Article pubs.acs.org/ac
Counting Bacteria Using Functionalized Gold Nanoparticles as the Light-Scattering Reporter Xiao Xu,† Yang Chen,† Hejia Wei,‡,§ Bin Xia,†,‡,§ Feng Liu,† and Na Li*,† †
Beijing National Laboratory for Molecular Sciences (BNLMS), Key Laboratory of Bioorganic Chemistry and Molecular Engineering of Ministry of Education, Institute of Analytical Chemistry, College of Chemistry and Molecular Engineering, Peking University, Beijing, 100871, China ‡ Beijing NMR Center, Peking University, Beijing 100871, China § College of Life Sciences, Peking University, Beijing 100871, China S Supporting Information *
ABSTRACT: A simple and rapid bacterial counting method was developed based on dark-field light-scattering imaging of bacteria and gold nanoparticle (AuNP) reporter simultaneously. Commercially available DH5α E. coli strain was used as the model bacterium to demonstrate the feasibility of the proposed method. With antibody-conjugated AuNPs, the simple sample treatment and target E. coli strain recognition can be finished within 15−30 min, with a detection limit of 2− 6 × 104 colony forming unit per milliliter (CFU/mL). By using 90 nm AuNPs as the light-scattering signal reporter, the bacterial recognition and counting can be easily performed with low-cost instrumentation such as an entry-level dark-field microscope setup and a common tungsten lamp as the light source. An automatic image analysis algorithm was also developed to facilitate robust and fast bacterial counting. The preliminary results of water, milk, and fruit juice sample analysis showed that this simple, fast, and cost-effective method can be easily adopted for routine bacterial detection.
T
Furthermore, current methods of E. coli assay are usually performed in bulk solution, using the calibration curve for quantification.19 For optical assay methods, the variation of shape and size of the same type of bacterium may result in fluctuation of measured optical signal (fluorescence or scattered light) of one bacterium, thus, signals from a large number of bacteria are required to achieve an acceptable signal-to-noise ratio, which is the major obstacle to improve the detection limit for calibration curve based methods. Bacterial counting method provides an approach for highly sensitive detection at the single-cell level20 and is fundamental to establishing such a method by identifying an adequate optical probe. Fluorophores such as organic molecules21,22 and quantum dots23,24 are good optical probes, but the sensitivity is worse than required for single-cell detection. Current bacterial counting methods, such as flow cytometry and fluorescence imaging, usually employ laser or high power white light as the light source,21,22 which significantly increases the cost of instrumentation. Another drawback of fluorescence methods is that only the signal from fluorophores can be obtained in the measurements and information about nonfluorescent target cells or tissues is unavailable. Although bright-field imaging, as complementary to fluorescence measurements, can be used to obtain the image of the cell and tissue, the sample must stay still
he detection of foodborne pathogens is of utmost importance for food safety and clinical diagnosis. Among the foodborne pathogens (bacterial, viral, and parasitic pathogens), enterohemorrhagic E. coli (EHEC) can produce verotoxins or Shiga-like toxins and cause symptoms such as hemorrhagic colitis and hemolytic-uraemic syndrome,1 with the infection dose of 0.3−10 CFU/g viable bacteria.2,3 The U.S. EPA regulation requires that no viable pathogens are allowed in certain foods, such as drinking water,4,5 thus, rapid, sensitive, and low-cost bacterial detection methods are urgently needed for food safety and environmental monitoring. Microbiological,6,7 immunological,8−11 polymerase chainreaction (PCR),12,13 and biosensing methods14−18 have been used for the detection of bacteria. Microbiological methods are time-consuming and usually take several days. Most immunoassays are rapid, with detection time in minutes to hours, but suffered from low sensitivity, with the limit of detection from 103 to 106 cells or CFU per mL, as well as the inability to determine infectivity of the organism. PCR and its modified methods are sensitive and need only a few hours to complete; however, sample preparation and enzymatic amplification of the target nucleic acid sequence are tedious.4 Furthermore, these techniques detect the component of the pathogen rather than the intact cell, thus, correlation with the number of cells in the sample is difficult.5 The current biosensor-based methods, using a variety of biorecognition moieties and signal transduction methodologies, can achieve a short assay time, but the limit of detection is typically greater than the effective dose. © 2012 American Chemical Society
Received: April 17, 2012 Accepted: October 4, 2012 Published: October 4, 2012 9721
dx.doi.org/10.1021/ac302471c | Anal. Chem. 2012, 84, 9721−9728
Analytical Chemistry
Article
between fluorescence measurements and bright-field imaging, which make flow analysis difficult. Dark-field imaging, as a highly sensitive method, can be used to acquire both the optical signal and the image of the test sample at the same time.25 As a common optical phenomenon, light-scattering occurs with most objects. With high efficiency light-scattering probes, signals of both bacteria and the optical probe can be obtained simultaneously using a common tungsten lamp with a simple microscope setup. The cost of instrumentation is much lower and has the same quality of recognition performance as the fluorescence imaging. Gold nanoparticles (AuNPs) are excellent light-scattering probes, due to the featured localized surface plasmon resonance (LSPR) phenomenon,26 and have been used for biosening purposes.27,28 According to Mie theory, the intensity of scattered light of an AuNP is proportional to the fourth power of the diameter of the particle.29,30 The light-scattering intensity of an AuNP with the diameter of 60 nm equals the fluorescence intensity of 2.7 × 105 fluorescein molecules.29 AuNPs with the diameter larger than 40 nm are visible under a simple dark-field microscopic setup with 20 W tungsten lamp irradiation. Therefore, recognition moiety-functionalized AuNPs would be suitable for recognition and dark-field imaging of bacteria. In this article, we have demonstrated a rapid bacterial recognition and counting method based on dark-field imaging of bacteria and the antibody-functionalized AuNP reporter simultaneously. The whole detection procedure of E. coli in liquid samples, from sample pretreatment to microscopic imaging, can be finished in 15−30 min. An automatic image analysis algorithm was also developed to explore the feasibility of fast bacterial counting. Our preliminary results with practical sample analysis showed that this method has the potential for bacterial detection in drinking water, milk, and fruit juice.
