Detecting and tracking nosocomial methicillin-resistant Staphylococcus aureus using a microfluidic SERS biosensor

Abstract

Rapid detection and differentiation of methicillin-resistant Staphylococcus aureus (MRSA) is critical for the early diagnosis of difficult-to-treat nosocomial and community acquired clinical infections and improved epidemiological surveillance. We developed a microfluidics chip coupled with surface enhanced Raman scattering (SERS) spectroscopy (532 nm) “lab-on-a-chip” system to rapidly detect and differentiate methicillin-sensitive S. aureus (MSSA) and MRSA using clinical isolates from China and the United States. A total of 21 MSSA isolates and 37 MRSA isolates recovered from infected humans were first analyzed by using polymerase chain reaction (PCR) and multilocus sequence typing (MLST). The mecA gene, which refers resistant to methicillin, was detected in all the MRSA isolates and different allelic profiles were identified assigning isolates as either previously identified or novel clones. A total of 17,400 SERS spectra of the 58 S. aureus isolates were collected within 3.5 hours using this optofluidic platform. Intra- and inter-laboratory spectral reproducibility yielded a differentiation index value of 3.43 to 4.06 and demonstrated the feasibility of using this optofluidic system at different laboratories for bacterial identification. A global SERS-based dendrogram model for MRSA and MSSA identification and differentiation to the strain level was established and cross-validated (Simpson index of diversity of 0.989) and had an average recognition rate of 95% for S. aureus isolates associated with a recent outbreak in China. SERS typing correlated well with MLST indicating that it has high sensitivity and selectivity and would be suitable for determining the origin and possible spread of MRSA. A SERS-based partial least-squares regression model could quantify the actual concentration of a specific MRSA isolate in a bacterial mixture at levels from 5 to 100% (regression coefficient, > 0.98; residual prediction deviation, >10.05). This optofluidic platform has advantages over traditional genotyping for ultrafast, automated and reliable detection and epidemiological surveillance of bacterial infections.

INTRODUCTION

Methicillin-resistant Staphylococcus aureus (MRSA) is a leading pathogen associated with serious hospital and community-acquired disease ¹. Early diagnosis of MRSA infection in humans is critical to determine a suitable treatment to reduce morbidity and mortality ². MRSA can be detected using chromogenic and selective media ³, and various DNA-based genotyping techniques, including commercially available BD GeneOhm and Xpert MRSA assays ⁴, polymerase chain reaction (PCR) ⁵, pulsed-field gel electrophoresis (PFGE) ⁶ and multilocus sequence typing (MLST) ⁷. However, all of these techniques are time consuming, laborious, require highly trained personnel, dedicated facilities and sometimes provide false-negative or false-positive results ⁸. Biosensor and bioelectronics techniques offer an alternative to these more traditional detection methods for rapid detection of MRSA ⁹.

Confocal micro-Raman spectroscopy provides the basis for epidemiological typing method for various pathogenic bacteria, such as Pseudomonas aeruginosa ¹⁰, Mycobacterium spp. ¹¹, Campylobacter spp. ¹², and S. aureus ¹³. Conducting SERS to detect bacteria at the single cell level without cultivation shows great potential ¹⁴. However, the major disadvantage of SERS to date has been the requirement for metal nanocolloids to serve as the SERS active substrate which, for bacterial detection, often results in poor spectral reproducibility due to variations in the properties of different batches of nanocolloid and phenomenon that bacterial detection is dependent upon nanocolloid concentration and this relationship needs to be empirically established for different analytes ¹⁵. Signal enhancement effects can vary within the same batch of nanocolloid because of the nonhomogeneity and unpredictable aggregation properties of the nanoparticles within the colloidal suspension ¹⁶. Thus, a high number of spectra for each biological sample are required to confirm the spectral features and subsequently establish and validate reliable chemometric models for prediction of analyte concentration in samples of unknown composition ¹⁷.

Optofluidic platforms with SERS detection have been constructed for detection of chemical and biological samples in a well-controlled microenvironment ¹⁸. For example, Walter et al. ¹⁹ designed a SERS microfluidic system to rapidly discriminate between nine strains of Escherichia coli. This optofluidic system showed substantial advantages compared to traditional SERS detection under a static condition. Spectral measurement taken under controlled flow conditions reduces problems with differing scattering geometry of the nanoparticles, instable mixing times, and localized heating ²⁰. In addition, a microfluidic system with this design could reliably handle small sample volumes and a high sample throughput ²¹. By creating a segmented flow format using mineral oil as the segregation medium ²², memory effects were reduced ²³ as were peak tailing ²⁴ and baseline drift ²⁵.

