multiple antibiotic concentrations in a predetermined manner along with cell encapsulation, incubation, and growth monitoring. This facilitates testing a spectrum of antibiotic concentrations within a single device, ensuring precise MIC determination without extensive repetition. Secondly, the inherent swiftness of single-cell AST stems from the confinement of individual bacteria within pico-droplets, promoting the capability to detect small changes in a number of bacteria and thereby leading to shortened assay times. Additionally, the analysis of 10 000 droplets for each antibiotic concentration permits the disclosure of phenotypic heterogeneity as well as statistically reliable and clear differentiation of minute bacterial growth changes due to antibiotics within just 180 minutes. Thirdly, the establishment of image processing and enumeration workflow with customized algorithms accelerates the turnaround time of AST performed. It also allows the investigation of the clinical isolates without an additional labeling process, which makes this method clinically amenable with minimized human error. By successfully integrating these three features, this methodology enhances scalability and maintains the speed of droplet microfluidic-based single-cell AST.

Furthermore, this approach unveils phenotypic heterogeneity at the single-cell level with precision, contrasting with conventional methods that primarily focus on the MIC to infer bacterial growth inhibition. This droplet AST proves the importance of the complex dynamics of antibiotic resistance and the critical need to understand bacterial subpopulations. Recognizing cells exhibiting different growth patterns, even above MIC levels, suggests the importance of tailored strategies to address these subpopulations for effective antibiotic resistance combat and treatment optimization. Thus, our method, by accurately identifying the antibiotic concentration that genuinely inhibits bacterial growth through phenotypic heterogeneity analysis, provides a more informed basis for prescribing drugs. This precision guides clinicians towards more effective treatment strategies, potentially curbing antimicrobial resistance development and enhancing patient outcomes by customizing treatments to the unique characteristics of bacterial populations.

Experimental methods

Materials

Photoresists (SU-8 3005, 3025 and 3050) and polydimethylsiloxane (PDMS, Sylgard 184) for device fabrication were purchased from Kayaku Advanced Materials (MA, USA) and Dow Corning (MI, USA), respectively. For droplet generation, HFE 7500 oil was purchased from 3M (MN, USA) and PFPE–PEG–PFPE triblock copolymer surfactant was synthesized following a previous study.⁵⁸ The fluorescein sodium salt, Mueller–Hinton (MH) broth, Luria-Bertani (LB) broth, tryptic soy broth (TSB), antibiotics, and all reagents not mentioned were purchased from Sigma-Aldrich (MO, USA).

Design of the microfluidic device

The microfluidic device for rapid AST must conduct a series of actions including an injection of reagents, mixing of antibiotic and bacteria solution, formation of concentration gradient, and encapsulation of bacteria in emulsion droplets. The overall microfluidic device mainly consists of the upper layer, lower layer, and the glass substrate coated with PDMS. The upper layer of the device has three inlets for the oil phase and two aqueous phases, each coupled with the microchannel that serves as a reservoir and distributes a fluid into a flow-focusing channel, and outlets.^59,60 The microchannels placed in a horizontal direction connect all inlets for eight sets of flow-focusing droplet generators in a ladder geometry, facilitating the injection of solutions simultaneously.

The channels organizing a microfluidic network of flow-focusing droplet generators lie in the lower layer of the microfluidic device. The microchannels in the lower layer are designed to have a flow resistance much greater than that of the microchannels in the top layer to ensure the production of monodisperse droplets. Specifically, we apply a previously established design rule to realize the even distribution of fluids flowing from the microchannel in the top layer into each set of flow-focusing network as described below,

2N_f(R_b/R_t) < 0.01

where R_t and R_b are the flow resistance along the microchannel in the upper layer and the lower layer, respectively; and N_f is the number of flow-focusing generators. As eight sets of flow-focusing generators are used in the microfluidic network, we designed the microchannels in the top layer with width, length, and height of 3 mm, 47.7888 mm, and 0.75 mm, respectively (Fig. S3†).

