A rapid antimicrobial susceptibility test based on single-cell morphological analysis

See allHide authors and affiliations

Science Translational Medicine  17 Dec 2014:
Vol. 6, Issue 267, pp. 267ra174
DOI: 10.1126/scitranslmed.3009650


A rapid antibiotic susceptibility test (AST) is desperately needed in clinical settings for fast and appropriate antibiotic administration. Traditional ASTs, which rely on cell culture, are not suitable for urgent cases of bacterial infection and antibiotic resistance owing to their relatively long test times. We describe a novel AST called single-cell morphological analysis (SCMA) that can determine antimicrobial susceptibility by automatically analyzing and categorizing morphological changes in single bacterial cells under various antimicrobial conditions. The SCMA was tested with four Clinical and Laboratory Standards Institute standard bacterial strains and 189 clinical samples, including extended-spectrum β-lactamase–positive Escherichia coli and Klebsiella pneumoniae, imipenem-resistant Pseudomonas aeruginosa, methicillin-resistant Staphylococcus aureus, and vancomycin-resistant Enterococci from hospitals. The results were compared with the gold standard broth microdilution test. The SCMA results were obtained in less than 4 hours, with 91.5% categorical agreement and 6.51% minor, 2.56% major, and 1.49% very major discrepancies. Thus, SCMA provides rapid and accurate antimicrobial susceptibility data that satisfy the recommended performance of the U.S. Food and Drug Administration.


Fast and accurate diagnosis of bacterial infection and subsequent administration of antimicrobials are necessary to prevent spread of infection and antibiotic resistance. Although antibiotic susceptibility tests (ASTs) are common clinical procedures that help select appropriate antibiotic treatments, these tests take too long to complete (usually 16 to 20 hours) (1). Because of the long lead times of ASTs, the initial antibiotic treatments are still mostly based on a physician’s guess (2). The widespread overuse and misuse of antibiotics coupled with the decreasing number of new U.S. Food and Drug Administration (FDA)–approved antibiotics could result in serious global health challenges (35). The dangerous methicillin-resistant Staphylococcus aureus (MRSA), once mainly found in hospitals and nursing homes, has become prevalent in households as well (6). Thus, it is important to reduce time for AST without compromising the accuracy of the tests (7, 8).

The main reason for the long lead times of ASTs is the limited sensitivity of the AST measurements (9). An AST determines the antimicrobial susceptibility of bacterial isolates from patient’s blood, wound specimens, or urine based on changes in the bacterial population in test wells when antimicrobials are added. Conventional AST systems indirectly sense the change in bacterial population by measuring the optical density (OD) of the antibiotic-dosed pathogen culture. If the bacterial population increases—indicating resistance to the antibiotic condition—the culture medium becomes opaque and the OD increases. Because the limit of detection for OD for a conventional AST is 107 colony-forming units (CFU)/ml (10), 16 to 20 hours must elapse for the bacterial population to reach the minimum detectable growth level. However, the only information that is needed to determine antibiotic susceptibility is whether the pathogen is dividing after the antibiotic is administered.

To overcome the limitations of conventional AST systems, researchers have developed various methods to observe bacterial division at early stages, in so-called rapid ASTs (1117). Rapid ASTs have been designed to measure, for example, the number of bacteria in a microfluidic channel (11, 1517); the rotation rate of magnetic beads, which is inversely proportional to bacterial mass (12, 13); or the fluorescent signal from the metabolic activity of bacteria cells in droplets (14). We previously reported that tracking single cell growth in microfluidic channel determined drug susceptibility by calculating the bacteria-occupying area in the images (18). These tests determine antimicrobial susceptibility based on the simple observation of whether the bacteria are growing. However, the antimicrobial responses of bacteria are very heterogeneous and specific to different antibiotic conditions (19, 20). We therefore hypothesized that a more accurate characterization of the “response” would include morphological changes.

Here, we propose and demonstrate the clinical application of a rapid AST with imaging-based single-cell morphological analysis (SCMA). SCMA can determine antimicrobial susceptibility by analyzing the morphological changes of single bacterial cells under various conditions. To adapt the SCMA to a high-throughput format, we developed a microfluidic chip that molds bacteria-mixed agarose to a thin, flat microscale slab. Thousands of morphological change patterns were acquired, performing time-lapse bright-field imaging of single cells using the microfluidic chip. From the results, we categorized the response of bacteria into several different morphological patterns. These patterns from our SCMA system were then correlated with the broth microdilution (BMD) test, which serves as the clinical gold standard. To eliminate human error in morphological analysis, we developed an automated image-processing and image-classifying algorithm that determines bacterial resistance to antibiotics.

Using the SCMA test, we determined the minimal inhibitory concentrations (MICs) of relevant antibiotics. Four different standard strains from the Clinical Laboratory Standard Institute (CLSI) and 189 clinical samples from two different hospitals were tested. The AST results were obtained in only 3 to 4 hours, thus meeting the FDA standards in clinical setting.