microscope with an Olympus DP-72 true color CCD (Japan). For quantitative detection, the DRM-700 CELL-VU CBC hemacytometer was used as the sample carrier. Otherwise, glass slide and coverslip were used for convenience. Dark-Field Imaging Based Bacterial Detection and Counting. The bacterial recognition and counting were based on the shape, color, and spatial information of bacteria and the attached, antibody-functionalized AuNPs were retrieved from the dark-field images. With the shape and color information, an object in the image can be identified as a bacterium, an AuNP, or interference entity. The spatial information can then be used to identify the AuNPs attached to the bacteria. Based on the selective recognition between the antibody-functionalized AuNPs and antigen on the surface of the bacteria, a bacterium can be distinguished as the target when AuNPs are attached to the bacterial surface. The bacterial counting can be performed manually as well as automatically. Preparation of AuNPs. AuNPs were prepared by a twostep, seed-mediated growth method.31 About 13 nm diameter gold seeds with a final concentration of 3 nmol/L were synthesized by reducing HAuCl4 with sodium citrate. Then 0.05 mL of gold seed colloid was added to 2.4 mL of 0.01% (wt) HAuCl4 solution, and 0.1 mL of 40 mmol/L NH2OH·HCl solution was added subsequently. The mixture was stirred for 1 min and 0.1 mL of 1 wt % HAuCl4 solution was added. With stirring, the reaction was complete in 10 min to produce a brownish red colored mixture. The AuNP average size was 91 ± 10 nm (Figure S-1), as characterized with a Hitachi H9000 transmission electron microscope. Functionalization of AuNPs. Functionalization of AuNPs was carried out at room temperature. The strategy and steps involved in functionalization of AuNPs are shown in Scheme S1. Gold colloid (2 mL) was mixed with 2 mL of 2% Tween-20 in PBS buffer and stirred for 30 min. Then 2 mL of 11-MUDA solution (5 mg 11-MUDA in 40 mL alcohol) was added with gentle stirring. The mixture was incubated in the dark for 4 h at 100 rpm and then centrifuged at 9000 rpm for 5 min, followed by washing 5× with 2 mL of purified water to remove the excess Tween-20 and 11-MUDA. The 11-MUDA-capped AuNPs were redispersed in water to a volume of 2 mL, and 0.067 mL of 1 mmol/L EDC/NHS solution was quickly added. The mixture was kept in the dark for 30 min to activate the carboxyl group of 11-MUDA. Antibody solution (0.5 mL) was added, and the resulting solution was stored in the dark for 4 h to complete the reaction. The mixture was then centrifuged at 9000 rpm for 5 min, and washed 3× with 2 mL of purified water to remove the excess EDC/NHS and antibody. The antibody-conjugated AuNPs were redispersed in 2 mL of PBS buffer to a final concentration of 0.082 nmol/L. Reaction of Antibody-Functionalized AuNPs with E. coli and Acquisition of Dark-Field Images. E. coli were incubated in Lysogeny broth (LB) media at 37 °C overnight, and then the E. coli suspension was diluted to yield an OD600 value about 0.2−0.3 with a Hitachi U-3010 spectrophotometer. Then 0.5 mL of diluted E. coli suspension was mixed with 1 mL of antibody-modified AuNPs. The mixture was incubated at room temperature for 15 min. Then 10 μL of each mixture was placed on a microscope slide and covered by a coverslip. All the samples were observed under the Olympus BX53 microscope with a 40× objective (dry lens). For best spatial resolution, all dark-field images were acquired using “pixel-shifting” mode of Olympus cellSens software, with a resolution of 4140 × 3096
■
EXPERIMENTAL SECTION Reagents and Instrumentation. 1-Ethyl-3-[3dimethylaminopropyl]carbodiimide hydrochloride (EDC, 99%), auric chloride dehydrate (A. R.), and Tween-20 (C. P.) were obtained from Sinopharm Chemical Reagent Co., Ltd.. Sodium citrate, hydroxylamine hydrochloride, absolute ethanol, potassium dihydrogen phosphate, disodium hydrogen phosphate, sodium chloride, and potassium chloride, all of A. R. grade, were obtained from Beijing Chemical Works. N-Hydroxy succinimide (NHS, 97%) and 11-mercaptoundecanoic acid (11-MUDA, 95%) were obtained from J&K Chemical Ltd. The phosphate buffered saline (PBS) was prepared by mixing 8.0 g NaCl, 0.2 g KCl, 1.44 g Na2HPO4, and 0.24 g KH2PO4 in 1 L of purified water (pH adjusted to 7.4). The E. coli strains, DH5α, BL21, and Rosetta, were purchased from Beijing