Integrated micro-optofluidic platform for real-time detection of airborne microorganisms

We demonstrate an integrated micro-optofluidic platform for real-time, continuous detection and quantification of airborne microorganisms. Measurements of the fluorescence and light scattering from single particles in a microfluidic channel are used to determine the total particle number concentration and the microorganism number concentration in real-time. The system performance is examined by evaluating standard particle measurements with various sample flow rates and the ratios of fluorescent to non-fluorescent particles. To apply this method to real-time detection of airborne microorganisms, airborne Escherichia coli, Bacillus subtilis, and Staphylococcus epidermidis cells were introduced into the micro-optofluidic platform via bioaerosol generation, and a liquid-type particle collection setup was used. We demonstrate successful discrimination of SYTO82-dyed fluorescent bacterial cells from other residue particles in a continuous and real-time manner. In comparison with traditional microscopy cell counting and colony culture methods, this micro-optofluidic platform is not only more accurate in terms of the detection efficiency for airborne microorganisms but it also provides additional information on the total particle number concentration.

may not satisfy the specific growth requirements of all species of interest, and also requires prolonged incubation (i.e., > 24 hours) 16 . Recently, detection and enumeration of airborne microorganisms has been accomplished using techniques based on polymerase chain reaction (PCR) and enzyme-linked immunosorbent assays (ELISA), which are highly sensitive and quantitative techniques [17][18][19][20][21] . However, additional pretreatment processes (such as particle condensation/purification), and elaborate sample handling by well-trained operators in a clean environment are required 22 . Although these PCR and ELISA techniques have been applied to microfluidic chips to simplify the treatment processes, [23][24][25] most such microfluidic systems have been developed as disposable chips for point-of-care diagnosis of target microorganisms, which often leads to a limited number of samples or poor accuracy or rates of detection 26 .
Novel real-time methods for the rapid detection and continuous monitoring of airborne microorganisms have the potential to address the public health problems associated with bioaerosols. One of the most frequently used real-time detection techniques for detection of airborne microorganisms exploits auto-fluorescence following illumination with ultraviolet (UV) light 27,28 . Auto-fluorescence is caused by metabolites and structural components of living cells 29 . Although this technique allows continuous real-time monitoring and detection of bioaerosols directly in the air stream, problems remain due to low fluorescence intensity, which leads to poor limits of detection, and requires precise optical systems for the measurements. For these reasons, such devices are typically not portable, and cannot provide an integrated "micro-total analysis system".
To overcome the limitations discussed above, cell and particle detection methods based on microfluidic flow-cytometry have been reported [30][31][32][33] . Microfluidics benefits from miniaturization, low power consumption and high sensitivity. Microfluidic methods of counting water-borne microorganisms include tracking the population of marine algae and quantification of bacterial cells using pre-staining [34][35][36] . Chung et al. (2014) developed a real-time single-cell detection system using target aptamer-conjugated fluorescent nanoparticles in a microfluidic flow-cytometry platform for microbial diagnostic applications 37 . However, there have been no reports of micro-optofluidic platforms for real-time continuous detection and monitoring of airborne microorganisms.
Here, we demonstrate a micro-optofluidic platform for real-time detection and quantitative analysis of airborne microorganisms. Our optofluidic system involves the following steps: (1) sampling of airborne microorganisms; (2) mixing and reacting in a microchannel for staining; and (3) real-time detection and analysis of the particle by means of light scattering (SC) and bacterial fluorescence (FL). From these optical signals, we may discriminate and quantify airborne microorganisms, enabling simultaneous measurement of the total particle concentration. The performance in terms of particle detection is evaluated using standard particles and three bacterial taxa, and is compared with that of conventional microscopy cell counting and colony counting methods.

Results
Design and operation of the micro-optofluidic platform. Figure 1 shows a schematic diagram of the micro-optofluidic platform. The 150-μ m-deep polydimethylsiloxane (PDMS) microfluidic channel was fabricated using conventional soft-lithography processes 38 . The micro-optofluidic platform is based on single-layer PDMS channel and has four inlets (one sample inlet, one dye inlet, and two hydraulic focusing sheath inlets), and one outlet for waste. It consists of four main components: sample and dye inlets, a mixing region, a hydraulic focusing zone, and a sample optofluidic detection component (see Fig. S1). We used SYTO82 dye (Molecular Probes Inc., Eugene, OR, USA) with an excitation wavelength of 541 nm and an emission wavelength of 560 nm for nuclear staining of live cells in the microfluidic mixing zone. SYTO82 provides cell-permeant nucleic acid staining and exhibits low intrinsic fluorescence in cell-free systems, with marked enhancement upon binding to deoxyribonucleic acid (DNA) or ribonucleic acid (RNA) 39,40 . Following injection of the sample and dye using a syringe pump (KDS200; KD Scientific Inc., Holliston, MA, USA), the two fluids mixed rapidly at the entrance to the mixing zone.