MAC integration with a 96-well format

Bacterial cells that are “resistant” to an antibiotic can divide in the presence of the drug. In contrast, “susceptible” bacterial cells are not able to divide because their growth is inhibited by bacteriostatic or bactericidal effects. Therefore, observing bacterial cell division is a fundamental method for determining bacterial resistance or susceptibility to antimicrobial agents. Turbidity, as determined by OD, is typically measured either by eye or by spectrophotometer to determine the bacterial cell division under antimicrobial conditions. Conventional clinical AST platforms typically use this OD method, where changes in OD reflect bacterial growth and, hence, drug resistance. However, owing to the high limit of detection of OD-based methods (≥107 CFU/ml), ASTs take 16 to 20 hours for sufficient bacterial growth (Fig. 1A).

Fig. 1. The rapid AST platform uses single bacterial cell morphology tracking in microfluidic agarose channels.

(A) Comparison of an AST based on SCMA with the conventional method using OD measurements. In principle, the SCMA method reduces the AST time compared with conventional OD-based AST. In conventional AST by OD measurement, the OD value does not change until the bacterial concentration reaches 107/ml, so OD measurement cannot be performed for 6 to 8 hours [data obtained from (10)]. However, in single-cell tracking using a microscope, changes in bacterial cells can be detected as soon as cells divide, so antibiotic susceptibility can be determined in 3 to 4 hours. (B) Schematic of the microfluidic agarose channel (MAC) chip integrated with a 96-well platform for high-throughput analysis. The MAC chip is composed of microfluidic channels containing bacteria in agarose, and a well to supply antibiotics and nutrients. The imaging region was the interface between the liquid medium and the microfluidic channel. The immobilized bacterial cells on bottoms of channels were monitored for SCMA by time-lapse bright-field microscopy. Detailed well dimensions and interfacing with 96-well plates are in fig. S1. (C) Experimental procedure for the MAC chip. Bacterial cells were mixed with agarose and loaded into a microfluidic channel, where the cells were immobilized by gel solidification. Liquid nutrients, some spiked with antimicrobials, were then loaded into the wells. These liquid samples diffuse into the agarose through openings between the channels and the wells. Time-lapse imaging was performed in the imaging region (yellow box).

To overcome this limitation, methods were developed to track the growth of single bacterial cells. However, it is difficult to apply such methods to ASTs because many types of motile bacteria cannot easily be observed and tracked. To observe and track single bacterial cells in clinic-ready AST systems, we needed to (i) immobilize bacteria, (ii) provide a stable supply of nutrients and antimicrobial agents, (iii) make imaging convenient and easy, and (iv) make the AST high-throughput. We designed a microfluidic chip that immobilizes bacterial cells in agarose, to facilitate image-based monitoring of single bacterial cell growth (18), that integrated it with a 96-well format for high-throughput testing (Fig. 1B and fig. S1).

In the microfluidic agarose channel (MAC) chip, nutrients and antimicrobial agents were supplied to the bacterial cells in the channel without any external equipment via diffusion (Fig. 1B) (2123). To perform a rapid AST in 3 hours, the diffusion of antimicrobials into the imaging region needs to be complete in 20 to 30 min, considering the division time of bacterial cells. The region of interest for imaging in the MAC chip was a square region (200 μm × 200 μm) of the interface between the agarose matrix and the antibiotic well. Approximate diffusion times of the antibiotic penicillin, and amino acids and proteins [lysozyme, formyl-norleucyl-leucyl-phenylalanine (FNLLP), bovine serum albumin (BSA)] into the imaging area were <5 min, indicating that the diffusion is sufficiently fast for AST, considering bacterial dividing times (~20 min) (table S1).

The bacterial cell samples were mixed with 1.5% liquid agarose at 37°C and then loaded into the well inlets to form MACs surrounding the liquid medium sample wells (Fig. 1C). As the agarose solidified in the channels at room temperature, the bacterial cells became immobilized. The liquid medium samples [Mueller Hinton broth (MHB) containing various antimicrobial agents at different concentrations] were then applied to the different wells. The agents and nutrients diffused from the antibiotic well toward the bacterial cells in the channels (characterized for Rhodamine B in fig. S2). About 30 combinations of antimicrobial agents (table S2) and test strains were loaded at the same time and subjected to SCMA by time-lapse imaging at 0, 2, and 4 hours for Gram-positive strains S. aureus and Enterococcus spp. and at 0, 1.5, and 3 hours for Gram-negative strains Escherichia coli, Pseudomonas aeruginosa, and Klebsiella pneumoniae.

The bacterial cells growing on the bottoms of the channels in the boundary areas between the liquid medium samples and the MACs were imaged by time-lapse method to monitor the change in bacterial morphology (Fig. 1, B and C). We took bacterial growth images of six different areas in the boundary region (1 mm × 1 mm) between the microfluidic channel and the well under or at MIC conditions. The pattern of bacterial growth did not exhibit any differences in these four areas in the vicinity of the well with the antibiotics; however, bacteria in regions farther from the antibiotics were unresponsive (fig. S3). We therefore chose the region closest to the antibiotic well to take images and obtain reliable AST results. We took one image (area 200 μm × 200 μm) per well for SCMA, with each field of view containing tens of bacterial cells.

