1. Introduction

Early diagnosis of viral infection and disease control have always been critical. Currently, there are three general approaches to detecting viruses: analyzing the host organism’s response to the virus, detection of a virus’s molecular fingerprints, and direct sensing of whole viral particles. The oldest methods for detecting the presence of viruses are based on the identification of intact viruses. The plaque assay [1] and the hemagglutination assay [2] both served as the gold standard for viral detection. The plaque assay and the hemagglutination assay are difficult to calibrate and cannot identify certain virus types. New methods are necessary for directly detecting and enumerating infectious viruses. Recent advances in optics and micro/nanofabrication provide more accurate approaches to viral quantitation. For example, the dynamic light scattering method can directly size virus-like particles (VLPs) [3]. Direct observation of stained viruses can be achieved by labeling the viral particle using specific antibodies with fluorophores [4]. The quartz crystal microbalance (QCM) method measures the mechanically resonant changes when target bio-molecules are captured by the surface antibodies [5]. The surface plasmon resonance (SPR) on a gold-coated prism [6–9] and localized SPR using antibody-immobilized gold nanoparticles (gold nanostructures) can detect viruses with a detection limit of several orders of magnitude better than conventional methods [10–12]. The SPR method is a label-free and real-time detection technology. It needs only a single sample-introduction step in the whole detection process. Therefore, the SPR technique is a good candidate for direct virus detection. However, the traditional SPR sensing technology utilizes a prism to induce SPR on a thin noble metal film and requires precise incident angle measurement. Such configuration makes it challenging for large-area, chip-based, and high throughput detections.

The SPR can also be excited by using metallic nanostructures. There are various types of nanoplasmonic nanostructures have been developed for the SPR sensing applications, such as plasmonic nanoparticle aggregates [13,14], nanohole [15,16], nanoslit arrays [17–20] or nanorod arrays [21,22]. They exhibit high sensitivity to the refractive index of the surrounding environment. Nevertheless, most of the measurements employ a spectrometer, which hinders the wide applications of nanoplasmonic sensing chips. Recently, we have developed a self-referencing two-color analysis technology by using two bandpass filters near the SPR wavelength [23,24]. The SPR signals can de directly detected using a simple transmission optical setup and a black/white CCD camera. In this work, we propose the direct detection of VLPs on multiple nanoplasmonic sensing arrays using a color image sensor without the additional bandpass filters. The built-in R and G filters in the color CCD were used as the two bandpass filters. The SPR wavelength of the sensing chip is designed at the crossing area of the R and G bands. The experimental result shows the detection limit for SARS-CoV-2 VLPs is lower than 2 pg/ml, which is much better than the complicated enzyme-linked immunosorbent assay (ELISA) approach. The platform with dual-band (G and R) images and a self-referencing two-color software analysis can also reduce the noise. Large area reading and cloud transmission data can also be used for home inspection. Our work provides a large area of high-throughput screening, low-cost process, and easy-to-use method for detecting and monitoring various pathogens, which is promising for point-of-care (POC) applications.

2. Principle of the color image detection

Figure 1 shows principle of the color image detection method for a nanoplasmonic chip. Figure 1(a) shows the optical image of 5 × 5 gold nanoslit arrays on a plastic substrate. The topography of the nanoslits is shown in the cross-sectional profile of the Atomic Force Microscope (AFM) image. The nanoslits has a period, P, a slit width, w and slit depth, d. The nanoslit arrays is coated with gold thin film with a thickness of t. The polarization of the incident light is perpendicular to the nanoslits for the gap plasmons to be excited in the gold nanoslits. The gap plasmons in nanoslits will form resonances due to the cavity effect. The resonance wavelength was estimated by a Fabry−Perot cavity and is expressed as follows [25,26]:

 