Numerical analysis was carried out using the commercial computational fluid dynamics (CFD) software package CFD-ACE (ESI US R&D Inc., Huntsville, AL, USA) to identify the optimal design and operating conditions. The Navier-Stokes equations were solved with Fick's law of diffusion to describe the dynamics and homogenization of the mixing process in the channel. Figure 1 (B-1,B-2) show the results of the CFD simulations and optical micrographs of the mixing of the sample and dye in the micro-mixer part. The numerical simulations show that the mixing of the two fluids was complete approximately 0.1-0.4 mm from the entrance of mixing zone with flow rate conditions for each fluid in the range 2.5-25 μ L/h (simulation duration, ~72-180 ms) (Fig. S2). The mixing zone was 136.8 mm long and 50 μ m wide. The residence time for mixing (and hence for the reaction between the sample and dye) was in the range 25-246 s for total flow rates in the range 5-50 μ L/h. After passing through the mixing zone, the mixed sample fluid was focused horizontally using a dual sheath flow with sterilized distilled water (SDW) in the hydraulic focusing component (see Fig. 1 (A)). The focused core flow was 7.5 μ m wide, which is one-twentieth the width of the channel. This minimized variation in the optical signals among locations in the particle detection zone 30 . Figure 1 (C-1) shows a schematic diagram of real-time detection of target particles in the core flow, as well as the optical setup for continuous detection and signal quantification. A diode-pumped solid-state (DPSS) laser with a wavelength of 532 nm and an output power of 2.0 mW (CNI laser, Changchun, China) was coupled with a single-mode optical fiber (core diameter: 3.5 μ m, cladding diameter: 125 μ m).   The laser beam was focused on the core fluid using an objective lens (RMS20x-PF; 20x/0.5; Thorlabs) located beneath the microfluidic channel (~7 μ m beam spot size at full-width half-maximum). A highly sensitive multi-pixel photon counter (MPPC) (C10507-11-50U; Hamamatsu Photonics, Hamamatsu City, Japan) module was used to detect the SC and FL signals from the target particles. The particle FL was collected using the objective lens, and entered the MPPC via the optical long-pass filter and focusing lens. To bring the SC light from the particles, multi-mode optical fiber with a core diameter of 105 μ m and a cladding diameter of 125 μ m (Thorlabs) was carefully inserted into the channel with the axis of the fiber at 90° to the direction of flow, and linked to the MPPC.
A data acquisition (DAQ) unit (NI USB-6356; National Instruments, Austin, TX, USA) was employed to collect the analog data from the MPPC modules, which were recorded using a desktop computer. An oscilloscope (DPO 2012B; Tektronix, Beaverton, USA) was used to monitor the signal in real-time, as shown in Fig. 1 Performance evaluation of the micro-optofluidic platform using standard PSL particles. The particle detection performance of the optofluidic platform was evaluated under various microfluidic and sample conditions, such that the sample flow rate and the mixing ratios of FL and non-FL particles were varied. We used standard spherical polystyrene latex (PSL) and fluorescence PSL (FLPSL) particles that were 1 μ m in diameter. Figure 2 shows the particle FL and SC signals with flow velocities in the range 1.42-14.3 mm/s. The FLPSL particle suspension (~1.45 × 10 3 /μ L) was injected into sample inlet. The ratio of the flow rate of the FLPSL particle suspension to that of the SDW was 1:19 in the narrow core region (which had dimensions of 7.5 × 150 μ m). As the flow velocity increased from 1.42 to 14.3 mm/s, the frequency of the particle SC and FL signals increased in proportion to the particle flow velocity, and the average intensity of the measured SC and FL signals (SC/FL) decreased from ~140 arbitrary unit (a.u.)/~60 a.u. to ~46 a.u./~30 a.u., respectively ( Fig. 2 (A)). This is attributed to the decreased photon integration time of the photodetector with decreasing residence time of the particles in the detection zone of microchannel 41 . Figure 2 (B) shows that the event rate (i.e., number of counts per second) of the particle SC and FL signals followed a linear relationship with the particle flow velocity; for FL we have y = 1.04x + 0.37, with r 2 = 0.996, and for SC we have y = 1.07x + 0.52, with r 2 = 0.998. The index of coincidence between the total frequencies of particle SC and FL signals was ~96.5% over the entire range of particle flow velocities.