TransGen Biotech Co., Ltd. The polyclonal antibody for anti-E. coli DH5α produced in rabbit was purchased from Beijing Biosynthesis Biotechnology Co., Ltd., and redispersed in PBS buffer solution to a concentration of 0.1 mg/mL. Purified water was used throughout the study. For real sample detection, Dole apple juice (product of China) and the Country Goodness fresh pure milk (product of New Zealand) were used. UV−visible absorption spectra of AuNP colloids and E. coli suspensions were measured with a Hitachi U-3010 spectrophotometer (Japan). The size of AuNPs was characterized using a Hitachi H-9000 transmission electron microscope (Japan). All the dark-field images were captured using an Olympus BX-53 9722
dx.doi.org/10.1021/ac302471c | Anal. Chem. 2012, 84, 9721−9728
Analytical Chemistry
Article
The automatic analysis algorithm, as outlined in Scheme 1, was composed of four steps: color space conversion, lightness
pixels. And one pixel in the dark-field image equaled an actual dimension of 0.15 × 0.15 μm2. Verification of AuNPs Surface Functionalization. E. coli DH5α (target) and BL21(control) strains were used to verify surface modification of AuNPs. To evaluate 11-MUDA modification efficiency, E. coli strains dispersed in purified water were mixed with unmodified and 11-MUDA-capped AuNPs, respectively. To evaluate the antibody conjugation efficiency, strains dispersed in PBS buffer were mixed with 11MUDA-capped AuNPs and antibody-conjugated AuNPs, respectively. Optimization of the AuNP-to-Bacterium Concentration Ratio. To find the concentration of AuNPs most suitable for bacterial recognition and imaging, 0.25, 0.50, 1.0, 2.0, and 4.0 mL of antibody-conjugated AuNP colloids were each mixed with 1.0 mL of diluted E. coli DH5α suspension (OD600 at about 0.2−0.3), and the mixtures were diluted to a volume of 5.0 mL with PBS, and incubated at room temperature for 15 min. The bacterial recognition performance was then evaluated with a dark-field microscope. Recognition Performance of Antibody-Functionalized AuNPs. To test the performance of anti-DH5α antibodyconjugated AuNPs, three E. coli strains, DH5α, BL21, and Rosetta, were reacted with unmodified, 11-MUDA-capped and antibody-conjugated AuNPs, respectively. Quantitative Detection. For quantitative detection, the original E. coli DH5α suspension with the bacterial concentration at the level of 108 CFU/mL was diluted sequentially with PBS buffer to different concentrations. Then the diluted suspensions were quantified by both plating and microscopic imaging, respectively. For plating, the original suspension was diluted sequentially by 10-fold, then 60 μL of each diluted sample was evenly spread on the LB agar plate and cultivated at 37 °C for 18−24 h, and bacterial colonies were counted. For microscopic imaging, the original suspension was diluted by 30×, 100×, 300×, 1000×, 3000×, and 10000×, respectively. A total of 20 μL of antibody-conjugated AuNPs was mixed with 100 μL of E. coli suspension. The mixture was incubated at room temperature for 15 min. A total of 5 μL of the mixture was then placed on the hemacytometer for microscopic imaging and counting. Sampling area was at least 1 mm2, with bacterial events equal to or greater than 10; if less than 10 events were observed, sampling area was increased such that the bacterial events were at least 10. Milk and Fruit Juice Sample Treatment. E. coli DH5α (target) and BL21 (control) strains were used to evaluate the feasibility of this proposed method with representative real used samples. Briefly, bacteria were incubated until OD600 reached approximately 1. Then 0.5 mL of the E. coli suspension was added to each 4.5 mL of the milk or apple juice sample. The spiked apple juice and milk samples were centrifuged at 3000 rpm for 5 min. For the apple juice sample, the precipitate was redispersed with PBS buffer to a volume of 5.0 mL. For the milk sample, the precipitate was washed with 5 mL of PBS buffer one time and redispersed with PBS buffer to a volume of 5.0 mL. Automatic Bacterial Counting. The AuNPs and bacteria presented different colors and shapes as the dark-field images. The AuNPs were typically round or elliptical with a yellow or red color. Bacteria were hollowed rods with white or slight green color. The counting algorithm for automatic analysis was developed based on these chroma and geometrical characters.