Fig. 1. Transmission-type SPR sensing chip and COMSOL Multiphysics simulation of the spectrum for gold nanoslit arrays. (a) The color and AFM images of 25 sensing regions in a plastic chip. The cross-sectional plot by the AFM shows the profile of nanoslits. (b) The measured transmission spectra of 430-nm-period gold nanoslit arrays. The spectra show a typical Fano resonance profile with a clear peak and dip resonance. The resonance is red-shifted when the refractive index increased from 1.333 to 1.416. It is noted the dip resonance occurs at the crossing area of green and red spectra of a color CCD. (c) The simulated field distribution for the gold nanoslit array with and without the PS beads. The field is not significantly changed by the PS beads. The parameters for the gold nanoslit array are PC w_1 = 165nm, PC w = 100nm, d = 200nm, t = 80nm, P = 430nm. (d) The calculated transmission spectra of gold nanoslit arrays with different numbers of PS beads on the surface. (e) The relationship between intensity changes of green (I_G), red (I_R) and contrast signals (I_GR). The contrast signal, I_GR = [I_G(λ) − I_R(λ)] / [I_G(λ) + I_R(λ)], can enhance the signal and reduce the common noises from the fluctuation of the light source. (f) The calculated contrast signal as a function of the number of dielectric nanoparticles (diameter = 100nm) on the gold nanoslits surface.

Download Full Size | PDF

To verify that transparent particles like VLPs can also induce redshift of the Fano resonance, we applied the COMSOL Multiphysics simulation. The structure parameters were P = 430 nm, w = 100 nm, d = 200 nm, and t = 80 nm. The boundary condition was periodic in the x-direction and a perfect matching layer (PML) was applied in the z-direction. The incident light was x-polarized. The maximum element size was 102.3nm, the minimum size was 3.07 nm, and the resolution of narrow regions was 0.4 nm for both x and z-direction. The substrate was a plastic film (n_b = 1.581) and the surface was covered with water (n = 1.333) as shown in Fig. 1(c). It is noted that the dip wavelength matches well with the BW-SPP wavelength as indicated in Eq. (2). The BW-SPP mode is a SPR which is sensitive to outside environmental changes. An increase of the n will increase the λ_SPR. The surface refractive index and the transparent nanoparticles can result in the redshift of the BW-SPP mode. Figure 1(d) shows the redshift of the Fano spectrum as a function of 100-nm polystyrene (PS) spheres on the gold surface. The dip resonance was found to have an apparent redshift with the increase of the particle number. The transmission intensity increases on the left side of the resonant spectrum and decrease on the right side. When two different color filters fall in the left band and right band of the spectrum, the change of the surface refractive index can be directly read using the contrast signal.

Measuring the intensity change at a fixed wavelength near the resonant wavelength is the conventional way to read the SPR image signals. This method requires a reference intensity at the initial time. It also suffers from intensity fluctuation from the light source and detector. Therefore, the signal-to-noise (S/N) level is poor for the intensity interrogation method. Two intensity signals in this approach come from the same source and detector. The noise level can be substantially reduced as indicated in the following equation.

We redefine the left intensity as I_G and the right intensity as I_R to meet the green and red pixels of the color CCD, respectively. In the method, I_G increased and I_R decreased with increasing particle number, as demonstrated in Fig. 1(e). Such a push-pull type setup can double the signal and reduce the noise. Hence, the S/N was enhanced using the contrast signal. Figure 1(f) shows the calculated I_GR as a function of the different quantities of polystyrene pellets (zero, second, sixth, tenth beads) and more (Figure S1). In this calculation, the F_L (λ) is the transmission spectrum of the green filter and F_R (λ) is the transmission spectrum of the red filter. T(λ) is the transmission Fano spectrum of different amount of polystyrene (n_PS= 1.55) beads distributed on the nanoslit arrays under water environment (n = 1.333). The COMSOL simulations verify that a simple color CCD can read the transmission SPR images with good linearity and sensitivity.