Furthermore, the same approach is applied to the design of flow-focusing generators. Due to the microchannels positioned in the upper layer of the microfluidic device, the length of the channel from the inlet to the junction for two aqueous phase solutions is different. To minimize the effect of the difference in channel length between two solutions on the generation of concentration gradient, the height of a junction channel is decreased to 10 μm, compared to the height of the other structures (50 μm) in flow focusing generators. This design consideration results in the generation of a concentration gradient in a highly precise manner. Consequently, the effect of the difference in channel length between two aqueous solutions is negligible, enabling the control of a predetermined profile of concentration gradient by simply adjusting the width of the channel for each aqueous solution.

Fabrication of microfluidic device

The microfluidic device is fabricated through standard photolithography and multilayer soft lithography.^61,62 First, the master mold for the top layer is patterned on a 750 μm thick photoresist. In brief, 6 mL of negative photoresist SU-8 3050 is poured onto a 4-inch silicon wafer and soft-baked overnight. The following steps required for photolithography including UV exposure and post exposure baking are conducted. On the other hand, the master mold of the bottom layer is prepared by two-step photolithography using SU-8 3005 for the junction channel structures where two aqueous phase solutions converge (10 μm), and SU-8 3025 for the remaining part of the flow-focusing generators (50 μm). For the replica molding process (Fig. S6†), a mixture of PDMS oligomer and crosslinker with a ratio of 10 : 1 (w/w) is poured onto both master molds and degassed within a vacuum chamber. Then, the PDMS mixture is thermally cured in a 65 °C convection oven for 6 hours. The replica is carefully detached from the mold and the inlet ports of eight flow-focusing droplet generators in the bottom layer are produced by punching holes using a biopsy punch of 0.75 mm diameter. Subsequently, both layers are aligned and bonded to each other through a plasma treatment. Finally, the inlet and outlet through-holes of the PDMS assembly are punched and bonded to a PDMS-coated glass slide by a second plasma treatment.

A rectangular-shaped droplet screening grid is also prepared by photolithography on the glass slide for the monitoring of bacteria within the droplets in a fixed position. The droplet grid consists of 9 × 8 microwells with the dimension of 925 μm × 690 μm. As the average diameter of the droplet produced from the microfluidic device is approximately 60 μm, the height of the walls surrounding each well is set as 70 μm to prevent the squeezing of droplets when they are laid in a monolayer form.

Droplet generation and characterization of concentration gradient

To visualize the formation of a concentration gradient, we use a 50 μM fluorescein sodium salt solution and distilled water as two aqueous solutions. The solutions are filtered using a 0.22 μm pore syringe filter to remove the debris or dust in the solution. HFE 7500 oil containing 2 wt% PFPE–PEG–PFPE surfactant is prepared as a continuous phase for droplet generation. The solutions are introduced into the microfluidic device at the volumetric flow rate of 5 mL h⁻¹ and 2 mL h⁻¹ for the continuous phase and the aqueous phase, respectively. During the generation of fluorescent droplets, the bright-field optical image and the fluorescence image at the junction channels within each flow-focusing generator are captured and merged to quantify the relative concentration of fluorescein compared to the input solution.

Once droplets are generated, they flow into the outlet and are collected into Eppendorf tubes. We then measure the average intensity of more than 1000 droplets per channel and estimate the concentration gradient for the calibration of antibiotic concentration afterward.

Bacterial cell culture and sample preparation

In this study, we use E. coli and S. aureus as a representative of Gram-negative and Gram-positive strains for the demonstration of AST. For the proof of concept, E. coli O157:H7/pCM18, which constitutively expresses green fluorescent protein (GFP), and S. aureus NCTC 8325-4/phc48, which constitutively expresses red fluorescent protein (RFP), are used. For the analysis of clinical samples in droplet AST, 10 strains of methicillin-resistant S. aureus (MRSA) and multidrug-resistant S. aureus, provided by Kangdong Sacred Heart Hospital in Hallym University Medical Center, are used.

A single colony of E. coli and S. aureus is inoculated into 4 mL of LB medium and TSB medium, respectively. They are incubated overnight in a shaking incubator at 37 °C with constant agitation at 200 rpm. On the day of the experiment, we exchange and dilute the overnight bacterial culture in a fresh medium and incubate for 3 hours to reach the exponential phase of the culture. We then measure the absorbance of the cultured bacterial suspension at a wavelength of 600 nm (OD₆₀₀), which is used for the estimation of bacterial concentration according to previously established calibration. 1 mL of bacterial solution is transferred into an Eppendorf tube, centrifuged at 10 000 rpm for 2 min, and washed using 1 mL of fresh MHB medium. Finally, the solution is further diluted into the concentration corresponding to OD₆₀₀ = 0.02.