Here, we report an automated PDMS-based optofluidic platform combining microfluidics and SERS to rapidly detect and investigate the epidemiology of MRSA isolated from infected humans. SERS typing results correlated well with reference genotyping methods, including PCR and MLST. To the best of our knowledge, this is the first study conducting an optofluidic based “lab-on-a-chip” detection system to perform epidemiological monitoring of contagious bacterial infections.

EXPERIMENTAL SECTION

The detailed information about chemicals and reagents, bacterial sample collection and preparation, DNA extraction, detection of mecA by PCR, and MLST was in Supplementary Information section.

Microfluidic chip fabrication

The fabrication process is shown in Fig. S1 and the detailed processing procedures were in Supplementary Information section.

Lab-on-a-chip device for SERS measurements

The microfluidic system with T-junction geometry was designed and fabricated to produce droplets using standard soft lithography technique ²⁶. A microfluidic syringe pump system (NE-1002X, New Era Pump Systems Inc., Farmingdale, NY) was used to inject the reagents and control flow at ultra-low flow rates of less than 0.01 μl/s. The programmed linear gradients were flow-rates of 0.008 μl/s for the mineral oil, 0.006 μl/s for the silver nanocolloid solution, 0.001 μl/s for 0.5 molar sodium chloride solution (aggregation agent), and 0.01 μl/s for the bacterial suspension. The principle of silver nanocolloid preparation is based on the reduction of silver cations and the synthesis procedure was mainly followed by previous method ²⁷. Multiple batches of fresh silver nanocolloids were prepared in each laboratory to validate the SERS-based pattern recognition. Under these conditions, the bacterial suspension formed a series of droplets in the carrier oil in the microchannel. These droplets were then mixed with silver nanocolloid solution as well as the aggregation agent. Chaotic advection inside the droplets enhanced mixing within the designed winding channels ²⁸.

Confocal micro-Raman spectroscopic instrumentation

Two confocal micro-Raman spectroscopic systems were set up at two different laboratories, one in China (Renishaw inVia) and one in the United States (WITec alpha300) as described in detail in our recent publication ¹². Both systems have a charge-coupled device (CCD) array detector. A 532-nm double Nd:YAG laser (10 mW, 0.5 s integration time) was introduced through a 100× objective into the microchannel and focused on the sample droplet (Fig. 1) for spectral collection over a simultaneous wavenumber shift range of 3700 to 200 cm⁻¹ in an extended mode. During the ongoing segmented flow of the droplets, the individual Raman spectra of mineral oil and bacterial samples within the mineral oil were collected accordingly. Spectral processing and subtraction could be employed to remove the spectral interference from mineral oil, as described previously ^21–22.

Figure 1.

Scheme of the optofluidic platform including a confocal micro-Raman spectroscopic system coupled with microfluidic “lab-on-a-chip” system. The microfluidic chip was connected with syringe pumps and placed in the measurement stage, where the droplets inside of the microchannel were illuminated with the incident laser (green color) and the scattered light (red color) was separated by the notch filter to remove Rayleigh scattering light and only allow Raman signal (red color dash line) to pass through. The Raman signal was dispersed on a diffraction grating and collected using a charge-coupled device (CCD) detector. The principle of the confocal technique was demonstrated that only the scattering light derived from the focal point could pass through the aperture.

The detailed information about spectral processing, discriminatory power calculation and chemometric analysis was in Supplementary Information section.

RESULTS

Optofluidic system

We fabricated the microfluidic chip using PDMS (Fig. S1) and detailed processing steps can be found in Experimental Section. A detailed scheme of the optofluidic platform including a confocal micro-Raman spectroscopic system coupled with microfluidic “lab-on-a-chip” system is shown in Fig. 1 and the configuration of the real platform is shown in Fig. S2a. Mineral oil was used as a carrier fluid in the microchannel. Droplets of bacteria suspension were generated using a T-junction structure, as shown in Fig. S2b. Depending on the fluid properties and flow rates, droplets can form due to three distinct mechanisms: squeezing, dripping and jetting. In our case, the capillary number Ca, which measures the ratio of viscous force and surface tension force, is estimated to be 10⁻³. It has been reported³² that the transition from squeezing to dripping at a T-junction happens at Ca = 10⁻². Therefore, in our case, the droplets were generated via the squeezing mechanism, meaning that surface tension forces dominated shear stresses and the size of the droplets was controlled by the ratio of the flow rates of the two fluids only. Based on the equations proposed by Garstecki et al. ²⁹, the droplet length in our case is estimated to be 0.9 mm, which is very close to the experimentally measured value of 1.1 mm. Once the bacteria suspension droplets were formed, they were transported downstream to mix with another droplet containing silver nanocolloid and NaCl using the dosing technique developed by Henkel et al. ³⁰. The merged droplets were mixed thoroughly when they were pushed through a winding microchannel within a short time period. The fast mixing can be credited to the microfluidic baker’s transformation that are best characterized by a series of stretches, folds, and reorientations that decrease the striation length.