Scheme 1. Outline of the Bacterial Counting Algorithm for Automatic Analysis
thresholding, and image segmentation, as well as shape and color analysis. Color space conversion transferred the original dark-field images into a color space for further processing. Lightness thresholding operation reduced background interference in the image. Region-based and color-based segmentations were then applied to split a full image into a series of subimages which were simpler and easier to analyze. In the final step, shape and color analysis was performed on subimages to identify bacteria and AuNPs. Optimization of the parameters in lightness thresholding and image segmentation was required prior to automatic analysis of a series of images with the same capture condition (see Supporting Information for details). Detailed description of the automatic analysis algorithm is as follows. Color Space Conversion. The images captured by microscope were stored in sRGB color space and were converted to CIE LAB color space to perform the color analysis. The D65 white point was used as the reference for color space conversion. Lightness Thresholding. The total lightness levels (L* element in CIE LAB color space) of the images may vary as capture condition alters, thus, resulting in background interference to subsequent image segmentation and shape analysis. Therefore, thresholding was required to simplify the image. In this step, pixels with the lightness lower than a given threshold were assigned to black. The threshold value was carefully selected to reduce background noise the most, keeping the shape characters of AuNPs and bacteria at the same time. Detailed description is provided in the Optimization of Algorithm Parameters section in the Supporting Information. Image Segmentation. The image segmentation contained two individual steps, the region-based and color-based 9723
dx.doi.org/10.1021/ac302471c | Anal. Chem. 2012, 84, 9721−9728
Analytical Chemistry
Article
Figure 1. Dark-field images of unmodified, 11-MUDA-modified, and anti-DH5α antibody modified AuNPs reacted with E. coli DH5α and BL21 strains.
segmentations. Region-based segmentation was applied first to split the full image into a series of subimages (S1−Sn) with each containing only one intact object. As the previous thresholding process may remove the pixels belonging to bacteria and AuNPs, “intact” was defined as “the distance between two nonblack pixels was equal or less than a given distance (r) measured by the number of pixels”, to best keep the shape characters of a single object (an AuNP, a gold-bacterium composite, or a bacterium). The color-based segmentation extracted the gold color channel from above subimages. For each subimage, all the nonblack pixels were compared with a color designated as the color of gold using the CIE Delta E 2000 algorithm.32 The pixels with the color differences less than a given threshold value are collected as the gold color channel image (Gc1−Gcn). Shape and Color Analysis. The shape and color analysis was performed on each subimage and gold color channel image to identify the content of each subimage. To facilitate shape analysis, all subimages (S1−Sn) and gold color channel images (Gc1−Gcn) were first converted to binary images (B1−Bn, BG1− BGn), by assigning the black pixel to 0 and other pixels to 1. For each binary subimage in B1∼Bn, ⌈r/2⌉ (rounding up) rounds of dilation operations were then applied to make “intact” object connected. Subsequently, the shrinking operation33 was applied until there was no change in the binary subimage occurring in a single round of shrinking. With dilation and shrinking operation, the profile of the E. coli cells in the final binary subimages (denoted as Bf1−Bfn) was an oval ring as the cells were hollow rod shaped. The identification of an E. coli cell was performed based on the analyses of both the ring profile obtained from the shrinking operation and the color associated. In ring profile analysis, oval ring characteristic parameters in final binary subimages (Bf1−Bfn), such as the orientation angle, the ratio of the minor axis to the major axis (axis ratio),34 and the perimeter35 were calculated (see Supporting Information for detail). Additionally, as a distinctive shape character of the ring obtained from an E. coli cell, the profile of two sides in the longitudinal direction of an oval ring formed a pair of parallel line segments. Hough transform36 was applied on Bf1−Bfn to retrieve the direction and other parameters of the line segments (see Supporting Information for detail). In color analysis, all the nonblack pixels in the corresponding subimage (S1−Sn), that the distance between the pixel and the ring was equal to or less
than half of the thickness of the actual bacterial membrane, are extracted as a possible bacterial cell image. With results from the above analyses, a subimage containing ring profiles in the corresponding final binary subimage was considered to contain an E. coli cell if the following criteria were met: (1) AuNPs and E. coli color dominates the color of one of the possible bacterial cell images. Specifically, the percentage of pixels with the same color as either AuNPs or E. coli cells around the ring was greater than a given value (60% was typically observed and has been shown to be the appropriate value). (2) For the subimage meeting criterion (1), the perimeter of the corresponding ring profile was greater than a given value and the axis ratio was less than a given value, to match the geometric profile of an actual bacterium. (3) For the subimage meeting criterion (1) but not (2), the orientation angle of the corresponding ring profile matched the line direction parameter from Hough transform. For each subimage, pixels near the rings which meet the above criterions were extracted as the E. coli cell image (E1− En). The recognition of AuNPs was much simpler. For each binary gold color channel image (BG1−BGn), region-based segmentation was also performed to extract connected components. The perimeter and the axis ratio of each component were calculated and the ones with perimeter or axis ratio less than a given value (compared with 90 nm AuNP) were excluded. The rest components were identified as AuNPs and collected as the AuNP images (G1−Gn). After the identification of E. coli cells and AuNPs, each E. coli cell image (E1−En) was merged with the corresponding AuNP image (G1−Gn). For each pair of images, if the distance between any nonblack pixel in the AuNP image and in the E. coli image was equal or less than a given value (typically zero, or a few pixels in cases of poor color extraction), the corresponding subimage was considered to contain an E. coli with AuNP(s) attached to its surface. The CIE LAB color tuple in the program was coded in double-precise floating-point format, and the automatic bacterial counting program was written in C#. MATLAB was used to test if color space conversion was correctly 9724
dx.doi.org/10.1021/ac302471c | Anal. Chem. 2012, 84, 9721−9728
Analytical Chemistry
Article
implemented (see Implementation of Algorithm section in the Supporting Information for detail).