All of the antibiotic solutions are prepared as stock solutions according to the manufacturer's instructions. They are sterilized using a 0.22 μm pore syringe filter and stored in a −20 °C freezer. The solutions are thawed and diluted with MHB medium into proper concentration before use.

Droplet antimicrobial susceptibility testing (AST) in the microfluidic device

We prepare the bacterial cell suspension and selected antibiotic solution into individual 1 mL syringes connected with Tygon tubing. The volumetric flow rates of the solutions are controlled by syringe pumps (PHD 2000, Harvard Instruments, MA, USA). The initial flow rates for all oil phase and aqueous phases are set as 4 mL h⁻¹. After filling the microchannels in the top layer of the microfluidic device, the volumetric flow rates are changed into 5 mL h⁻¹ and 2 mL h⁻¹ for the oil phase and aqueous phase solutions, respectively.

The microfluidic device is placed on the microscope stage in a custom-made incubator with a built-in temperature control module. Droplets are collected into eight Eppendorf tubes. After collecting droplets from the microfluidic device for 5 min, we carefully load them into a specially designed grid to arrange them in a monolayer within a predetermined area. Assisted by the microwell structure within the grid, we can monitor the same droplet pools over time without indexing of droplets. To effectively monitor the change in the number of cells within the droplets, we commence our observation from the topmost droplets and proceed downward through the grid. This approach facilitates the comprehensive collection of cumulative growth dynamics occurring within droplets. A pair of bright-field optical images and fluorescence images are captured every 30 minutes for the demonstration of AST using E. coli and S. aureus, while the time-lapse phase contrast images are acquired for the analysis of AST using clinical isolates.

Broth microdilution (BMD) test

The standard BMD method is conducted according to the criteria provided by CLSI and used as the gold standard to compare and assess the performance of our microfluidic AST platform. Stock solutions of antibiotics are prepared by thawing and diluting into the appropriate concentration recommended by CLSI. 100 μL of the prepared antibiotic solutions are pipetted into each well of a 96-well plate, followed by the addition of bacterial solution with the final concentration of 5 × 10⁵ CFU mL⁻¹. The BMD tests are performed in triplicate. After 20 hours of incubation at 37 °C, the MIC values of the selected antibiotics are read as the concentration where the measured absorbance is below 0.1 at the wavelength of 625 nm. If the results from the triplicate test are not identical, the majority results are selected as the MIC.

Image processing and statistical analysis

Accurate and quantitative enumeration of bacteria within droplets is critical for investigating phenotypic differences in bacterial resistance. Here, we have employed a detailed image processing workflow using phase contrast microscopy to analyse bacterial growth and response to antibiotics.

1. Image acquisition: phase contrast images of droplets containing bacterial cells were captured using a 20× magnification objective lens. These images were acquired at regular intervals to monitor bacterial growth and response to antibiotics over time.

2. Preprocessing: the acquired images undergo preprocessing to enhance contrast and reduce background noise. This step included normalization and background subtraction techniques to improve the visibility of individual bacterial cells.

3. Cell segmentation: we use a segmentation algorithm specifically optimized for phase contrast images to distinguish bacterial cells from the background. This algorithm identifies cell edges based on distinct phase contrast features, allowing for accurate cell boundary detection without the need for fluorescent labels.

4. Cell counting: once cells were segmented, we use FIJI ImageJ software, an open-source image analysis tool, to count the number of bacterial cells in each droplet. The software automatically detects and enumerates the segmented cells based on their size and shape, ensuring reliable cell counting across different samples.

5. Statistical analysis: we plot the average growth curves obtained from at least 1000 droplets for each antibiotic concentration after normalizing bacterial numbers to the values of the first data point (t = 0). Error bars in the growth curves indicate the standard error of the mean (SEM). The minimum inhibitory concentration (MIC) of antibiotics in droplets is determined according to the established criteria.

