Introduction

Foodborne diseases commonly caused by pathogenic bacteria are vital concerns to public health worldwide. Because bacteria proliferate over time, sensitive detection of bacteria at early stage is critical for preventing food-borne diseases1. Conventional detection methods based on microbial cultivation are accurate and reliable and considered to be golden standard methods. However, they are time-consuming and labor-intensive, so their application is limited to laboratory measurements2. Several methods for rapid detection of bacteria without cultivation have been reported including polymerase chain reaction (PCR)3,4, quartz crystal microbalance (QCM)5,6, surface plasmon resonance (SPR)7,8, electrochemical impedance spectroscopy (EIS)9, surface-enhanced Raman scattering (SERS)10,11 and fluorescence spectroscopy12. However, they still require complex pretreatment procedures to separate bacteria from food matrices.

This need for complex pretreatment may be mitigated by adopting immunomagnetic assays that use antibody-functionalized magnetic nanoparticles to capture and separate bacteria from food matrices under an external magnetic field. To determine the concentration of bacteria, conventional immunomagnetic assays label captured bacteria with fluorescent molecules or quantum dots to distinguish the bacteria-magnetic nanoparticle complexes from free magnetic particles13,14,15. Size-based separation techniques are good alternative approaches which do not require a complicated labeling procedure. For instance, larger bacteria-magnetic nanoparticle complexes are easily separated from smaller free magnetic nanoparticles using filter membranes16. However, filtration has a drawback of a high background noise due to inefficient separation of free magnetic nanoparticles, which degrades the detection sensitivity.

Instead of the use of filter membranes, size-based microfluidic separation methods have been reported for isolation of red blood cells17, circulating tumor cells18 and microparticles19. Among the microfluidic separation methods such as surface acoustic waves20, inertial focusing21 and deterministic lateral displacement22, the inertial focusing method based on Dean drag force has attracted much attentions because it is easy to control the operation condition with no external force and small chance of physically damaging the cells during separation. Cells or particles can be separated by inertial focusing using spiral microchannels fabricated in a two-dimensional PDMS substrate23,24,25. However, in a spiral channel on a flat substrate, the radius of curvature changes; therefore its Dean number changes, so the flow behavior is difficult to predict and separation is hard to control.

In this paper, we used stereolithography26,27 to fabricate a helical microchannel around a cylindrical chamber. The vertically designed device offers a constant radius of curvature and compact size. We applied the 3D-printed microfluidic device for rapid and facile detection of E. coli (EC) bacteria in a real food matrix. After the capture of E. coli bacteria in milk using antibody-functionalized magnetic nanoparticle clusters (MNCs), the free MNCs and MNC-EC complexes were separated using the 3D helical microchannel device. Combined with UV-vis spectroscopy, our method could detect the presence of pathogenic bacteria in 10 mL of a milk sample with a sensitivity of 100 cfu/mL (colony-forming units per mL). The separation was verified by using dynamic light scattering (DLS) and ATP luminescence measurements. Selectivity of the assay was tested against Salmonella typhimurium and Staphylococcus aureus bacteria and we verified that this detection method can capture and isolate desired target only.

Results and Discussion

Characterization of the MNCs and MNC-EC complexes

Figure 1(a) and 1(b) show a scanning electron microscopy (SEM) image of free MNCs and a transmission electron microscopy (TEM) image of a MNC-EC complex, respectively. The average size of the MNCs was approximately 150 nm and each consisted of a few hundred 15 nm Fe3O4 nanoparticles. The large size of MNC promotes more effective magnetic separation from the analyte compared to small Fe3O4 nanoparticles because the magnetic force experienced is proportional to the volume of a particle. An E. coli bacterium is about one order of magnitude larger than an MNC and this size difference is the main driving force to separate EC-MNC complexes from free MNCs under the conditions of helical flow.

Figure 1
figure 1

(a) SEM image of MNCs, (b) TEM image of an MNC-EC complex.