Bacterial area measuring method incorrectly informing the MIC

We tested four clinically relevant (24, 25) and standard CLSI strains—E. coli ATCC (American Type Culture Collection) 25922, S. aureus ATCC 29213, P. aeruginosa ATCC 27853, and Enterococcus faecalis ATCC 29212—in our MAC chip with imaging. The antimicrobial agents to be tested on each strain were selected according to the lists of commercial AST kits from VITEK 2 systems (bioMérieux Inc.) and MicroScan WalkAway (Siemens Healthcare Diagnostics). Time-lapse images of bacterial cell growth were obtained to determine the MIC for the four strains. Bacterial imaging for AST was performed 3 hours after drug administration to Gram-negative strains and 4 hours after for Gram-positive strains, because the growth rates differ. From the time-lapse images, the MICs were first determined using a previously developed image-processing method called BAM [bacterial area measuring method (18)], which determined the antimicrobial susceptibility only by measuring the cell area in the images (11, 16, 17). If the bacterial area in the images increased, the case was determined as resistant, and if the bacterial area was static, the case was determined as susceptible.

In the case of gentamicin against P. aeruginosa ATCC 27853, bacterial area in the images increased with incubation time for concentrations less than 1.0 μg/ml. With concentrations ≥1.0 μg/ml, the bacterial area did not change. Therefore, 1.0 μg/ml was determined as the MIC value by the BAM method. The value agreed with the MIC from the BMD, and the MIC was within the CLSI quality control (QC) range [0.5 to 2 μg/ml (26)] (Fig. 2A).

Fig. 2. MIC determination for P. aeruginosa ATCC 27853 by analyzing bacterial number and size in response to antibiotics.

P. aeruginosa was exposed to three different antibiotics for 3 hours, and susceptibility was measured in the MAC with SCMA, or by standard BMD, and results were compared to the CLSI QC range. Images of the cells from the SCMA are shown on the left (scale bar, 25 μm). (A) Gentamicin. At concentrations lower than 1.0 μg/ml, bacterial cells divided. However, at concentrations equal to or higher than 1.0 μg/ml, the bacterial cells stopped dividing. MIC was determined as 1.0 μg/ml. A.U., arbitrary unit. (B) Ceftazidime. At concentrations equal to or higher than 2 μg/ml, ceftazidime induced filament formation (cell elongation), but the bacterial cells did not divide. The MIC was determined to be 2.0 μg/ml. (C) Imipenem. At concentrations lower than 1.0 μg/ml, the bacterial cells divided. At concentrations equal to or higher than 1.0 μg/ml, imipenem induced cell swelling, resulting in cell bursting, and bacterial cells did not divide, indicating it was the MIC. Additional morphological data are provided in table S4.

However, BAM did not lead to accurate MIC determination for other cases. For ceftazidime against the same strain of P. aeruginosa, bacterial area in the images increased at all concentrations, showing filamentary formations, and the MIC would have been considered >8.0 μg/ml by the BAM method. The MIC value was not matched to the BMD result (2 to 4 μg/ml) and was not within the CLSI QC range (1 to 4 μg/ml) (Fig. 2B). The mismatch of MIC values also occurred in the case of imipenem against the same bacterial strain, this time showing swelling (Fig. 2C).

We found that in the case of the β-lactam antimicrobial agent against Gram-negative strains, such as E. coli ATCC 25922 and P. aeruginosa ATCC 27853, the MIC was discordant with the BMD test because filamentary or swelling formation caused increase of the bacterial occupying area at or above the MIC value (movies S1 and S2). Swelling or filamentary formations were incorrectly classified as growth. It suggests that the BAM method is not able to determine the MIC and susceptibility in accordance with the BMD test. Thus, we designed a new AST method that could cope with heterogeneous bacterial responses to different antibiotics (19, 20).

SCMA for rapid and accurate AST

We propose and demonstrate here an SCMA as new image-based AST that meets the accuracy requirements of susceptibility testing by observing morphological changes of bacterial single cell under various antimicrobial conditions, including swelling and filament formation. Instead of applying the area and mass criteria of BAM, we classified antibiotic responses of bacteria based on cell mass and morphology. The correlation between SCMA and the BMD test results was investigated, and the MIC values from SCMA were within the CLSI QC range for various antimicrobials at different concentrations.

With our SCMA in the MAC chip, in the case of ceftazidime against P. aeruginosa ATCC 27853, the cell numbers increased, whereas the sizes were not changed at concentrations <2 μg/ml. At concentrations ≥2 μg/ml, there was no change in the cell number, but the cell sizes increased showing the filamentary formation, as described. Considering the BMD result (2 to 4 μg/ml) and the CLSI QC range (1 to 4 μg/ml), 2 μg/ml was determined to be the MIC value (Fig. 2B). In the case of imipenem against the same P. aeruginosa strain, the cell numbers increased, whereas the cell sizes did not change at concentrations <1 μg/ml. At concentrations ≥1 μg/ml, the cell numbers were not changed, but the cell sizes increased showing swelling formation. The determined MIC value was 1 μg/ml. The MIC value was slightly different from the BMD result (2 to 4 μg/ml), but the value was within the CLSI QC range (1 to 4 μg/ml) (Fig. 2C).