3. Chip fabrication and measurement system

The plasmonic nanostructures were fabricated using the plastic compression-injection molding method. First, a 430-nm period nanowire array with an area of 5 mm × 5 mm was made on a silicon mold by electron beam lithography; the pattern was then transferred to the Ni-Co mold by electroplating method [28]. This metal stamp, with an area of 9 mm², was exploited to reproduce 25 nanostructure arrays on a polycarbonate film using hot-embossing nanoimprint lithography. A Ni-Co mold with 25 nanoslit arrays was constructed after electroplating the large-area plastic template. Plastic nanostructure chips were massively manufactured with this protruding area of metal mold using compression-injection molding (H3000R, Mitsubishi Engineering-Plastics Corporation) [29]. The chips were further deposited with 80-nm thick gold film by using a thermal evaporator. Figure S4(a) shows the flow chart of the fabrication process. Figures S4 (b) and (c) show optical images of 5 × 5 arrays on a polycarbonate (PC) chip and the topographic image of the periodic nanoslit arrays measured by an atomic force microscope (AFM) (Bruker VEECO Dimension 3100) operated in tapping mode in the air. The groove depth was 200 nm with a period of 430 ± 1.8 nm. The aspect ratio type of the AFM probe was 7:1 (guaranteed 5:1). The typical tip radius of curvature < 10 nm. The edge region of the nanostructure could not be well resolved due to the limit of the probe diameter and the curvature [30]. It is noted that the typical gold film thickness for the excitation of SPR is about 50 - 60 nm. In the gold nanoslits, due to the boundary shadow effect, the thickness of gold film in the bottom of nanoslits is about 60 nm (see Figure S5). The gap plasmons are directly excited in the nanoslits and coupled to the BW-SPPs on the outside gold surface. In addition, the 80-nm-thickness gold film can effectively reduce the direct light transmission from the gold film. It reduces the background light and makes the transmission measurement dominant by the Fano resonance mode.

The biosensing capability of the proposed color imaging method was tested by a commercial transmission scanner (Epson Perfection V800 Photo), as shown in shown in Fig. 2. Figures 2(a) and 2(c) show the 5 $\times$ 5 array sensing biochip. The chip was comprised of the gold nanoslit arrays, 40-um thickness Double-sided tape (Ds-tape) layer for the fluidic channels and polymethyl methacrylate (PMMA) plate for the cover. We used a laser engraving machine to fabricate the fluidic channels and wells in the tape. There were 25 sensing wells in the biochip, and the dimension of each sensing region was 5 $\times$ 5 mm². There were five channels and each chamber covered five sensing wells. The fluidic biochip connected five inner diameters of 0.25 mm and the outer diameter of 0.76 mm Tygon hoses, respectively. The fiver fluidic channels were connected to a syringe pump to inject the samples into the chip, as shown in Fig. 2(b). The multiple chambers design was suitable for high-throughput virus screening experiments. Color images were obtained using the scanner with 5000 × 5000 pixels. The horizontal and vertical resolution were 3200 dpi×3200 dpi. Furthermore, a water refrigeration circulation cooling system was placed on the glass plate of the scanner and in contact with the biochip. The temperature was maintained at 24°C for optimal antibody and virus interactions. It is noted that the incident light should have a single polarization (TM-polarized) for the sensing chip. A polarization sheet (extinction ratio: 9000:1 for average @400 nm -700 nm; wavelength range:450 nm∼680 nm; polarization efficiency (%) >99.98) was placed under the white LED light source to achieve the polarization. Polarization of the incident light was perpendicular to the nanoslits; hence, the gap plasmon resonances in nanoslits and surface plasmon waves on both sides of the periodic gold surface can be excited. The images obtained using the scanner were analyzed on the laptop using an in-house developed MATLAB program to calculate the image contrast signals.

 

Fig. 2. (a) The biochip design in this experiment. The biochip comprised of PMMA, Double-sided tape, and 25 gold nanoslit arrays. (b) The setup of the measurement system. The color images were obtained by scanning the biochip. The chip was placed in a water circulation cooling plate to maintain 24°C temperature. A polarization sheet was placed right after the LED line source for provide TM-polarized light into the biochip. Multiple samples were injected into the biochip using the syringe pump. (c) The photo image of the biochip.