The number of bacteria encapsulated into droplets indeed follows a Poisson distribution. Our experiments also confirmed this distribution pattern for bacteria encapsulation (Fig. S7†). To minimize the impact of variations in initial bacterial concentration on the AST results due to this distribution, we specifically selected droplets that contained a single bacterium. This approach allowed us to standardize the initial conditions and focus our analysis on these uniformly prepared droplets throughout the entire procedure.

Data availability

The data supporting the findings of this study are available within the article and its ESI.†

Author contributions

Jae Seong Kim: data curation, formal analysis, methodology, validation, visualization, writing – original draft. Jingyeong Kim: data curation, formal analysis, validation, methodology, Jae-Seok Kim: conceptualization, resources, validation. Wooseong Kim: conceptualization, data curation, validation. Chang-Soo Lee: conceptualization, methodology supervision, resources, funding acquisitions, writing – original draft, writing – review & editing, project administration.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (the Ministry of Science and ICT) (No. 2021R1A2C3004936 and 2021R1A5A8032895).

References

1. M. Frieri, K. Kumar and A. Boutin, J. Infect. Public Health, 2017, 10, 369–378.
2. A. Y. Hwang and J. G. Gums, Bioorg. Med. Chem., 2016, 24, 6440–6445.
3. S. Al-Saidi, F. S. Mjalli, M. Al-Azzawi, T. Al-Wahaibi, B. Abutarboosh and M. A. AlSaadi, Korean J. Chem. Eng., 2024, 41, 1399–1413.
4. J. S. Molton, P. A. Tambyah, B. S. Ang, M. L. Ling and D. A. Fisher, Clin. Infect. Dis., 2013, 56, 1310–1318.
5. H. M. Zowawi, P. N. Harris, M. J. Roberts, P. A. Tambyah, M. A. Schembri, M. D. Pezzani, D. A. Williamson and D. L. Paterson, Nat. Rev. Urol., 2015, 12, 570–584.
6. O. J. Dyar, B. Huttner, J. Schouten and C. Pulcini, Clin. Microbiol. Infect., 2017, 23, 793–798.
7. J.-K. Kim, H.-Y. Yun, J.-S. Kim, W. Kim, C.-S. Lee, B.-G. Kim and H.-J. Jeong, Appl. Microbiol. Biotechnol., 2024, 108, 2.
8. H. Li, K. Hsieh, P. K. Wong, K. E. Mach, J. C. Liao and T.-H. Wang, Nat. Rev. Methods Primers, 2023, 3, 6.
9. L. Varadi, J. L. Luo, D. E. Hibbs, J. D. Perry, R. J. Anderson, S. Orenga and P. W. Groundwater, Chem. Soc. Rev., 2017, 46, 4818–4832.
10. B. Behera, G. K. Anil Vishnu, S. Chatterjee, V. V. Sitaramgupta, N. Sreekumar, A. Nagabhushan, N. Rajendran, B. H. Prathik and H. J. Pandya, Biosens. Bioelectron., 2019, 142, 111552.
11. J. Dietvorst, L. Vilaplana, N. Uria, M.-P. Marco and X. Muñoz-Berbel, TrAC, Trends Anal. Chem., 2020, 127, 115891.
12. I. Wiegand, K. Hilpert and R. E. Hancock, Nat. Protoc., 2008, 3, 163–175.
13. J. B. Patel, K. Alby, R. Humphries, M. Weinstein, J. D. Lutgring, S. N. Naccache and P. J. Simner, J. Clin. Microbiol., 2023, 61, e01154-22.