Design principle and characterization of the MNCs

Figure 2(a) shows the 3D CAD design of the device and the separation scheme for isolating MNC-EC complexes. A sample solution containing free MNCs and MNC-EC complexes is injected into the outer inlet of the device. A sheath flow, if needed, is injected into the inner inlet of the device. When a fluid passes through a curved microchannel, secondary flow that consist of two vortices is generated (Figure 2(b)). These vortices are known as Dean vortices and their magnitude can be expressed by using the dimensionless Dean number (De) and Dean drag force (FD)28,

where ρ is density, Uf is fluid velocity, Dh is hydraulic diameter, μ is dynamic viscosity, R is the radius of curvature of the channel and ap is the hydrodynamic diameter. Because De is inversely proportional to the radius of curvature and because the magnitude of FD experienced by the particle increases with the increasing ap and De, separation of particles with two different sizes can be facilitated by increasing the size difference between particles and by using a helical channel with small radius of curvature. In addition to FD parallel to the Dean flow direction, the particles in a fluid flow also experience a lift force FL perpendicular to the primary flow direction. FL can be expressed as a function of particle size and particle position across the channel cross-section29.

where G is the fluid shear rate and CL (assumed to be 0.5)21 is the lift coefficient as a function of the particle position across the channel cross-section. Because the MNC-EC complexes are one order of magnitude larger than the free MNCs, FD and FL have significantly stronger effect on MNC-EC complexes than on free MNCs, enabling separation of MNC-EC complexes from free MNCs. The velocity of the fluid varies within the channel cross-section, so the magnitude and direction of both FD and FL also vary as a function of the particle's relative position. As a result, large particles become focused at the equilibrium position near the inner wall of the microchannel where FD and FL are balanced.

Figure 2
figure 2

(a) Schematic illustration of separation of captured bacteria by inertial focusing. (b) Illustration of Dean vortices in a channel with trapezoid cross-section. (c) Photograph of the 3D printed device.

To enhance the effectiveness of separation, the microchannel was designed to have a trapezoidal cross-section and an additional inlet was added to the device to allow injection of a sheath flow. The asymmetric nature of the trapezoidal channel cross-section (Fig 2(b)) induces strong Dean vortex cores to form close to the outer wall and traps smaller particles within the cores30. The sheath flow was used to induce initial focusing of both free MNCs and MNC-EC complexes towards the outer wall of the microchannel. After the sample was passed through the helical microchannel, the size-sorting effect of FD and FL caused only the MNC-EC complexes to migrate towards the inner wall. Figure 2(c) shows the photographic image of 3D-printed microfluidic device used in this study. The fabricated device incorporated 10 loops of helical microchannel to provide sufficient length needed for particle migration31. The width of the trapezoidal channel cross-section was 1000 µm and the inner and outer heights were 250 µm and 500 µm, respectively. Such dimensions were determined by the minimum resolution of the 3D-printer used for fabrication of the device and the experimental results exhibited satisfying performance of the device. The inlet and outlet are in the same plane. However, the inlet is connected to the top of the helix with a vertical channel. The reason that the inlet is not in above plane is to avoid overhang structure which is not favorable for 3D-printing. The inlets and outlets were connected with tubing by barbed luer lock connectors to provide easy tight seal and prevent leakage.

MNC separation performance of the device

The MNC separation performance of the microfluidic device with or without the sheath flow was characterized using DLS measurements. To evaluate the effectiveness of separation without the sheath flow, only one inlet was used; the other inlet was sealed with a luer cap. A sample containing free MNCs and MNC-EC complexes (obtained from milk spiked with 105 cfu/mL E. coli) in 5 mL of buffer solution was introduced into the separator at the flow rate of 10 mL/min and the size distribution of the collected samples from each outlet was measured using DLS.

Figure 3(a) shows the number mean size distribution of the sample solution before passing through the device. The two peaks at 150 nm and 1020 nm correspond to free MNCs and MNC-EC complexes, respectively. The peak amplitude at 150 nm is bigger than the one at 1020 nm because the concentration of free MNCs are much higher than that of MNC-EC complexes. The peak at 1020 becomes more conspicuous in the intensity mean size distribution shown in the inset of Figure 3(a) because the scattering intensity is proportional to the sixth power of the particle diameter. After the inlet solution was passed through the microchannel and divided into two outlet solutions, the size distribution curves were obtained for both the inner outlet and outer outlet solutions. The sample solution from the outer outlet showed only one peak at 150 nm (Figure 3(b)), which corresponds to free MNCs and indicates that no MNC-EC complexes were present in this solution. However, the inner outlet solution showed two peaks (Figure 3(c)), which implied that the separation was not successful and both free MNCs and MNC-EC complexes remained in the solution. The incomplete separation of particles was attributed to the fact that the free MNCs that were initially distributed randomly throughout the entire inlet area were too small to be migrated by Dean drag force under the experimental conditions.

Figure 3
figure 3

DLS number mean size distribution curves of the solution (a) at the inlet, (b) at the outer outlet and (c) at the inner outlet in case of no sheath flow input.