■
RESULTS AND DISCUSSION Dark-Field Images of AuNPs and E. coli Cells. Under the dark-field microscope, E. coli cells were slightly green colored hollow rods, and AuNPs were yellow or red colored spheres (Figure 1). The lightness of AuNPs was comparable to or higher than that of the E. coli cell. Both the signal reporter and the bacterial target images were obtained at the same time, indicating that the prepared AuNPs were suitable probes for dark-field imaging bacterial detection. The number of events of bacteria in an image is associated with the concentration of bacteria suspension used. Enough bacterial events should be collected to obtain reliable counting results. Therefore, a high concentration of bacteria suspension (∼107 CFU/mL) was used in method development for labor-saving purposes. At such a concentration level, more than 100 events could be collected from one original dark-field image. Verification of AuNP Surface Functionalization. Selectivity of anti-DH5α strain polyclonal antibody functionalized AuNPs was evaluated by reacting the unmodified, 11MUDA-capped, and anti-DH5α antibody-conjugated AuNPs with E. coli DH5α and BL21 strains (Figure 1). Due to the interaction between gold and glyco-compounds, unmodified AuNPs were extensively adsorbed onto both E. coli DH5α and BL21 strains in purified water (Figure 1a,c). The 11-MUDAcapped AuNPs did not show any adsorption to the bacterial surface (Figure 1b,d), indicating successful modification of AuNPs. Antibody required adequate pH and ionic strength for the best recognition performance, thus, 11-MUDA-capped AuNPs and antibody-conjugated AuNPs were reacted with bacteria in PBS to verify antibody modification. In PBS buffer, 11-MUDA-capped AuNPs remained separated from bacteria (Figure 1e,g), while antibody-conjugated AuNPs selectively bound to E. coli DH5α (Figure 1f,h), demonstrating that antibody was successfully immobilized on the surface of AuNPs with sufficient binding activity retained. AuNP-to-Bacterium Concentration Ratio. The goal of this study was to confirm the feasibility of a dark-field imaging and counting method for E. coli detection, so optimization was carried out mainly on the recognition reaction and imaging performance rather than AuNP modification and the performance of antibody. The AuNP-to-E. coli concentration ratio was the major factor affecting the detecting performance under the microscope. To facilitate easy concentration ratio optimization, batch-to-batch variation in the size of AuNPs was not considered. Manual counting of the E. coli cells was employed for quantification, and the value of “recognition percentage (the proportion of the bacteria with AuNPs in total bacteria)” was used to quantify the recognition performance of antibody modified AuNPs. The AuNPs-to-E. coli concentration ratio was presented for convenience as the ratio of volumes. As can be seen in Figure 2, recognition percentage increased with the volume ratio. A volume ratio greater than 1.0 was adequate for bacterial detection. Because high concentrations of AuNPs resulted high density images that were difficult for bacterial identification and counting and AuNPs were more apt to aggregate in high concentration, the volume ratio of 2.0 was chosen for the following experiments. To be consistent with the definition of a colony forming unit (CFU), bacteria adhered in clusters were considered as one bacterium.
Figure 2. Recognition percentage at different AuNP-to-bacterium ratios. Data were an average of four batches of experiments. In the final suspensions, the concentrations were 4.1, 8.2, 16, 33, and 66 pmol/mL for AuNPs, and ∼107 CFU/mL for E. coli.
Selectivity of Anti-DH5α Strain Polyclonal Antibody Functionalized AuNPs. The selectivity of antibody-modified AuNP was evaluated by reacting with the target strain (E. coli DH5α) and control stains (E. coli BL21 and Rosetta). All three strains were reacted with unmodified AuNPs, 11-MUDAcapped AuNPs, and antibody-conjugated AuNPs in PBS buffer, respectively. Different from the results in purified water, unmodified AuNPs did not react with any E. coli strains in PBS buffer (Figure S-2, left column). Ionic strength at this level might have helped to stabilize and to prevent unmodified AuNPs from adsorption onto bacteria. 11-MUDA-capped AuNPs were not attached to E. coli cells in PBS buffer, although slight aggregation was observed (Figure S-2, middle column). It is clear that anti-DH5α antibody functionalized AuNPs selectively reacted with the target bacteria (E. coli DH5α), showing that reasonable selectivity was achieved (Figure S-2, right column). Manual Bacterial Counting. Manual counting was performed for all the dark-field images of the target strain (E. coli DH5α) and control stains (E. coli BL21 and Rosetta). Three batches of bacterial detection were performed for each bacterial strain. The results (Table 1) showed that 77% DH5α Table 1. E. coli Detection by Manual Counting number of bacteriaa
a b
samples
DH5α
batch 1 batch 2 batch 3 recognitionb (%)
152 (61) 344 (26) 239 (118) 77 ± 14
BL21 35 50 27 15
(141) (232) (297) ±6
Rosetta 22 19 24 13
(111) (156) (186) ±3
The number of the bacteria with and (without) AuNPs attached. The ratio of the bacteria with AuNPs to the total number of bacteria.