McClain et al. reported that only ~5% of the particle SC signals were unaccompanied by coincident FL signals, which was attributed to unresolved FL peaks due to the relatively low signal intensity 36 . When the number of counts of the particle SC and FL signals are plotted as histograms as a function of the signal intensity, both sets of signals form Gaussian-like distributions, as shown in Fig. S3. The variation in the intensity results from variations in the particle size and particle position in the detection zone. As expected, the particle SC signal was significantly stronger than the FL signal.
We evaluated the particle detection performance of the micro-optofluidic platform with various mixing ratios of FLPSL and PSL particles. The total particle concentration was fixed at ~1.45 × 10 3 /μ L, and the total flow rate of the mixed suspension was 25 μ L/h, giving a flow velocity of ~7.1 mm/s. The mixing ratio of FLPSL and PSL particle suspensions was varied in the range 0 -100%. As shown in Fig. 3 (A), the ratio of the count rate of FL signals to SC signals increased in proportion to the mixing ratio of FLPSL particles. Figure 3 (B) shows fluorescence microscopy images of the FLPSL particles in the mixing suspension. When the mixing ratio of FLPSL particles was 0%, the FLPSL particle number concentration in the micrographs was 0/μ L; at a mixing ratio of 50%, the FLPSL particle number concentration was 688/μ L (~95% of the initial FLPSL particle concentration); and at a mixing ratio of 100%, the FLPSL particle number concentration was 1406/μ L (~97% of the initial FLPSL particle concentration). PSL particles could not be enumerated using normal bright-field microscopy because of the limit of resolution and small field of view.
From this performance evaluation (Figs 2 and 3), we may conclude that the micro-optofluidic platform demonstrated not only real-time and continuous particle detection, with suitable performance for standard PSL particles, but also quantitative and accurate discrimination between FL and non-FL particles.
Real-time detection of airborne microorganisms. To simulate a hazardous airborne bacteria-contaminated environment, we prepared a 1 × 1 × 1 m test chamber containing live, airborne Escherichia coli, Bacillus subtilis and Staphylococcus epidermidis bioaerosols separately. Details of the experimental setup can be found in the Methods and Supplemental Information (Fig. S4). Figure 4 shows the size and morphology of the bacterial bioaerosols in the test chamber. Real-time particle size distribution data of the bacterial bioaerosols were obtained using an aerodynamic particle sizer (APS; 3321; TSI Inc., Shoreview, MN, USA). As shown in Fig. 4 (A), the bacterial bioaerosols exhibited mono-modal curves with a specific geometric mean diameter (GMD), peak diameter, and geometric standard deviation (GSD). Table 1 lists these data for the three bioaerosols; the GMDs of E. coli and B. subtilis were similar at ~1 μ m, and that of S. epidermidis was somewhat smaller, at ~0.8 μ m.
Airborne bacterial bioaerosols in the test chamber were collected using a BioSampler (SKC Inc., Eighty Four, PA, USA), which is a highly efficient collection device that traps airborne microorganisms in a swirling liquid for subsequent analysis 42 . The airborne particle collection efficiency of the BioSampler was ~94.3% for the 1-μ m-diameter PSL particles (Fig. S5). Figure 4  Samples containing the bacterial suspension were injected into the micro-optofluidic platform using a syringe pump to evaluate the device performance for the real-time detection of airborne bacterial particles. Figure 5 (A) shows the acquired SC and FL signals of the test bacterial bioaerosols. The average SC/FL signal intensity ratios were ~555/240 for E. coli, ~565/227 for B. subtilis, and ~433/142 for S. epidermidis. Because of the size of these bacteria (i.e., 0.5 -1.6 μ m), they may be considered to lie within the Mie scattering regime, and the intensity of the SC signals may be assumed to be proportional to the cross-sectional area of particles 36 . In the same manner, the FL signal intensity increased proportionally with the size of the bacteria because larger particles have a higher concentration of nucleic-acid fluorescent dye 33,36 . The particle number concentration was determined based on the frequency data for the SC and FL signals. Table 2 lists the SC signal concentration, which corresponds to the total particle number concentration. This was higher than the FL signal concentration, which corresponds to the total concentration of microorganisms. The difference between the total particle count (SC signals) and the total bacterial concentration (FL signals) was ~7% for E. coli, ~16% for B. subtilis, and ~23% for S. epidermidis. It follows that the sample included other particles; i.e., non-microorganism particles or impurities, such as non-dyed particles or small debris.