Raman spectral features of MRSA and MSSA

The optofluidic-based SERS spectra of S. aureus were collected along with the segmented flow in the microchannel and the representative processed SERS spectral features of MRSA and MSSA are shown in Fig. S3. The band assignments are summarized in Table S3 and could be referred to previous publications ^{12, 31}. MRSA and MSSA had similar SERS spectral features but some variations of band assignments were observed between MRSA and MSSA. For example, there were several band shifts between MRSA and MSSA, including a shift of the α-helix of amide I band from 1659 cm⁻¹ for MRSA to 1667 cm⁻¹ for MSSA, a shift of the C-C stretch of carbohydrate band from 911 cm⁻¹ for MRSA to 923 cm⁻¹ for MSSA and a shift of the C-H bending of tyrosine band from 1176 cm⁻¹ for MRSA to 1210 cm⁻¹ for MSSA (Fig. S3). These band shifts may be due to a different interaction mode between biological samples and the silver nanocolloids ^{12, 14b, 14e}. In addition, MRSA and MSSA had unique bands. For nucleic acids, MRSA had unique bands at 1373 cm⁻¹ (ring breathing modes of the DNA/RNA bases) and 1093 cm⁻¹ (DNA backbone-phosphate backbone) while MSSA had a unique band at 1084 cm⁻¹ (phosphodiester groups in nucleic acid). For proteins, MRSA had unique bands at 935 cm⁻¹ (α-helix of protein) and 819 cm⁻¹ (protein band) while MSSA had a unique protein band at 1270 cm⁻¹ (amide III). Additionally, MSSA had a unique lipid band at 966 cm⁻¹.

The spectral reproducibility (D_y1y2 value) was calculated using the wavenumbers between 1800 cm⁻¹ to 400 cm⁻¹ and the D_y1y2 values were between 3.43 and 4.06, indicating a better spectral reproducibility than our previous work where confocal micro-Raman spectroscopy was conducted to record the spectral features of Campylobacter ¹². Preliminary experiment was conducted using the commercially available Klarite substrate (Renishaw Diagnostics) for the SERS detection of MRSA. We found that the spectral reproducibility (D_y1y2 values between 11.63 and 16.29) was worse than the current optofluidic SERS system, indicating that the current engineering design significantly alleviates the problems of inhomogeneity for SERS.

The correlation between optofluidic-based SERS typing and PCR typing for S. aureus antibiotic resistance

The result of the PCR amplification of the mecA gene in 38 S. aureus isolates from China and 20 S. aureus isolates from the United States is summarized in Table S3 (37 MRSA isolates and 21 MSSA isolates). Amplification of the mecA gene for selective S. aureus isolates is shown in Fig. S4.

A global supervised DFA dendrogram model on the basis of SERS spectral features of intact S. aureus cells was established using the 38 S. aureus isolates from China and cross-validated using the 20 S. aureus isolates from the United States (Fig. 2). The spectral feature of each bacterial isolate in the supervised DFA dendrogram model has an average of 300 spectra collected from three independent experiments. Therefore, this model contained a total of 17,400 Raman spectra. The interclass distance was calculated based on Mahalanobis distance measurements between groups in the supervised DFA dendrogram model and the values were higher than 8.50. Clusters with interclass distance values higher than 3 are believed to be significantly different from each other ³². This result further validated the discrimination between MRSA and MSSA using optofluidic-based SERS typing.

Figure 2.

Composite dendrogram representing the global chemometric model to identify MRSA and MSSA from China and the United States. Entries in black represent the training data using MRSA and MSSA samples from China, and those in red denote the validation data using MRSA and MSSA samples from the United States. The model was established and validated as described in the text.