To determine the MIC values of the other antimicrobial agents, the morphological analysis was applied by considering the BMD results and the CLSI QC range (Table 1, fig. S4, and tables S3 and S4). The cases of Gram-positive strains (S. aureus ATCC 29213 and E. faecalis ATCC 29212) mainly consisted of dividing or nondividing cases without size change. However, the growth rates of bacteria in the various antimicrobial agents were different. When the growth rate of the strains differed, the MIC values were determined by comparing the relative growth rate in different concentrations of antimicrobial agents (Table 1, figs. S5 and S6, and tables S5 and S6).

Table 1. Accurate MIC determination for the four CLSI standard strains using SCMA.

For validation of the SCMA, four standard bacterial strains were tested against antimicrobial agents that are commonly used in clinical areas: (A) Gram-negative E. coli ATCC 25922 and P. aeruginosa ATCC 27853; (B) Gram-positive S. aureus ATCC 29213 and E. faecalis ATCC 29212. The MIC values were determined using SCMA after time-lapse imaging. The MIC values of SCMA were compared with the MIC ranges (QC ranges) provided by CLSI. Each test was performed in triplicate unless otherwise noted with an asterisk (*, performed twice).

View this table:

The discordant MIC values from the BAM method were reevaluated, and all of the MIC values were found to be within the QC ranges from the CLSI, except E. faecalis ATCC 29212 with erythromycin. In this case, the MIC value was lower than the CLSI MIC QC range. A lower MIC may reflect slow E. faecalis ATCC 29212 growth under erythromycin (table S6), making it difficult to observe the change in 4 hours. However, susceptibility was determined using the clinical strain of E. faecalis with erythromycin (supplementary file SummaryDataAST.xlsx).

Evaluating cell morphology using SCMA

Bacterial cell morphologies were categorized into six types for SCMA: dividing, no change, filamentary formation, swelling, coexistence of filamentary formation and dividing, or swelling formation and dividing (Fig. 3). The typical dividing pattern in the antibiotic-free and the antibiotic-resistant conditions involved single bacterium dividing into two cells, which increased both the number of bacterial cells and the OD value of the BMD test (Fig. 1A and Fig. 3A). The typical pattern in the susceptible condition involved no change in the bacterial cells (Fig. 3B).

Fig. 3. Morphological categorization of single cells against antibiotics.

After time-lapse imaging of the single bacterial cells, their growth patterns against the antibiotics were analyzed and classified into four groups. (A) Typical division in the antibiotic-free or the antibiotic-resistant conditions, in which bacterial cells divide into two cells. (B) Typical antibiotic-susceptible conditions, in which bacteria do not grow. (C) Filamentary formation conditions for Gram-negative strains in response to β-lactam antibiotics, in which the bacterial cells show filamentary growth, but do not divide. (D) Swelling formation conditions for Gram-negative strains in response to imipenem and meropenem, in which the bacterial cells are swollen but do not divide. Cases in (C) and (D) were considered to be susceptible. (E) Coexistence of filamentary formation and dividing. (F) Coexistence of swelling formation and dividing. Cases in (E) and (F) were considered as resistant. *For Gram-positive strains in (A) and (B), the time-lapse images were taken at 0, 2, and 4 hours compared with 0, 1.5, and 3 hours for Gram-negative strains in (C) to (F). Scale bar, 20 μm.

Filamentary formation occurred for Gram-negative strains in response to most β-lactam antimicrobial agents, with the exception of the penem class of drugs (Fig. 3C and fig. S4B). These bacterial cells typically showed filamentary formation but no division. Swelling patterns were observed for Gram-negative strains in response to the penem class of drugs (imipenem and meropenem) (Fig. 3D and fig. S4C). These bacterial cells were swollen but did not divide. In the illustrated cases of filamentary and swelling formations (Fig. 3, C and D), we chose to determine the bacteria to be susceptible to the antibiotics because there was no OD change. This is because no OD change in the conventional BMD test would also give a read-out of susceptible. For cases in which we observed the coexistence of filamentary formation and dividing or swelling formation and dividing, we determined that the bacteria were resistant (Fig. 3, E and F).

Automated susceptibility determination based on SCMA

Determination of bacterial susceptibility to antimicrobials based on morphological assessment by a human examiner could be subject to human error. We therefore developed an automated image-processing and classification program to automatically determine susceptibility of various bacterial strains against different antimicrobial agents. Figure 4 shows the schematic representation of our automated analysis. To determine the six morphological patterns in Fig. 3, the total areal occupancy of cells (Ai), the number of bacterial cells (Fi), and the total length of bacteria cells (Li) were automatically evaluated from each time-lapsed image. Details of the image processing, such as digital filtering, binary conversion, and line detection, are explained in figs. S7 and S8 and table S7.

Fig. 4. Automated image-processing and data interpretation.