14. M. F. Chellat, L. Raguz and R. Riedl, Angew. Chem., Int. Ed., 2016, 55, 6600–6626.
15. V. I. Band, D. A. Hufnagel, S. Jaggavarapu, E. X. Sherman, J. E. Wozniak, S. W. Satola, M. M. Farley, J. T. Jacob, E. M. Burd and D. S. Weiss, Nat. Microbiol., 2019, 4, 1627–1635.
16. C. Murray, O. Adeyiga, K. Owsley and D. Di Carlo, Lab Chip, 2015, 15, 1226–1229.
17. R. Mohan, A. Mukherjee, S. E. Sevgen, C. Sanpitakseree, J. Lee, C. M. Schroeder and P. J. Kenis, Biosens. Bioelectron., 2013, 49, 118–125.
18. H. Nicoloff, K. Hjort, B. R. Levin and D. I. Andersson, Nat. Microbiol., 2019, 4, 504–514.
19. W. Postek, N. Pacocha and P. Garstecki, Lab Chip, 2022, 22, 3637–3662.
20. K. P. Kim, Y.-G. Kim, C.-H. Choi, H.-E. Kim, S.-H. Lee, W.-S. Chang and C.-S. Lee, Lab Chip, 2010, 10, 3296–3299.
21. C. Li, S. McCrone, J. W. Warrick, D. R. Andes, Z. Hite, C. F. Volk, W. E. Rose and D. J. Beebe, Lab Chip, 2023, 23, 2005–2015.
22. J. D. Besant, E. H. Sargent and S. O. Kelley, Lab Chip, 2015, 15, 2799–2807.
23. K. Mulroney, J. Hall, X. Huang, E. Turnbull, N. Bzdyl, A. Chakera, U. Naseer, E. Corea, M. Ellington and K. Hopkins, Sci. Rep., 2017, 7, 1903.
24. K. Chotinantakul, W. Suginta and A. Schulte, Anal. Chem., 2014, 86, 10315–10322.
25. D. Yang, Z. Yu, M. Zheng, W. Yang, Z. Liu, J. Zhou and L. Huang, Lab Chip, 2023, 23, 3961–3977.
26. W. Kang, S. Sarkar, Z. S. Lin, S. McKenney and T. Konry, Anal. Chem., 2019, 91, 6242–6249.
27. K. Churski, T. S. Kaminski, S. Jakiela, W. Kamysz, W. Baranska-Rybak, D. B. Weibel and P. Garstecki, Lab Chip, 2012, 12, 1629–1637.
28. J. Choi, J. Yoo, M. Lee, E.-G. Kim, J. S. Lee, S. Lee, S. Joo, S. H. Song, E.-C. Kim, J. C. Lee, H. C. Kim, Y.-G. Jung and S. Kwon, Sci. Transl. Med., 2014, 6, 267ra174.
29. W. Wu, G. Cai, Y. Liu, Y. Suo, B. Zhang, W. Jin, Y. Yu and Y. Mu, Lab Chip, 2023, 23, 2399–2410.
30. P. Juskova, S. Schmitt, A. Kling, D. G. Rackus, M. Held, A. Egli and P. S. Dittrich, ACS Sens., 2021, 6, 2202–2210.
31. Ö. Baltekin, A. Boucharin, E. Tano, D. I. Andersson and J. Elf, Proc. Natl. Acad. Sci. U. S. A., 2017, 114, 9170–9175.
32. I. Bennett, A. L. B. Pyne and R. A. McKendry, ACS Sens., 2020, 5, 3133–3139.
33. W. Postek and P. Garstecki, Acc. Chem. Res., 2022, 55, 605–615.
34. Y. Li, D. T. Burke, R. Kopelman and M. A. Burns, Biosensors, 2014, 4, 76–89.
35. Y. Yang, K. Gupta and K. L. Ekinci, Proc. Natl. Acad. Sci. U. S. A., 2020, 117, 10639–10644.
36. A. Bolotsky, R. Muralidharan, D. Butler, K. Root, W. Murray, Z. Liu and A. Ebrahimi, Biosens. Bioelectron., 2021, 172, 112615.
37. D. C. Spencer, T. F. Paton, K. T. Mulroney, T. J. Inglis, J. M. Sutton and H. Morgan, Nat. Commun., 2020, 11, 5328.
38. S. Hannah, E. Addington, D. Alcorn, W. Shu, P. A. Hoskisson and D. K. Corrigan, Biosens. Bioelectron., 2019, 145, 111696.