This DFA dendrogram model achieved a classification sensitivity of 96.5% and a specificity of 92.6% using Wards cluster algorithm at the cut-off value established at 99% similarity. The ROC curves were generated to further evaluate the group segregation (Fig. S5). Each point on the ROC curve represents a sensitivity/specificity pair corresponding to a particular decision threshold. The AUC under the ROC curve is 0.989, which indicated that the optofluidic SERS typing can well distinguish between two diagnostic groups (i.e., MRSA and MSSA).

The numerical index of discrimination of the optofluidic-based SERS typing was determined to be 0.989 by calculating the Simpson’s index of diversity. A value higher than 0.9 indicates a good discrimination typing method for bacterial samples ¹². Bayesian probability analysis was then employed to compare the top 15 significant features with PCs determined by using the DFA model. We found good agreement in the results of the two methods. The stability of the DFA model was determined using three different Monte Carlo estimations, one using the 15 most significant features, a second with the 15 least significant features, and one in which all features were selected and compared. The highest stability of the DFA model was derived from the use of the 15 most significant features (0.65 ± 0.12). A model with a slightly lower stability (0.58 ± 0.08) was achieved when all features were selected, indicating the contribution of non-significant features and possibly interference and noise. The lowest stability model was obtained when the 15 least significant features (0.02 ± 0.01) were used.

Prediction from a recent nosocomial S. aureus outbreak

An outbreak associated with multiple S. aureus isolates recently occurred in second hospital affiliated to Tianjin Medical University in summer 2012. Twenty S. aureus isolates were obtained from both patients and staff members in this hospital which may or may not have comprised different strains. Both PCR typing and optofluidic-based SERS typing were conducted and the spectra were incorporated into our cross-validated global dendrogram model (Fig. 2). The recognition results for each isolate are summarized in Table S4. The average recognition rate was 95% for these twenty isolates including 12 MRSA isolates encoding the mecA gene and 8 MSSA isolates without the mecA gene.

The correlation between optofluidic-based SERS typing and MLST for S. aureus epidemiology

MLST was conducted to determine the epidemiology of S. aureus isolates from the United States. The allelic fragments of S. aureus isolates for MLST are shown in Fig. S6 and the properties of the sequence types of S. aureus isolates from the United States are summarized in Table S2. The unsupervised HCA based dendrogram model is shown in Fig. 3, indicating a good concordance between optofluidic-based SERS typing and MLST. The optofluidic SERS spectral features of four different patterns of MRSA are shown in Fig. S7.

Figure 3.

Hierarchical cluster analysis of MRSA clinical isolates from the United States. Full details of the numbers of PCs determined at each level and the associated percentages of explained variance are shown. This dendrogram shows the reproducibility and concordance of optofluidic-based Raman typing and MLST.

Identification of MRSA in a bacterial mixture sample

Two clinical S. aureus isolates (i.e., MRSA S-93 and MSSA S-N7) were mixed in a ratio of 5% to 100% MRSA isolate in the mixture. The corresponding SERS spectra of these 11 different bacterial cell mixtures were recorded using the optofluidic platform and a PLSR model was established and a leave-one-out cross-validation conducted (Table S5). This supervised chemometric model included a total of 3300 spectra (11 MRSA-MSSA mixtures, 300 spectra for each mixture) with a high R and RPD values for both calibration and validation models, low RMSE values for both calibration and validation models and 6 of latent variables (Table S5). These parameters indicated that the PLSR model could accurately determine and predict the actual concentration of MRSA in a mixture of MRSA and MSSA strains.

DISCUSSION

Microfluidic chip design

Microfluidic systems have obvious advantages over traditional analytical methods by providing a miniaturized device in the form of a “lab-on-a-chip” system that integrates sample preparation, analyte separation, biochemical reaction and detection together ²¹, providing a well-controlled microenvironment. Recently, microfluidic systems have been employed to couple nucleic acid extraction and amplification for rapid diagnosis of MRSA from clinical specimens ³³.

Most microfluidic platforms for detection are manufactured with glass or a glass-PDMS-glass three-layer composite ³⁴. PDMS is a soft and biocompatible material and is well known for its interfacial adhesion property to assist in bonding ^{20, 35} that is easy to fabricate and is extremely precise ²⁶. Besides, the bonding process, which uses oxygen plasma treatment, is strong and reliable and has no leaking problems ³⁶.