Download Full Size | PDF

4. Measurement results

We first measured the refractive index sensitivity of the gold nanoslit arrays using the color analysis of the R and G bands to test the performance of the sensing platform. Figure 3(a) shows the color images under different refractive index solutions. The test samples were glycerin solutions with different refractive indexes from 1.333 to 1.416. The colors of the sensing areas were slightly changed. The intensity of the red color decreased and the green color increased when the refractive index increased. The measurement results can be explained by the transmission spectra shown in Fig. 1(b). When the 430-nm-period gold nanoslit arrays were covered with water, the dip wavelength was 570 nm and redshifted with an increase of the glycerin/water mixtures. The intensity of green part was increased and the red part was decreased due to the redshift of the resonant dip. However, it is hard to directly identify the slight change in the color image (see Fig. 3(a)). The contrast signals of R and G bands were introduced to quantitatively and sensitively determine the change of color. Figure 3(b) shows the self-referencing two-color analysis using G and R channels (I_GR) as a function of time. We sequentially injected the glycerol mixtures into the 25 sensing wells. The signals increased with the increase refractive index of the mixture for all sensors. Figure 3(c) shows the correlations between I_GR and good linearity with the environmental refractive index from 1.3330 to 1.4160. The average slope was about 4.04827 ± 0.075 RIU⁻¹, and the coefficient of determination (COD) was about 0.99828. Figure 3(d) shows the refractive index sensitivities and Adj. R-Square of all the sensing arrays. The average sensitivity slope and Adj. R-Square were ≈4 RIU⁻¹ and ≈1, respectively. There was a low-sensitivity sensor (number 21) due to the defects in the gold nanoslit array. Figure 3(e) shows the noise levels and the limit of detections (LOD) for 25 arrays on a biochip. The mean noise level was 0.85×10⁻⁴ and the average LOD was 2.96×10⁻⁵ RIU. Such sensitivity is comparable to that of the intensity measurement at the most sensitive wavelength using a spectrometer (1∼5×10⁻⁵ RIU) and that of the traditional prism-based SPR image method (1∼3×10⁻⁵ RIU) [31]. According to Fig. 3, The signal (I_GR) and detectable RIU varied about 1∼1.5 times depending on the position of the sensors. Due to the defects in the mold, there were variations of the sensitivities for 25 sensors on a single chip. However, the chip was fabricated by the same mold. Using the injecting molding process, sensors at the same position of different chips have the similar sensitivity. We have checked different chips and verified that variation is systematic.

 

Fig. 3. Transmission color images of gold nanoslit arrays with a period of 430 nm under different refractive index conditions. (a) Transmission color image of gold nanoslit arrays in different deionized water/glycerin Mixtures (1.3330∼1.4160). The area of each array is 5mm × 5 mm. The refractive index of the mixture was confirmed using a commercial refractometer. (b) The contrast signals between G and R channels (I_GR) as a function of time and, (c) the contrast signals (I_GR) against the refractive index for 25 sensing arrays. I_GR shows a good linearity for the refractive indices from 1.3330 to 1.4160. (d) The measured sensitivity and adjusted R-square as a function of the 25 sensors in a biochip. (e) The measured limit of detection (LOD) and noises for each sensing array.