(A) In the case of Gram-positive strains, the data of total area occupancy of cells (Ai) were obtained from the processed images. The value of bacterial occupancy area in third and final image at 4 hours (A3) was divided by those of first image at 0 hours (A1), and the calculated value was compared with the first threshold value (T1). If A3/A1 was larger than T1, the case was determined as resistant. If the value was smaller than T1, the second thresholding processes comparing with the second (T2) and third (T3) thresholds were performed to determine susceptibility. Two calculated values from A3/A1 and A3/A2 (A2, the second image at 2 hours) were compared with T2 and T3, respectively. If both values were larger than T2 and T3, respectively, then the case was determined as resistant. If not, it was susceptible. (B) In the cases of Gram-negative strains with filamentary formation, all cases with increased area of bacterial cells compared with the threshold values (T4, T5, and T6) were sent to filament checks. Total length of bacterial cells (L3) was divided by the number of bacterial cells (F3), and the calculated value was compared with T7. If the value was larger than T7—implying filamentary formation—the case was determined as susceptible; if not, it was regarded as resistant, implying cell division. A1–3 are the values of bacterial growth area according to incubation times. T1–7 are the threshold values determined individually for each strain with different growth rates under various antimicrobial agent conditions. Scale bar, 20 μm.

In the cases of Gram-positive bacteria, there are only two morphological patterns: the dividing case (resistant) and the “no change” case (susceptible). For no-change cases, the change in area only needed to be examined. When area increased to a certain threshold rate, it was determined to be resistant (Fig. 4A). In the cases of Gram-negative bacteria, there are six morphological patterns: the dividing case (resistant), no-change case (susceptible), filamentary formation (susceptible), swelling formation (susceptible), coexistence of filamentary formation and dividing (resistant), or swelling formation and dividing (resistant). For Gram-negative bacteria, all parameters—area, number of cells, and length of cells—needed to be evaluated (Fig. 4B). Although area may have been increasing, it may have been due to an increase in length; therefore, the case would not be resistant but susceptible.

To determine the threshold values, we examined each raw image from the CCD (charge-coupled device) camera and suggested the thresholds (T1-T6) for the ratio between the areas of bacteria (A) and the ratio between the length (L) and number (F) of filament-shaped bacteria (T7) to match the susceptibility results from human inspection (Fig. 4). The proposed threshold values were refined by comparing the results from human inspection with the results from our image-processing program.

SCMA for determination of antimicrobial susceptibility in clinical samples

We tested 189 clinical isolates (42 E. coli, 34 P. aeruginosa, 30 K. pneumoniae, 45 S. aureus, and 38 Enterococcus spp.) (Fig. 5A). Clinical strains from various specimens (blood cultures, wound specimens, and urine) were collected from the laboratories of Seoul National University Hospital (SNUH; 149 strains) and Incheon St. Mary’s Hospital (ISMH; 40 strains). The collection included various phenotypes with special resistance mechanisms: 37 extended-spectrum β-lactamase (ESBL)–producing Enterobacteriaceae spp., 17 imipenem-resistant P. aeruginosa (IRPA), 24 MRSA, and 16 vancomycin-resistant Enterococci (VRE). The clinical bacterial samples were prepared on either sheep blood agar or Mueller Hinton agar (MHA) plates. Before testing, each isolate was subcultured on cation-adjusted MHA for 20 to 24 hours. Our ASTs using SCMA were performed in the MACs simultaneously with the gold standard BMD test as reference AST.

Fig. 5. Discrepancy rates and CA rates for SCMA using clinical samples.

(A) Distribution of the 189 clinical samples from SNUH and ISMH. (B and C) The SCMA AST results were compared with the clinical gold standard BMD test. Discrepancy rate and CA rates were calculated. (B) Summary of the discrepancy and CA rates. Total values were calculated with all AST results from Gram-positive and Gram-negative cases. For BMD test: S, susceptible; I, intermediate; R, resistant. For the SCMA: mE, minor error; ME, major error; VME, very major error. TTR, time to results. (C) CA rates according to the different clinical strains in response to β-lactam and non–β-lactam antimicrobial agents. Error bars represent 95% confidence intervals determined by Wilson’s binomial method. P values were determined using a χ2 test.

To produce discrepancy rates, the AST results from SCMA were compared with the results from the BMD test (Fig. 5B). To examine the accuracy of our clinical interpretations, category agreement, minor errors (mEs), major errors (MEs), and very major errors (VMEs) were calculated, as defined by the FDA guidance document for each organism (27). Category agreement is the agreement of interpretive results (SIR) between a new device under evaluation (SCMA) and a standard reference method (BMD test). The total category agreement for all isolates was 91.5%, and the mE rates, ME rates, and VME rates were 6.51, 2.56, and 1.49%, respectively, for SCMA (Fig. 5B). (All data are in supplementary file SummaryDataAST.xlsx.) The SCMA AST took only 3 to 4 hours and required only 10 to 20 bacteria in a single image (200 μm × 200 μm) to produce AST results, and the error and categorical agreement (CA) rates satisfied the FDA requirements for AST systems (mE ≤ 10%, ME ≤ 3.0%, VME ≤ 1.5%, and CA ≥ 90%).