cells were attached with AuNPs, and the binding percentage was less than 15% for the other two strains. It is understood that the antibody was not produced against intact E. coli cells, thus, binding affinity may not be the best. As a result, incomplete recognition (binding percentage of 77%) toward the target strain resulted with the AuNP-to-bacterium ratio that was adopted to optimize imaging results in this study. Also, nonspecific binding of antibody with control strains was observed. This may be due to the recognition performance of the polyclonal antibody used in the experiment, which was 9725
dx.doi.org/10.1021/ac302471c | Anal. Chem. 2012, 84, 9721−9728
Analytical Chemistry
Article
Milk and Apple Juice Sample Detection. To evaluate sample matrix effect on the bacterial detection, milk and fruit juice samples were analyzed with the proposed dark-field imaging method by spiking the real sample with the bacterial strain suspensions. The high concentration of glyco-compounds in apple juice may bind with AuNPs, thus, interfering with AuNP binding to bacteria and, thus, the microbial detection. Proteins, lipids, and other water-insoluble organic compounds in milk could induce light-scattering to interfere with dark-field imaging analysis and bacteria counting. Centrifugation was used to remove the potential interferences from the matrix of the samples. E. coli strains, DH5α and BL21, were detected in spiked apple juice and milk samples (Figure 4). The recognition percentage of target E. coli strain (DH5α)
produced by using the whole cell proteins of E. coli DH5α. The specificity of polyclonal antibody is generally worse than monoclonal antibody and showed binding with two other control strains. Nevertheless, a reasonable difference in recognition between the target and the control strains was achieved with good reproducibility; furthermore, nonspecific binding did not affect the comparison between manual and automatic counting of bacteria in our study. The results of manual bacterial counting indicated that the AuNP-based darkfield imaging and counting method were feasible for bacterial recognition. Quantitative Detection. The calibration curve was constructed for DH5α, and results were compared with the plating method. As shown in Figure 3, the microscopic imaging
Figure 4. Bacterial counting results of E. coli DH5α and BL21 strain spiked milk and apple juice samples (data were an average of duplicate experiments).
was 71% for apple juice sample and 73% for milk sample, comparable with that (77%) in water. The level of nonspecific binding with control E. coli strain (BL21), 25% for apple juice sample and 26% for milk sample, was also comparable with the results from purified water. These results indicated that this detection method could effectively identify E. coli strains in real samples with very simple sample pretreatments. Automatic Bacterial Counting. Among the published methods for bacterial recognition, light-scattering based single bacterium recognition was usually carried out by multiangle scattering pattern based recognition on flow cytometer,39,40 because the multiangle scattering pattern provides enough information for identification. Microscopic imaging can only acquire the geometric profiles of bacteria, which is insufficient for general bacteria identification. Thus, with shape analysis, microscope-based methods can only be applied to samples with very few coexisting types of bacteria.41−43 By introducing AuNPs as the light-scattering label, target bacteria can be identified by the spatial relationship between AuNPs and the bacterial cell. And the requirement for shape analysis is only to differentiate bacteria, gold nanoparticles, and interferences with each other to prepare for the abovementioned spatial relationship judgment, not to identify the type of bacteria. Thus, it was easy to select profile features for shape analysis based on the observation. Automatic bacteria recognition algorithm was established using the training images (see Optimization of Algorithm Parameters in Supporting Information for details), and the automatic recognition and counting were carried out with the first batch of dark-field images that were used for manual
Figure 3. Comparison between plating and microscopic imaging results (data were an average of duplicate experiments).
results of total bacteria at each concentration level exhibited good correlation with the plating results with a slope of 0.956 (R2 = 0.9984). The E. coli concentration in the original suspension was determined to be 2.8 × 108 CFU/mL by plating and 2.9 × 108 CFU/mL by microscopic imaging. The microscopic imaging method gave a slightly larger number of bacteria than plating method. The reason might be that the microscopic imaging method counts all visible bacteria, while the plating method reflects only the number of viable bacteria. The ratio of AuNP attached bacteria to the total bacteria was consistent with previous results in this work. The dark-field light-scattering imaging and antibodyconjugated AuNPs ensure the identification of a single bacterium. From this point of view, the limit of detection is determined by the volume of sampling area. The hemacytometer provides a 3 × 3 mm squared (0.18 mm3) and a round (0.50 mm3) sampling fields for quantification. And more than 10 bacterial events are generally required to ensure the accuracy of bacterial counting.37 Therefore, for a sampling area containing 10 bacteria, the detection limit is about 2−6 × 104 CFU/mL, approximately 100 bacterial cells in the sample suspension applied on the hemacytometer. The limit of detection of our method is comparable to that of flow cytometry methods.38 Lower limit of detection can be achieved by sample preconcentration. 9726
dx.doi.org/10.1021/ac302471c | Anal. Chem. 2012, 84, 9721−9728
Analytical Chemistry
Article
Regardless of preconcentration, the detection limit for E. coli was ∼104 CFU/mL. And good recognition performance for the target strain in water, milk, and juice samples was obtained within only 15−30 min. Furthermore, an automatic bacterial counting algorithm was developed for robust and faster image analysis. Overall, this work has provided a new way for rapid and low-cost bacterial selective counting, which is the need of bacterial detection.