Intensity mean size distribution curves are shown as insets. No separation of particles was observed.

To improve the separation, a sheath flow was introduced into the inner inlet of the separator. The sheath flow pushes the particles to the outer wall side of the microchannel and helps their trapping in the strong Dean vortex cores. The solution obtained from E. coli-spiked buffer solution (105 cfu/mL) was injected into the outer inlet of the separator at 5 mL/min and the sheath flow was injected into the inner inlet of the separator at 5 mL/min (10 mL/min in total). The sheath flow was injected at the same flow rate as the sample flow because the outlet is bifurcated evenly; slower sheath flow would not push the particles towards outer wall beyond the center line of the channel and faster sheath flow would significantly dilute the sample. In contrast to the incomplete separation without the sheath flow, complete separation was achieved when the sheath flow was provided. The single peak in Figure 4(a) and 4(b) obtained from the inner outlet and outer outlet solutions corresponds to MNC-EC complexes and free MNCs, respectively.

Figure 4
figure 4

DLS number mean size distribution curves of the solution (a) at the inner outlet and (b) at the outer outlet.

Intensity mean size distribution curves are shown as insets.

Detection of E. coli

After separation of the MNC-EC complexes from the free MNCs, the concentration of E. coli was measured using a portable UV-vis spectrometer. Figure 5(a) and 5(b) show the light absorption spectra for the inner outlet solutions after the injection of various concentrations of E. coli in buffer and milk, respectively. The light absorbance intensity increased with the concentration of E. coli in both buffer and milk solutions. The peak absorbance intensity was observed at 484.9 nm. Figure 5(c) and 5(d) show variations in the light absorbance intensity at 484.9 nm with the concentration of E. coli buffer and milk, respectively. The detection limit was found to be 10 cfu/mL for buffer samples and decreased to 100 cfu/mL for milk samples due to the less efficient capture process by the presence of interferents.

Figure 5
figure 5

UV-Visible absorbance spectra of the inner outlet MNC-EC samples having a range of concentrations in (a) buffer and (b) milk samples. The spectrum from top to bottom represents E. coli concentrations of 105 cfu/mL (black), 104 cfu/mL (red), 103 cfu/mL (green), 102 cfu/mL (blue), 10 cfu/mL (brown) and 0 cfu/mL (orange). Variations in the light absorption intensities of the solution at 484.9 nm for (c) buffer and (d) milk samples.

Verification of MNC-EC complexes using a portable ATP luminometer

An ATP luminometer was used to verify that the separated samples obtained at the inner outlet were MNC-EC complexes. Because luminescence emission occurs only in the presence of living organisms containing ATP, the intensity of luminescence is directly related to the concentration of E. coli. After separating MNC-EC from sample solutions with various concentrations of E. coli bacteria in milk, the ATP luminescence for samples from each outlet was measured. Figure 6(a) shows that the luminescence signal was detected only from the inner outlet, confirming that the separation was successful. The limit of detection was found to be better than 100 cfu/mL, which is comparable to the results from light absorption measurements. The selectivity of the detection assay was examined using control samples of 10 mL milk solution spiked with 105 cfu/mL Salmonella bacteria or Staphylococcus bacteria. After incubation of E. coli antibody-functionalized MNCs in the control sample, the detection assay was conducted and the luminescence intensity was measured. As shown in Figure 6(b), the signal intensities from the samples containing Salmonella bacteria or Staphylococcus bacteria were lower than the noise level, indicating that nonspecific binding was negligible.

Figure 6
figure 6

(a) The intensities of the ATP luminescence measured for the inner outlet (red) and outer outlet (green) solutions after the injection of a milk solution spiked with different concentrations of E. coli bacteria. Concentrations of the control samples that are not spiked with bacteria were labeled “0” for convenience. (b) The intensities of the ATP luminescence, measured for the inner outlet solution after the injection of different bacteria at concentration of 105 cfu/mL at the inner outlet.

Conclusions

In summary, we have demonstrated a facile method to detect pathogenic bacteria using an immunoassay in combination with a 3D-printed helical microchannel device. After E. coli bacteria in milk was captured using antibody-functionalized MNCs, the free MNCs and MNC-EC complexes were separated by the 3D-printed microfluidic device based on the size difference between the particles. The detection limit of the assay for E. coli bacteria-spiked sample solutions was evaluated by UV-Vis absorption spectroscopy and found to be 10 cfu/mL in buffer and 100 cfu/mL in milk. Further, incorporation of 3D-printing technology enabled easy development of a microfluidic device with complex structures to facilitate the effective and efficient detection of pathogenic bacteria on site.