Overall, the AST results for Gram-positive strain indicated higher category agreement rates and lower error rates compared to Gram-negative strains (Fig. 5B). In the Gram-negative strains, β-lactam antimicrobial agents produced lower category agreement rates than did non–β-lactam antimicrobial agents. Reduced category agreement with BMD has been reported in the other AST systems (for example, VITEK 2 and MicroScan) because different inoculum concentrations and incubation times may influence the detection of resistance and the β-lactamase production (Fig. 5C) (28). For P. aeruginosa, the lowest category agreement rate was for β-lactam antimicrobial treatment. In the Gram-positive strains, there were no significant differences between β-lactam and non–β-lactam antimicrobial agents because there were no morphological differences between them.

SCMA reducing AST error rates compared with the BAM method

SCMA reduced error rates compared with BAM (18) for both β-lactam and non–β-lactam antibiotics against Gram-negative strains E. coli, P. aeruginosa, and K. pneumoniae (Fig. 6). For β-lactam antimicrobial agents, the ME rates were markedly reduced from 12.8 to 0.9% for E. coli and from 48.1 to 13.7% for P. aeruginosa. The mE rates were slightly reduced with SCMA from 10.3 to 9.7% for E. coli and from 6.7 to 4.1% for K. pneumoniae. Filamentary responses of Gram-negative strains to β-lactam antimicrobials were determined resistant by BAM, which only considers the change in bacterial area. However, the cases were deemed susceptible by the BMD test, resulting in a high ME rate. Using our SCMA, the filament cases were determined to be susceptible, in agreement with the BMD test, thus reducing the ME rates.

Fig. 6. SCMA reduces AST error rates compared with BAM.

Data are the AST error rates for clinical Gram-negative strains E. coli (n = 42), P. aeruginosa (n = 34), and K. pneumoniae (n = 30) tested with non–β-lactam and β-lactam antimicrobial agents (supplementary file SummaryDataAST.pdf). The error rates were calculated by comparing AST results from each method with BMD test. P values were determined using a χ2 test.

In 1.2% of cases, there was filamentary growth in SCMA (deemed susceptible) with increased OD values in the BMD test (deemed resistant) causing VME: 2 VME cases of 390 in E. coli, and 6 VME cases of 253 in P. aeruginosa. This could have been caused by the limited imaging area (200 μm × 200 μm). In these cases, we speculate that there was only filamentary formation or swelling formation in the imaged area, whereas cell division took place in the other areas.


Here, we assessed the response of four clinical pathogens representing various bacterial infections, including major antibiotic-resistant pathogens (ESBL-positive E. coli and K. pneumoniae, VRE, and MRSA), to clinically relevant antibiotics. The MAC system using SCMA produced accurate AST results in only 3 to 4 hours. The results were compared with the conventional AST using BMD, which takes ~66 hours in total, including the blood culture. In rapid AST with our SCMA system, the total testing time can be reduced to 52 hours, about 25% less than conventional methods, because it does not require relatively long incubation time. Reduced testing times will have a major impact on the care and outcome of hospitalized patients with bacterial infections (7).

The morphological changes of Gram-negative bacteria in response to β-lactam antibiotics have been studied previously (19, 20). Penicillin-binding proteins polymerize and modify peptidoglycans (the stress-bearing components of bacterial cell walls), leading to morphological deformations, including filamentary formation and swelling. The relationship between filamentary formation and MIC was observed by Buijs et al. (29), but they could not determine a MIC value because filamentary formation occurred over a broad range of antimicrobial concentrations that were variably below, equal to, and above the MIC. In our study, we used automated SCMA to objectively examine morphological changes for MIC determination. Cases of filamentary formation or swelling were regarded as susceptible because the OD value did not increase. In some cases, morphological deformation and division matched. These cases were regarded as resistant because some resistant bacterial cells dominated to increase the OD (Fig. 3).

Hospital ASTs such as VITEK 2 and MicroScan can test about 60 combinations of antimicrobial agents at different concentrations. We integrated microfluidic channels with 96-well plates to exclude syringes and tubing systems. In our system, nutrient and antibiotic solution is provided directly from well into the microfluidic agarose channel where bacterial cells are immobilized. However, an auto-pipetting system is needed to load bacterial cells and nutrient (antibiotic) solution to meet these high-throughput clinical requirements. In addition, imaging data acquisition of SCMA took 20 to 30 min per one test sample, which is slower than conventional ASTs based on the turbidity measurement, which takes less than 1 min (VITEK 2 and MicroScan). Bright-field imaging of an entire 96-well MAC chip with a 60× objective lens will take too much time to be clinically amenable. Instead, we performed single-spot 60× imaging per well (field of view, 200 × 200 μm2), which was sufficient to obtain a bacterial susceptibility result. Imaging 60 different combinations in the MAC chip required 60 different bright-field images, but because the imaging location is already known and predetermined by microfluidic chip design, it took about 20 min in manual setting and would take less than 5 min in automated system. To reduce the time for imaging, finding proper z-location (focusing) is important. We have implemented a high-resolution focusing mark in our microfluidic chip (fig. S1B).