counting and not overlapped with training images. After the parameters of the automatic analysis algorithms were optimized using one of the images prior to autorecognition, the automatic bacterial recognition and counting were performed (details described in Supporting Information). First, 60∼72% of the total bacteria (including bacteria with or without AuNPs) was identified with the automatic analysis (Figure 5), which may be
■
ASSOCIATED CONTENT
* Supporting Information S
Additional information, as noted in the text. This material is available free of charge via the Internet at http://pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Author
*Tel.: +86 10 62761187. Fax: +86 10 62751708. E-mail: lina@ pku.edu.cn. Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS This work was supported by the National Natural Science Foundation of China (Nos. 20975004, 21035005, and 21275011) and a Merieux Research grant. The authors gratefully thank Prof. Chengzhi Huang for suggestions in manuscript preparation and Dr. John Hefferren for English editing.
Figure 5. Comparison of manual and automatic counting results of total E. coli bacteria in dark-field images.
adequate for qualitative detection although the negative error was observed comparing with manual counting. In terms of target cell recognition percentage, automatic counting results were comparable to manual counting (Table 2).
■
(1) WHO Enterohaemorrhagic Escherichia coli (EHEC), http:// www.who.int/mediacentre/factsheets/fs125/en/ (accessed April 5, 2012). (2) Willshaw, G. A.; Thirlwell, J.; Jones, A. P.; Parry, S.; Salmon, R. L.; Hickey, M. Lett. Appl. Microbiol. 1994, 19 (5), 304−307. (3) Bolton, F. J.; Crozier, L.; Williamson, J. K. Lett. Appl. Microbiol. 1996, 23 (5), 317−321. (4) Batt, C. A. Science 2007, 316 (5831), 1579−1580. (5) Connelly, J.; Baeumner, A. Anal. Bioanal. Chem. 2012, 402 (1), 117−127. (6) Kim, H. C.; Bhunia, A. K. Appl. Environ. Microbiol. 2008, 74 (15), 4853−4866. (7) Yu, Y. G.; Wu, H.; Liu, Y. Y.; Li, S. L.; Yang, X. Q.; Xiao, X. L. Can. J. Microbiol. 2010, 56 (7), 585−597. (8) Reyes, L.; Morell, E.; Cabrera, M. A.; All, L.; Flores, J. Acta Cient. Venez. 1978, 29, 127−127. (9) Kerr, P.; Chart, H.; Finlay, D.; Pollock, D. A.; Mackie, D. P.; Ball, H. J. J. Appl. Microbiol. 2001, 90 (4), 543−549. (10) Neely, E.; Bell, C.; Finlay, D.; McCappin, J.; Wilson, I.; Ball, H. J. J. Appl. Microbiol. 2004, 97 (6), 1161−1165. (11) Sunwoo, H. H.; Wang, W. W.; Sim, J. S. Immunol. Lett. 2006, 106 (2), 191−193. (12) Perna, N. T.; Plunkett, G.; Burland, V.; Mau, B.; Glasner, J. D.; Rose, D. J.; Mayhew, G. F.; Evans, P. S.; Gregor, J.; Kirkpatrick, H. A.; Posfai, G.; Hackett, J.; Klink, S.; Boutin, A.; Shao, Y.; Miller, L.; Grotbeck, E. J.; Davis, N. W.; Limk, A.; Dimalanta, E. T.; Potamousis, K. D.; Apodaca, J.; Anantharaman, T. S.; Lin, J. Y.; Yen, G.; Schwartz, D. C.; Welch, R. A.; Blattner, F. R. Nature 2001, 409 (6819), 529− 533. (13) Koh, C. G.; Tan, W.; Zhao, M.-q.; Ricco, A. J.; Fan, Z. H. Anal. Chem. 2003, 75 (17), 4591−4598. (14) Chuang, H.; Macuch, P.; Tabacco, M. B. Anal. Chem. 2000, 73 (3), 462−466. (15) Gehring, A. G.; Patterson, D. L.; Tu, S. I. Anal. Biochem. 1998, 258 (2), 293−298.
Table 2. Comparison between Manual and Automatic Analysis No. of events (%)
a
strains
recognition percentage
missed target
false target
DH5α BL21 Rosetta
67/71%a 24/20%a 17/17%a
11 (7%) 2 (2%) 2 (2%)
4 (3%) 7 (7%) 4 (5%)
REFERENCES
Results from manual counting.
Missed target cases identified bacterium with AuNPs as one without AuNPs, resulting in negative errors. False target results showed recognition occurred in the opposite direction, resulting in positive errors. The total positive and negative errors were less than 10%, indicating good accuracy in judging the binding status of functionalized AuNPs with bacteria. Among the three types of strains tested, the maximum difference in recognition percentages between manual and auto recognition results was as low as 4%. Among the automatic results, only one case was an interfering object identified as a bacterium, indicating strong anti-interference capability of the automatic recognition algorithm (DH5α, data not shown).