39. D. Spencer, Y. Li, Y. Zhu, J. M. Sutton and H. Morgan, ACS Sens., 2023, 8, 1101–1108.
40. H. Shi, Z. Hou, Y. Zhao, K. Nie, B. Dong, L. Chao, S. Shang, M. Long and Z. Liu, Chem. Eng. J., 2019, 359, 1327–1338.
41. M. Zhao, J. Yang, Z. Li, Y. Zeng, C. Tao, B. Dai, D. Zhang and Y. Yamaguchi, Lab Chip, 2024, 24, 2280–2286.
42. C. Heuer, J.-A. Preuss, M. Buttkewitz, T. Scheper, E. Segal and J. Bahnemann, Lab Chip, 2022, 22, 4950–4961.
43. A. M. Kaushik, K. Hsieh, L. Chen, D. J. Shin, J. C. Liao and T.-H. Wang, Biosens. Bioelectron., 2017, 97, 260–266.
44. S. Y. Oh and C.-S. Lee, Korean J. Chem. Eng., 2024, 41, 2449–2458.
45. X. Zhao, X. Yan, Y. Li and B.-F. Liu, Microfluid. Nanofluid., 2019, 23, 1–10.
46. M. E. O'Sullivan, Y. Song, R. Greenhouse, R. Lin, A. Perez, P. J. Atkinson, J. P. MacDonald, Z. Siddiqui, D. Lagasca and K. Comstock, Proc. Natl. Acad. Sci. U. S. A., 2020, 117, 32423–32432.
47. S. Herbert, A.-K. Ziebandt, K. Ohlsen, T. Schäfer, M. Hecker, D. Albrecht, R. Novick and F. Götz, Infect. Immun., 2010, 78, 2877–2889.
48. G. N. Wee, E. S. Lyou, J. K. Hong, J. H. No, S. B. Kim and T. K. Lee, Front. Cell. Infect. Microbiol., 2022, 12, 913415.
49. P. L. Irwin, L.-H. T. Nguyen, G. C. Paoli and C.-Y. Chen, BMC Microbiol., 2010, 10, 1–14.
50. A. Mirzaie, N. Peirovi, I. Akbarzadeh, M. Moghtaderi, F. Heidari, F. E. Yeganeh, H. Noorbazargan, S. Mirzazadeh and R. Bakhtiari, Bioorg. Chem., 2020, 103, 104231.
51. H. M. Blumberg, D. Rimland, D. J. Carroll, P. Terry and I. K. Wachsmuth, J. Infect. Dis., 1991, 163, 1279–1285.
52. Z.-X. Luo, Y. Li, M.-f. Liu and R. Zhao, Front. Microbiol., 2023, 14, 1328947.
53. M. A. Sánchez-Romero and J. Casadesús, Proc. Natl. Acad. Sci. U. S. A., 2014, 111, 355–360.
54. M. Ackermann, Nat. Rev. Microbiol., 2015, 13, 497–508.
55. H. Li, P. Torab, K. E. Mach, C. Surrette, M. R. England, D. W. Craft, N. J. Thomas, J. C. Liao, C. Puleo and P. K. Wong, Proc. Natl. Acad. Sci. U. S. A., 2019, 116, 10270–10279.
56. G. Longo, L. Alonso-Sarduy, L. M. Rio, A. Bizzini, A. Trampuz, J. Notz, G. Dietler and S. Kasas, Nat. Nanotechnol., 2013, 8, 522–526.
57. L. Mahler, M. Tovar, T. Weber, S. Brandes, M. M. Rudolph, J. Ehgartner, T. Mayr, M. T. Figge, M. Roth and E. Zang, RSC Adv., 2015, 5, 101871–101878.
58. S. Ki and D.-K. Kang, Biosensors, 2020, 10, 172.
59. H. H. Jeong, V. R. Yelleswarapu, S. Yadavali, D. Issadore and D. Lee, Lab Chip, 2015, 15, 4387–4392.
60. M. B. Romanowsky, A. R. Abate, A. Rotem, C. Holtze and D. A. Weitz, Lab Chip, 2012, 12, 802–807.
61. S. Jang, B. Lee, H.-H. Jeong, S. H. Jin, S. Jang, S. G. Kim, G. Y. Jung and C.-S. Lee, Lab Chip, 2016, 16, 1909–1916.
62. H.-H. Jeong, B. Lee, S. H. Jin, S.-G. Jeong and C.-S. Lee, Lab Chip, 2016, 16, 1698–1707.

---

Footnote

† Electronic supplementary information (ESI) available: Fig. S1 to S5 and Table S1. See DOI: https://doi.org/10.1039/d4lc00629a

---