The MAC system with SCMA can be used to observe morphological reactions against antimicrobial agents with single-cell resolution, whereas conventional AST systems and many rapid AST systems can only be used to observe bacterial population behavior (susceptible or resistant) (11, 30, 31). This suggests that the MAC system can provide more information for clinical pathologists and researchers on antibiotic resistance mechanisms. Although those mechanisms were not investigated here, they may help improve our understanding of the mode of action and resistance of antibiotics.

Bacterial identification with AST result is necessary for accurate prescription of antimicrobials to the bacteria-infected patients. In commercialized AST platforms, AST is performed simultaneously with the bacterial identification assay. Our SCMA with the MAC will also require bacterial identification. The test time required for bacterial identification is generally shorter than the time for AST. Owing to the fast identification capabilities of mass spectrometry (32), there are some commercial AST platforms that have integrated with this analytical technique, such as the VITEK MS. The MAC system with SCMA could be similarly integrated with mass spectrometry–based platforms for bacterial identification.

Here, the five clinically important strains—E. coli, P. aeruginosa, K. pneumoniae, S. aureus, and Enterococcus spp.—were tested, and their morphological patterns under various antibiotics were analyzed to establish criteria for SCMA. We trained our SCMA algorithm to match with the BMD method—the current clinical standard for AST—to meet FDA requirements for accuracy. However, for other clinical strains under certain antibiotic conditions, other morphological patterns may arise; our current SCMA criteria may not be applicable to other patterns. To apply SCMA in clinical settings, more bacterial strains will need to be tested and their morphological patterns included in the SCMA criteria.

In conclusion, the SCMA can determine bacterial susceptibility antimicrobial drugs faster than conventional AST and the gold standard BMD, but with the same accuracy. This technology was validated with clinical samples (MRSA, VRE, IRPA, and ESBL-positive E. coli and K. pneumoniae), suggesting that it will be suitable for broad clinical use. For clinical translation of our rapid AST with SCMA, full automated system with sample preparation, image acquisition, and analysis will be needed. The additional information on individual cell morphology provided by the SCMA could in the future lend insight into bacterial responses to antibiotics as well as resistance mechanisms.


Study design

The objective of this study was to develop a rapid AST using single-cell imaging analysis. Our hypothesis was that the time required to perform AST could be reduced to a few hours using single-cell tracking and a morphological analysis of bacteria. To prove the hypothesis, four standard strains from CLSI were tested three times to determine their morphological patterns under various antimicrobial conditions. The antimicrobials were selected on the basis of the antimicrobial compositions of conventional AST systems, such as the VITEK 2 systems and MicroScan WalkAway. To obtain morphological patterns for a single cell, we coupled a 60× microscope lens with a high-resolution CCD camera. The image data were processed automatically and compared with BMD results. For further validation of SCMA, clinical strains (n = 189; 149 from SNUH, 40 from ISMH) composed of E. coli (n = 42), P. aeruginosa (n = 34), K. pneumoniae (n = 30), S. aureus (n = 45), and Enterococcus spp. (n = 38) were tested. For a blinded assessment of the outcomes, the MIC results from the SCMA method were determined automatically using an imaging-processing program without knowledge of the results from the BMD test.

Fabrication of the MAC chip

The MAC chip (96-well format) was designed using 3D CAD design software (SolidWorks v2010, Dassault Systèmes SolidWorks Corp.) and fabricated by a commercial injection molding company in the Republic of Korea, R&D Factory that used injection molding (Selex NT-2 130) of polycarbonate (SABIC Innovative Plastics). The bottom film (150 μm, SEJIN TS) of poly(methyl methacrylate) (PMMA) was bonded using a solvent process [80:20 (w/w), ethanol:1,2-dichloroethane; Sigma-Aldrich] under pressure at 35°C. Before the AST, a 1-min O2 plasma treatment (CUTE-MP, Femto Science) was used to make the chip hydrophilic.

Bacterial strains

The four CLSI standard strains were purchased from MicroBioLogics Inc. (E. coli ATCC 25922, S. aureus ATCC 29213, P. aeruginosa ATCC 27853, and E. faecalis ATCC 29212). One hundred forty-nine strains were isolated from 149 patients at SNUH. Forty strains were obtained from 40 patients at ISMH. Pure cultures of the clinical strains from blood agar plate and MHA (BBL, BD Biosciences) were inoculated on MHA plates and incubated for 20 to 24 hours. After incubation, several colonies were used to prepare bacterial stocks at concentrations of 1.5 × 108 CFU/ml. Bacterial identification was performed according to the protocols of the hospital where the strain originated: Gram-negative identified by VITEK 2 Systems and Gram-positive by MicroScan at SNUH; both Gram-negative and Gram-positive strains were identified by VITEK 2 Systems at ISMH.

Antimicrobial preparations

The antimicrobial agents (table S2) were purchased from Sigma-Aldrich and Santa Cruz Biotechnology Inc. Stock solutions were prepared using the method from the supplier and the CLSI guidelines (33). The stock solutions were stored at −70°C, and before the AST, the solutions were thawed to room temperature and diluted in cation-adjusted MHB (BBL, BD Biosciences). In the case of ceftazidime, the antimicrobial solutions were prepared directly from powder form following the manufacturer’s instructions.