In the present study, we employed a droplet-based segmented flow system for the optofluidic platform. Adjacent biological samples are separated in the continuous flow system by a mineral oil droplet. Deposition of nanoparticles inside the microchannel, which may adversely affect detection due to the conjugation of analytes to the optical window, creating a “memory effect” can be reduced using segmented flow ^{22, 37}. The method of Strehle et al.²³ using crystal violet as the analyte was used to check the “memory effect” of our microfluidic system. The flow was abruptly ceased at 200 s and the corresponding Raman spectrum collected for a sample of silver nanocolloid containing only water droplets and no obvious spectral features were detected (Fig. S8), indicating that the flow was appropriately segmented and the “memory effect” was minimal.

SERS implementation into microfluidic system

On-line detection techniques, including chemiluminescence and fluorescence require a tagging step ³⁸. The width of the fluorescence bands are broad and band overlapping may occur if several analytes are simultaneously determined ^18f. Raman spectroscopy provides an attractive alternative. In the present study, we used silver nanocolloid for signal enhancement from microbes flowing in the microchannel and optimized this system. Because individual single nanoparticles could only provide a faint Raman scattering signal enhancement ³⁹, aggregation by the addition of salt, such as nitrate and chloride, is required ²⁰ and incorporating silver and gold nanoparticles can amplify the weak Raman scattering signal significantly, theoretically 10¹⁴ times, placing detection of single molecule within the scope of this analytical method ^{12, 14a, 14e}. Compared to gold nanocolloid, silver nanocolloid is less stable and the size distribution is more difficult to control, resulting in lower homogeneity ^15b from various aggregation behaviors but this can be compensated for in a flow-through system by efficient mixing. However, the advantage of silver nanocolloid is that enhancement effect (factor) is 1000 times higher than that of gold nanocolloid ³⁹.

Here a higher intensity of green laser (532 nm, 10 mW, 0.5 s integration time) was introduced into the microchannel for SERS spectra collection than that employed by Walter et al.¹⁹ for SERS spectral collection (532 nm laser, 5 mW, incident 1 s integration time) for E. coli strains in a microfluidic system. Compared to fluorescence, photodecomposition is reduced for Raman spectroscopy because of rapid quenching of excited states for the analytes ^18f making rapid spectral collection possible. In addition, the fluidic condition can evenly disperse the energy from the incident laser light, which is the major factor requiring employment of a higher laser power compared to traditional Raman and SERS for this optofluidic system. Further, flow behavior was controlled, preventing sample burning.

Use of a confocal technique allowed removal of interfering scattering signals from PDMS which is a Raman active material. Each of our confocal micro-Raman spectroscopic systems (Renishaw system in China and WITec system in the United States) has an adjustable confocal slit coupled with the aperture located in the back image plane of the microscope objective lens (Fig. 1) to collect only the green laser (532 nm)-induced spectral signal from a definite focal volume (see red layer in Fig. 1), blocking diffracted light from outside the focal plane (see purple and yellow layers in Fig. 1). This confocal mode can effectively remove the background stray light derived from the out-of-focus regions of the PDMS ^{20, 35}.

Minor variations in SERS spectral features between MRSA and MSSA were observed (Fig. S3, Table S3). Spectral profiles of MRSA and MSSA differed in the mass range of m/z 500–3500 Da and cluster analysis could be constructed for visual segregation ⁴⁰ for matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectrometry. A MALDI-TOF method for phenol-soluble modulin α1 and α2 peptides as a biomarker for MRSA and vancomycin-intermediate S. aureus isolates has also been developed ⁴¹.

Timeline advantage

MRSA is the most frequently identified antibiotic-resistant pathogenic bacteria in the world and the infection rates have increased over the past decades. The prevalence of MRSA varies by country ⁴². The highest proportion of MRSA infection is in large hospitals and the transmission between patients is mainly responsible for the emerging deadly multidrug-resistant S. aureus strains ⁴³. The European Food Safety Authority has recently (October 5, 2012) published a scientific report recommending continuous monitoring and reporting program for MRSA in food-producing animals, food and feed. Improved analytical methodology for identification and typing is required for this zoonotic agent because MRSA colonization in production animals and the dissemination of this zoonosis into the food production chain is still not clearly understood ⁴⁴. Therefore, further study and surveillance of epidemiology and dissemination for MRSA, especially the development of new techniques for rapid detection of MRSA in clinical and now environmental specimens is critical to decrease infection rates acquired from hospital and community and from agricultural products, is necessary to improve environmental health.