Methods

Materials

Sodium citrate, polyacrylamide, Iron(III) chloride hexahydrate, urea, 3-amino-propyltriethoxysilane (APTES), glutaraldehyde and Tween 20 were purchased from Aldrich (St. Louis, Missouri) and were used without further purification. Deionized water (18.3 MΩ cm−1) was obtained using a reverse osmosis water purification system and was used to prepare the phosphate buffer (PB) solution. Monoclonal Salmonella antibody and goat anti-mouse IgG antibody were purchased from Abcam Inc. (Cambridge, Massachusetts).

Synthesis of antibody-functionalized magnetic nanoparticle clusters (MNCs)

Fe3O4 magnetic nanoparticle clusters were synthesized using a hydrothermal method, as described elsewhere32. Briefly, 4 mM FeCl3, 8 mM sodium citrate and 12 mM urea were added to 80 mL DI water. After mixing the solution for 1 min, 0.6 mM polyacrylamide was added under vigorous stirring at room temperature. The solution was gradually heated to 200°C in a Teflon-lined autoclave and maintained for 10 h. The resulting MNCs were magnetically separated and rinsed a few times with DI water and absolute ethanol. The surfaces of MNCs were treated with 1% APTES in ethanol and 0.5% glutaraldehyde in water, then 10 μg of goat anti-mouse IgG antibody was immobilized onto the MNCs as a linker to produce oriented immobilization of the Salmonella antibody. After passivating the surface of MNCs with 1 wt% Tween 20 in PB, 10 μg of the Salmonella antibody was immobilized.

Fabrication of the microfluidic device

The three-dimensional microfluidic device containing a helical microchannel with trapezoidal cross-section was fabricated using stereolithography as described elsewhere26,33. The device design was drafted using a 3D CAD software, Autodesk Inventor (San Rafael, California). The inner and outer heights of the channel with trapezoidal cross-section were designed to be 250 µm and 500 µm, respectively. The width of the channel cross-section was 1000 µm. The device consists of two inlets, two outlets and a 10-loop helical microchannel with a radius of 1.5 cm. The pitch of the helical channel is 1 mm. The device was 3D-printed by FineLine Prototyping, Inc. (Raleigh, North Carolina) using a 3D Systems Viper SL system (Rock Hill, South Carolina) in High-Resolution Mode with a natural finish. DSM Somos® WaterShed XC 11122 (Heerlen, Netherlands) was selected as the resin due to its transparency and water resistance.

Capture and separation of E. coli bacteria

E. coli-spiked samples were prepared at concentrations in the range of 10 to 105 cfu/mL in either PB buffer or milk; 100 μL of antibody-functionalized MNCs (100 μg/mL) were added to 10 mL of E. coli-spiked samples and incubated for 1 h at room temperature with gentle shaking. The MNC-EC complexes and free MNCs were magnetically separated using a permanent magnet, rinsed with buffer solution, then resuspended in 5 mL of PB buffer. The solution was injected into the microfluidic device with or without a sheath flow for separation of MNC-EC complexes. The flow rates for the sample inlet or buffer sheath flow were both fixed at 5 mL/min. The actual device has built-in luer lock connector that is not shown in Figure 2(a) for better presentation purposes. Syringe pumps were used to introduce constant flow in the channel through the luer lock connection and tubing.

Size distribution measurements

Samples obtained from the outlets were magnetically concentrated and resuspended in 1 mL PB buffer solution for DLS measurements. Size distributions of free MNCs and MNC-EC complexes were determined using DLS measurements. Scattered light in the sample was detected and analyzed by using a Zetasizer (Malvern Instruments Ltd., England) with a He-Ne laser at a detection angle of 173°.

Bioluminescence measurements

The samples of MNC-EC complexes obtained from the inner outlet and outer outlet of the microfluidic device were magnetically concentrated to 50 μL and added to 150 μL of benzalkonium chloride solution to extract ATP from E. coli bacteria29. The ATP-extracted solution was sequentially added to lyophilized luciferin and luciferase powder and the luminescence intensity was measured using a portable luminometer (Kikkoman PD-20). The photochemical reaction mechanism is described elsewhere34,35. In brief, luciferase catalyzed the adenylation of luciferin by ATP to form an intermediate, luciferyl adenylate, which is instantly oxidized by oxygen in solution to generate excited oxyluciferin. Finally, the excited oxyluciferin emits light while decaying to a stable state. Each ATP luminescence measurement was completed within 1 min.