■
CONCLUSION In summary, a bacterial counting detection method was established for the first time based on dark-field imaging that can obtain the signals of bacteria and the AuNP reporter simultaneously. By using the AuNP reporter, the detection can be performed with a common tungsten lamp and an entry-level microscope setup, thus, the cost of instrumentation was low. 9727
dx.doi.org/10.1021/ac302471c | Anal. Chem. 2012, 84, 9721−9728
Analytical Chemistry
Article
(16) Hnaiein, M.; Hassen, W. M.; Abdelghani, A.; Fournier-Wirth, C.; Coste, J.; Bessueille, F.; Leonard, D.; Jaffrezic-Renault, N. Electrochem. Commun. 2008, 10 (8), 1152−1154. (17) Pohanka, M.; Skladal, P.; Pavlis, O. J. Immunoassay Immunochem. 2008, 29 (1), 70−79. (18) Ruan, C. M.; Zeng, K. F.; Varghese, O. K.; Grimes, C. A. Anal. Chem. 2003, 75 (23), 6494−6498. (19) Byrne, B.; Stack, E.; Gilmartin, N.; O’Kennedy, R. Sensors 2009, 9 (6), 4407−4445. (20) McClain, M. A.; Culbertson, C. T.; Jacobson, S. C.; Ramsey, J. M. Anal. Chem. 2001, 73 (21), 5334−5338. (21) López-Amorós, R.; Castel, S.; Comas-Riu, J.; Vives-Rego, J. Cytometry 1997, 29 (4), 298−305. (22) Yang, W.; Pan, C.-Y.; Luo, M.-D.; Zhang, H.-B. Biomacromolecules 2010, 11 (7), 1840−1846. (23) Yang, L.; Li, Y. Analyst 2006, 131 (3), 394−401. (24) Su, X.-L.; Li, Y. Anal. Chem. 2004, 76 (16), 4806−4810. (25) Murphy, C. J.; Gole, A. M.; Stone, J. W.; Sisco, P. N.; Alkilany, A. M.; Goldsmith, E. C.; Baxter, S. C. Acc. Chem. Res. 2008, 41 (12), 1721−1730. (26) Willets, K. A.; Van Duyne, R. P. Annu. Rev. Phys. Chem. 2007, 58, 267−297. (27) Phillips, R. L.; Miranda, O. R.; You, C.-C.; Rotello, V. M.; Bunz, U. H. F. Angew. Chem., Int. Ed. 2008, 47 (14), 2590−2594. (28) Sanvicens, N.; Pastells, C.; Pascual, N.; Marco, M. P. TrAC, Trends Anal. Chem. 2009, 28 (11), 1243−1252. (29) Yguerabide, J.; Yguerabide, E. E. Anal. Biochem. 1998, 262 (2), 137−156. (30) Kelly, K. L.; Coronado, E.; Zhao, L. L.; Schatz, G. C. J. Phys. Chem. B 2003, 107 (3), 668−677. (31) Brown, K. R.; Walter, D. G.; Natan, M. J. Chem. Mater. 1999, 12 (2), 306−313. (32) Sharma, G.; Wu, W.; Dalal, E. N. Color Res. Appl. 2005, 30 (1), 21−30. (33) Pratt, W. K.; Kabir, I. Morphological binary image processing with a local neighborhood pipeline processor. In Proceeding of Computer Graphics Tokyo ’84; Springer: Tokyo, 1985; pp 321−343. (34) Pratt, W. K. Digital Image Processing: PIKS Scientific Inside, 4th ed.; John Wiley & Sons: New York, 2007. (35) Gray, S. B. IEEE Trans. Comput. 1971, C-20 (5), 551−561. (36) Hough, P. V. C. Method and Means for Recognizing Complex Patterns. U.S. Patent 3069654, 1962. (37) GB/T 4789.38−2012 Microbiological examination of food hygiene. Enumeration of Escherichia coli, National Standard of the People’s Republic of China, 2012. (38) Yang, L.; Wu, L.; Zhu, S.; Long, Y.; Hang, W.; Yan, X. Anal. Chem. 2009, 82 (3), 1109−1116. (39) Steen, H. B. Cytometry 1990, 11 (2), 223−230. (40) Rajwa, B.; Venkatapathi, M.; Ragheb, K.; Banada, P. P.; Hirleman, E. D.; Lary, T.; Robinson, J. P. Cytometry, Part A 2008, 73A (4), 369−379. (41) Tsibidis, G. D.; Burroughs, N. J.; Gaze, W.; Wellington, E. M. H. Micron 2011, 42 (8), 911−920. (42) Hiremath, P. S.; Bannigidad, P. Automatic classification of bacterial cells in digital microscopic images. Second International Conference on Digital Image Processing, Singapore, Singapore, Febuary 26, 2010; SPIE: Bellingham, WA, USA, 2010; pp 754613−754613. (43) Hiremath, P. S.; Bannigidad, P. Digital image analysis of cocci bacterial cells using active contour method. Signal and Image Processing (ICSIP), 2010 International Conference, Chennai, India, Dec 15−17, 2010; Institute of Electrical and Electronics Engineers (IEEE): Washington, DC, USA, 2010; pp 163−168.
■
NOTE ADDED AFTER ASAP PUBLICATION This paper was published on the Web on October 17, 2012. A revision was made to the Acknowledgments, and the corrected version was reposted on October 30, 2012.
9728
dx.doi.org/10.1021/ac302471c | Anal. Chem. 2012, 84, 9721−9728