Download Full Size | PDF

The refractive index change from the VLPs is like the immobilization of dielectric nanoparticles. We also tested 100-nm-diameter dielectric nanoparticles with various surface densities to have a better correlation with the VLPs. We tested amine-modified polystyrene (PS) beads. The amino group made the PS beads easily immobilized on the gold surface. The stability of the scanner system was investigated before the measurement. There were thermal noises from 0 min to 60 minutes and the system became stable after 60 minutes (see Figures S2 and S3 in Supplemental Document). Therefore, if the scanner system was turned on, it needs to be warmed up for at least 60 minutes. Figure 4(a) shows a schematic diagram of typical I_GR signal change in response to different concentrations of amine-modified PS beads (aqueous suspension, 0.1 µm mean particle size) [32]. We scanned the photo every minute for 1 pg/mL, 10 pg/mL, 1 ng/mL and 100 ng/mL, respectively. Furthermore, the samples were injected into the fluidic biochip using a syringe pump with a pumping speed of 1 µg/µL. The I_GR signals were gradually saturated from 50 minutes to 80 minutes. Figure 4(b) shows the measured results. The first 20 min were considered the reference value for deionized (DI) water. We added an 80 µL sample volume for each concentration of PS beads. The PS beads adhered to the biochip from 20 min to 50 min. DI water was injected into the biochip to wash the sample. Small signal jumps were observed. The increased jumps originated from the PS beads washing from the channels to the sensing areas. Figure 4(c) shows the mean signals and error bars for the 100-nm PS beads with concentrations of 1 pg/mL, 10 pg/mL, 1 ng /mL, and 100 ng/mL, respectively. The signals show a good linear relationship with logarithmic concentration. Moreover, the error bar was approximately 0.016. The larger error bar comes from the variations of sensitivity between the sensors as seen in Fig. 3(d). The distribution of PS beads on the gold nanoslit arrays for different concentrations of beads was observed using the AFM. The sample was dried using nitrogen gas, and the PMMA cover was removed before the AFM measurement. Figure 4(d) shows the AFM images for 1 pg/mL, 10 pg/mL, 1 ng/mL, and 100 ng/mL, respectively. The number of PS beads increased with the concentrations. We took several AFM images of each concentration to obtain the mean surface particle density. The mean particle counts in a 4 µm x 4 µm area is indicated at the upper axis of Fig. 4(c). The experimental tests show that the image contrast method can measure the 100-nm polystyrene particles with a surface density of fewer than 2 particles/µm². Figure 4(e) shows the cross-sectional profile of the particle on the gold surface. The size was approximately 100 nm, which is consistent with the size of the 100-nm PS beads.

 

Fig. 4. (a) Schematic distribution of amine-modified polystyrene (0.1µm mean particle size) on the gold nanoslit arrays. (b) The sensorgram as a function of time for four different concentrations of PS beads. (c) The mean contrast signals (I_GR) and error bars as a function of PS bead concentration. (d) The AFM images for four different concentrations (1pg/mL, 10pg/mL, 1ng/mL, 100ng/mL) of PS beads on the gold nanoslit arrays. (e) The cross-sectional plot of the immobilized particle.

Download Full Size | PDF

5. Direct detection of virus-like particles (0.1 µm mean particle size)

The previous section demonstrated that the gold nanoslit arrays using a commercial scanner can detect 100-nm dielectric spheres with good sensitivity and linearity. The typical virus size was also approximately 100 nm. It indicates that the same platform can be applied for virus detection. However, the real virus tests must be performed in the P3 laboratory. We used the virus-like particles as an alternative to prove the concept. In this experiment, the SARS-CoV-2 virus-like particles produced in HEK293T cells were tested. The VLPs were composed of four structural proteins: Membrane protein (M), Nucleocapsid protein (N), Spike protein (S), and Envelope protein (E). We first immobilized the anti-spike protein on the gold nanoslits surface to capture the VLPs through the spike protein. Figure 5(a) shows the surface modification process of the biochip. The gold surface was cleaned using oxygen (O₂) plasma (Start Plasma Pressure:35mTorr) for 20 min. We first coated a self-assembled thiolated-Protein G layer on the gold surface [33]. Furthermore, a 120 µL aliquot of thiolated-Protein G was injected into the sensing wells with a pump rate of 2 µL/min. A self-assembled Protein G layer was formed within 60 min interaction time. The common method to immobilize antibodies is the EDC/NHS method which activates the carboxyl group on MUA to bind antibodies. As illustrated in Fig. 5(a), the thiolated protein made the Protein G fixed on the gold surface [34] through the thiol-gold covalent bonding [35]. The Protein G has a high affinity to antibodies. It helps better standing of the antibodies to expose to their binding sites. Previous report [34] shows that such proteins can only bind antibodies through their nonantigenic (Fc) regions. This leaves the binding sites of the antibodies available to bind target antigens and increases the binding affinity.