One of the major incentives for conducting this research work was to shorten the diagnostic time. Traditional genotyping methods (e.g., PCR, PFGE and MLST) have a long turnaround time (2–3 days) and a low sample throughput for bacterial detection starting from a pure culture (Fig. 4a). Therefore, these methods are not suitable for routine uses in hospital and community surveillance. While, PCR is useful for rapid identification (~ 6 h) of MRSA or MSSA isolates from a clinical specimen including DNA extraction (1 h), PCR setup (1 h), PCR amplification (3 h) and gel electrophoresis (1 h), it does not permit determination of the source of an outbreak. Recently, recombinase polymerase amplification (RPA) has been widely conducted to rapidly detect bacteria; however, RPA cannot address the bacterial epidemiological surveillance issues. To address this concern, scientists have turned to MLST or PFGE ⁸. The time frame for MLST (or PGFE) is long, requiring DNA extraction (1 h), PCR setup (1 h), PCR amplification (3 h), PCR cleanup (1 h), sequencing (8 h), and alignment and sequence typing (5 h). An important caveat with these techniques is that they require an additional step to distinguish between MRSA and MSSA isolates (e.g., S-FF2 and S-FF18 in Table S2), that have identical sequence types.

Figure 4.

Optofluidic SERS biosensor shows advantages compared to traditional molecular typing. Panel (A): Time table to compare the detection speed of one S. aureus isolate starting from a pure culture using different methods (PCR, MLST and optofluidic-SERS). Purple color denotes DNA extraction, blue color denotes PCR setup, green color denotes PCR amplification, dark blue color denotes gel electrophoresis, yellow color denotes PCR clean up, grey color denotes DNA sequencing, orange color denotes alignment and sequence typing, red color denotes spectral collection and black color denotes data analysis. S. aureus was grown in LB broth at 37°C overnight before performing any detection method. Panel (B): Flow chart of comparing the working efficiency between optofluidic SERS typing and genotyping techniques to study bacterial epidemiology and antibiotic-resistant profiles.

The optofluidic platform described here provides a much more rapid analysis. On the basis of 25–30 mins for the generation time of MRSA ^1a, cultivation for 200 mins starting from single bacterial cell is enough (100 cells per droplet) for detection using SERS spectra. It will be important to optimize the relationship between bacterial cell density in each droplet and the SERS signal intensity and further studies with this system are needed. A decrease in cultivation time for less bacterial cell numbers necessary for accurate identification and classification based on spectral features in this optofluidic platform is proposed.

A total of 300 SERS spectra were recorded for each bacterial isolate to establish and validate the relevant chemometric model. Considering 0.5 s as the integration time for collecting each spectrum, a total of 3.3 min was needed for spectral recording, including approximate 1 min for spectral collection of mineral oil used for sample segregation which also provides an optical baseline to show that the flow and optical behavior within the optofluidic device remained stable.

Simultaneous determination of S. aureus antibiotic-resistance and epidemiological dissemination

PFGE, spa typing and MLST are the major tools to study the epidemiology and dissemination of S. aureus in the food production chain, community, and hospitals ⁴⁵. Even though most research agencies still used PFGE as the gold standard for S. aureus epidemiological studies, the application of MLST has increased significantly during recent years ⁴⁶. In addition, the detection of the mecA gene for MRSA has to be conducted individually using PCR or other related techniques, such as loop-mediated isothermal amplification. Our study indicated that optofluidic-based SERS typing can not only detect and differentiate MRSA from MSSA, but can also be used to successfully investigate the epidemiology of MRSA isolates (e.g., distinguish between isolates that were part of the outbreak and isolates that were not) in a fast turnaround time (Fig. 4b). Further, many clinical specimens may contain a mixture of Staphylococci and this may presents challenges for reliable and direct detection of MRSA from clinical specimens ⁴⁷. Our optofluidic-based SERS PLSR model could accurately determine the concentration of an MRSA strain in a mixture containing MSSA (Table S5), a finding with significant clinical application.

Finally, this method is robust. Experimental protocols were conducted in different laboratories by using two different confocal micro-Raman spectroscopic systems coupled with microfluidic platform for recording the SERS spectral features of S. aureus clinical isolates from China and the United States. We created a global dendrogram model from the data from these two laboratories and used this to monitor a recent S. aureus-associated infection outbreak (Table S4) denoting the possibility of employing this optofluidic technique as a reliable routine tool for rapid detection and epidemiological surveillance of MRSA in the world.