Broth microdilution test

The BMD test was used as a gold standard recommend by the CLSI (33). For the BMD test, the antimicrobial solutions were prepared from the stock solution or directly from powder form (ceftazidime). A 100-μl volume of the antimicrobial agents at the appropriate concentration, which was determined by CLSI recommendation (26), was pipetted into the bottom of 96 MicroWell plates (Falcon, BD Biosciences), and 10 μl of bacterial stock solution was inoculated in the well at a final concentration of 5 × 105 CFU/ml. The BMD tests were performed in triplicate. After 16 to 20 hours of incubation at 37°C, the MIC values of the microdilution wells were read as the concentration in which there was ≥80% reduction in growth as compared to the control by unaided visual inspection. If the results from triplicate tests were not identical, the majority result was selected as the MIC.

Single-cell tracking

Single cells were monitored in the MACs by an S Plan Fluor ELWD 60× (numerical aperture 1.49) lens on an inverted optical microscope (Eclipse Ti, Nikon; IX71, Olympus) integrated with a heating system. Bright-field micrographs were obtained by using an electron-multiplying CCD camera (QuantEM:512SC, Photometrics for Eclipse Ti). For the first image of the MAC, the imaging area was the vicinity of the boundary between the agarose and the antimicrobial well. The imaging area was 200 μm × 200 μm. Time-lapse images were acquired at the same area.

Image processing and data acquisition

The automated image analysis program was coded with MATLAB R2013a (MathWorks). The image analysis process is described in figs. S7 and S8, table S7, and Supplementary Materials and Methods (34, 35).

Statistical analysis

Wilson’s binomial method was used to calculate the 95% confidence intervals (36). The 95% confidence intervals for the proportion of CA, including the mE, ME, and VME between the SCMA, VITEK 2 Systems, and MicroScan and gold standard BMD method, were also calculated. A χ2 test was used to calculate the P values. All statistical analyses were performed with SAS (version 9.3).


Materials and Methods

Fig. S1. MAC chip.

Fig. S2. Diffusion characteristics of Rhodamine B in the imaging area of the MAC chip.

Fig. S3. Time-lapse images of bacteria at the different locations in the MAC chip.

Fig. S4. MIC determination for E. coli ATCC 25922 by analyzing bacterial number and size in response to antibiotics.

Fig. S5. MIC determination for S. aureus ATCC 29213 by analyzing bacterial number in response to antibiotics.

Fig. S6. MIC determination for E. faecalis ATCC 29212 by analyzing bacterial number in response to antibiotics.

Fig. S7. Automated image processing for representative Gram-positive strains.

Fig. S8. Automated image processing for representative Gram-negative strains.

Table S1. Diffusion times into the imaging region.

Table S2. Molecular weight of the antimicrobial agents tested in the clinical AST.

Table S3. Morphological characteristics of E. coli ATCC 25922 in response to different antimicrobial agents at concentration equal to or higher than the MIC.

Table S4. Morphological characteristics of P. aeruginosa ATCC 27853 for different antimicrobial agents.

Table S5. Morphological characteristics of S. aureus ATCC 29213 for different antimicrobial agents.

Table S6. Morphological characteristics of E. faecalis ATCC 29212 for different antimicrobial agents.

Table S7. Explanation of image-processing algorithm.

Movie S1. The Gram-negative P. aeruginosa ATCC 27853 divided at concentrations lower than the MIC of aztreonam.

Movie S2. The Gram-negative P. aeruginosa ATCC 27853 formed filaments at concentrations of β-lactam antimicrobial greater than the MIC.

Supplementary file [SummaryDataAST.xlsx]. Summary of antimicrobial susceptibility test results from SCMA and the commercial automated systems (VITEK 2 and MicroScan) using clinical strains of SNUH and ISMH.


  1. Funding: Korean Health Technology R&D Project, Ministry of Health & Welfare, Republic of Korea (grant numbers HI13C1468 and HI13C0866), the Institute for Basic Science in Korea, the Pioneer Research Center Program through the National Research Foundation (NRF) of Korea funded by the Ministry of Science, ICT & Future Planning (NRF-2012-0009555), and NRF grant funded by the Korean Government (2012M3A9B2030170). Author contributions: J.C., Y.-G.J., and S.K. contributed to the concept and design of the study and writing of the paper. J.C., J.Y., M.L., Y.-G.J., and S.K. contributed to the design and performance of experiments and analysis of data. J.C. performed the statistical analysis. E.-G.K. fabricated the MAC chip. S.L., S.J., S.H.S., and E.-C.K. provided general advice and contributed to analysis of data. J.S.L., J.C.L., and H.C.K. developed the image-processing algorithm. Competing interests: E.-G.K., J.Y., Y.-G.J., and S.K. at the submission of the paper were employed at QuantaMatrix Inc., which is commercializing the SCMA technology. E.-G.K., J.Y., Y.-G.J., and S.K. have equity interest in QuantaMatrix Inc. Data and materials availability: Clinical samples are available via a material transfer agreement.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article