To evaluate the particle detection efficiency for airborne microorganisms, the total concentration of bacteria obtained from the FL signal of the micro-optofluidic platform was compared with the conventional fluorescence microscopy cell counting and colony counting methods. Figure 5 (B) shows a comparison of the concentration of bacteria measured using fluorescence microscopy and colony counting, both of which were normalized to that measured using the micro-optofluidic platform. For E. coli the normalized cell concentration by microscopic cell counting was 88 ± 4.6%, and was 67 ± 8.5% using colony counting; for B. subtilis, the normalized cell concentration using microscopic cell counting was 88 ± 4.7%, and was 77 ± 6.6% using colony counting; and for S. epidermidis the normalized cell concentration using microscopic cell counting was 85 ± 11.1%, and was 73 ± 12.4% using colony counting. The concentration of bacteria measured using both conventional cell-counting methods was lower than that using the micro-optofluidic platform.
The micro-optofluidic platform exhibited the highest cell counts and the lowest standard deviation, which is indicative of superior performance. Photobleaching of fluorophores may occur during cell counting via conventional fluorescence microscopy, which results in a reduction in the fluorescence intensity. Furthermore, the low signal-to-noise ratio in the imaging process may lead to underestimation of the number of cells. An inappropriate image threshold intensity setting is required to discriminate fluorescence particles from the background, and this may exclude particles that either have a low fluorescence intensity or are slightly out of focus 43 . Furthermore, quantitative analysis of the total concentration of particles with sizes of less than 1 μ m is limited by the resolution and the field of view of bright-field microscopy. As shown in Figure 5 (B), the colony counting method resulted in the lowest cell concentration of the three methods. The colony counting method measures only culturable cells in the medium. Therefore, it is difficult to apply the colony counting method for the analysis of viable but non-culturable (VBNC) microorganisms, or microorganisms that require a specific growth environment and/or specific nutrients 44 .
The micro-optofluidic platform with integrated sample preparation and detection simplifies the system, can significantly reduce the measurement time, and can yield more accurate quantitative results  48 . The number concentration of the particles in air was recorded for each APS channel size, divided by the logarithmic interval of the corresponding particle size range and plotted as a function of the aerodynamic diameter 48 Table 1. Size characteristics of test airborne bacterial particles. The geometric standard deviation (GSD) is defined as exp( where d j is the diameter of an individual particle, n j is the number of particles in the j th group, N is the total number of particles, and ln d g is the natural logarithm of the geometric mean diameter (GMD) of the particles, defined as ∑ / n d N ln compared with the abovementioned conventional methods. Furthermore, our system provides additional information on the total particle number concentration of aerosols, which is difficult to obtain using conventional methods.

Discussion
We have demonstrated continuous, rapid and real-time detection of bioaerosols using a micro-optofluidic platform. The performance of our device was investigated using standard PSL particles with various flow speeds and mixing ratios with FLPSL particles. Accurate quantitative FLPSL particle discrimination with high efficiency was achieved. This first application of this integrated micro-optofluidic platform for the analysis of airborne microorganisms showed that our device could reduce the time for sample preparation and manual analysis compared with conventional microorganism-counting methods. The micro-optofluidic platform has potential applications in aerosol analysis, and can enable portable, highly sensitive, continuous real-time detection of airborne particles and microorganisms.
In future, we plan to include a 3D focusing stream to optimize the optical signals, which is expected to enhance the resolution and efficiency of the measurements, and to reduce the noise due to spatial deviations of the particles. Accurate particle sizing using the current micro-optofluidic platform is possible 30 ; furthermore, a micro-pump between the BioSampler and microfluidic channel could be included for continuous real-time analysis of airborne microorganisms. Various species of dye or marker for specific target materials, such as target-specific antibodies or aptamers, could be used to create an analysis system that provides data on the physical, chemical and biological properties of the airborne particles 37 . Note that the purpose of this work was to quantitate airborne microorganisms rapidly and in real-time using a fully integrated micro-optofluidic platform.

Methods
Sample preparation. Two types of standard uniform particles and three types of bacteria were used in this study. Fluorescence polystyrene latex (FLPSL) particles (F8820; Fluorescent microsphere; 1.0-μ m diameter; orange fluorescent (540/560); Invitrogen) and polystyrene latex (PSL) particles (4010A; Monosized microsphere; 1.0-μ m diameter; refractive index of 1.59; density of 1.06 g/cm 3 ; Duke Scientific Corporation) were used as standard particles for the performance evaluation of particle detection and quantification. As test airborne microorganisms, Gram-negative Escherichia coli (Korean Collection of Type Cultures (KCTC) 1039, Biological Resource Center, Republic of Korea), Gram-positive Bacillus subtilis (KCTC 1022), and Gram-positive Staphylococcus epidermidis (ATCC 12228), were used 45 . The bacteria were incubated in nutrient broth (Becton Dickinson, Franklin Lakes, USA) at 37 °C, and harvested using a centrifuge (Mini, Gyrozen, South Korea) at 6000 rpm for 10 minutes. Bacterial pellets were washed three times using SDW with a centrifuge to remove the residual medium. To create the bacteria suspensions, 30-mL aliquots were placed in a six-jet collision nebulizer (BGI Corp., USA) to aerosolize the bacteria.
Microchannel design and fabrication. The microchannel was a single-layer PDMS channel fabricated using conventional soft lithography. SU-8 (2025; Microchem Corp.), which is a negative photoresist, was spin-coated onto a 4.2-inch Si wafer. The photoresist was soft baked on a hotplate at 65 °C for 5 min, followed by 95 °C for 10 min. A chrome photomask pattern was installed using a UV aligner, and the photoresist was exposed to UV irradiation. Following this exposure, the resist was baked at 65 °C for 2 min and then at 95 °C for 10 min. The SU-8 photoresist pattern was developed using 1-methoxy-2-propyl acetate (Microchem Corp.) for 7 min. After completion of the developing process, the pattern was rinsed using isopropyl alcohol (IPA) and deionized (DI) water to yield a master mold. PDMS was poured over the developed wafer with the pattern, and cured in an oven at 65°C. The patterned PDMS channel was then removed from the wafer, and bonded onto glass using O 2 plasma to seal the microchannel. Fluorescence microscopy cell counting. To enumerate FLPSL particles and bacterial cells using the fluorescence microscopy method, 10-20 μ L of sample (following enrichment via centrifugation in the case of low particle concentrations) was loaded into the sample injection area of a disposable hemocytometer (DHC-N01; INCYTO, Republic of Korea), and visualized using a fluorescence microscope (B X 51; Olympus, Tokyo, Japan) with a U-MWG2 filter set; the excitation wavelength was in the range 510-550 nm, and the emission wavelength was > 590 nm. For each sample, images of at least 16 microscopic fields were captured using a CCD array camera. Microscopy cell counting was carried out using the ImageJ software package (http://imagej.nih.gov/ij/). Colony counting. Bacterial suspensions were serially diluted and spread onto the surface of nutrient agar (Becton Dickinson) in a petri dish, followed by incubation at 37°C for 24 hours. The resulting colonies were counted manually.
Scanning electron microscopy. The morphology of airborne bacterial particles was investigated using SEM (Nova Nano SEM 200; FEI Co., Hillsboro, OR, USA). The modified Karnovsky's fixation protocol was used 46,47 , and bacterial samples were fixed using 2% paraformaldehyde (15713; Electron Scientific RepoRts | 5:15983 | DOi: 10.1038/srep15983 Microscopy Sciences (EMS), Hatfield, PA, USA) and 2% glutaraldehyde (16020; EMS) in a 0.05 M sodium cacodylate buffer (SCB; pH 7.2) (11652; EMS) at 4°C for 2 hours, followed by washing with 0.05 M SCB at 4°C for 2 hours. Post-fixation, the samples were treated with 1% osmium tetroxide (19152; EMS) in 0.05 M SCB at 4°C for 2 hours, and then washed with distilled water at room temperature. The samples were dehydrated at room temperature in a 30-100% ethanol series with 10-minute exposure to each concentration. The ethanol was then replaced with hexamethyldisilazane (16700; EMS). Following dehydration, each sample was spread onto polycarbonate membrane filters with a pore size of 0.4 μ m (Isopore Membrane Filters HTTP01300; Millipore, Billerica, MA, USA) and coated with osmium by chemical vapor deposition (HPC-1SW; Vacuum Device Inc., Ibaraki, Japan).