 

Fig. 5. (a) Surface functionalization of the gold nanoslit arrays for the detection of SARS-CoV-2 VLPs. (b) The sensorgram SPR as a function of time under different surface conditions: Protein G-thiol, Anti-SARS-CoV-2 Spike mAb, BSA and SARS-CoV-2 VLPs. (c) The mean contrast signals and error bars as a function SARS-CoV-2 VLPs concentrations.

Download Full Size | PDF

The gold nanoslits surface was washed thoroughly in 100 µL PBS for 10 min with a syringe pump rate of 10 µL/min. Anti-SARS-CoV-2 monoclonal Spike mAb were then injected into the surface modified thiolated-Protein G (ProG-thiol) chip with syringe pump rate 1 µL/min under water cooling system for 2 h. The monoclonal antibody was derived from SARS-CoV-2 VLP immune mouse with a concentration of 100 µL (prepared in binding buffer pH 7.2). The chip was then washed with the binding buffer and incubated with 120 µL of 1 ug/mL BSA in PBS with a syringe pump rate of 2 µL/min for 60 min to block non-specific binding sites on the gold surface. The BSA can avoid non-specific biomolecules directly binding to the gold surface. The Protein G binds the excess SARS-CoV-2 antibody before the BSA blocking. The BSA cannot bind to the SARS-CoV-2 antibody, therefore most BSA blocking is on the gold surface.

VLPs samples with concentrations of 1 pg/mL, 10 pg/mL, 100 pg/mL, and 1 ng/mL were injected into the chip after washing with PBS. The injection volume was 60 µL. Figure 5(b) shows the typical response of the I_GR signals as a function of the time. There are multiple steps due to different surface conditions. The conditions were as followings: PBS baseline → ProG-thiol → PBS wash → Anti-SARS-CoV-2 Spike mAb → PBS wash → BSA → PBS wash → SARS-CoV-2 VLPs → PBS wash. Every step shows the increase in the surface refractive index due to the binding of the biomolecules. The entire process maintained a temperature of 24°C. We fixed the temperature because SARS-CoV-2 VLPs activity is longer and can be stably combined with Anti-SARS-CoV-2 Spike mAb at low temperatures [36]. The final step was capturing the VLPs through the anti-spike proteins [37]. The difference of signals in the PBS condition before and after the capture process was defined as the signal for the VLPs. Figure 5(c) shows the average signals and error bars for various concentrations of VLPs. The mean signal also showed a good linear correlation with the logarithmic concentration. The signal was much lower than the PS beads under the same concentration, as compared with Fig. 4(c). The reason is due to the lower refractive index of VLPs as compared to the amine-modified PS spheres. The variations in detection should be reduced to get a meaningful measurement. In the previous section, the major variation came from the sensitivity difference between each sensing array. For the sensors in the same position same array, the compression-injection molding process can produce sensors with very close sensitivity due to the same mold. Therefore, we measured the VLPs using the same array position in different chips. The error bars can be significantly reduced. Moreover, the error bar was approximately ±0.0009, as shown in Fig. 5(c). We found that VLPs with a concentration lower than 1 pg/mL can be detected under the signal and error bar condition. We have also tested the specificity by using Anti-Mouse IgG as an antibody. As seen in Figure S6, the Anti-Mouse IgG was well fixed by Protein G-thiol. When SARS-CoV-2 VLPs flew through the sensor, there was an abrupt change due to the change of bulk refractive index. However, after washing with PBS buffer, there is no significant change in signals. It indicates that the sensor has a good specificity for the antibody and antigen interaction. In the experiment, the PBS is acted as the buffer solution. The complexity of real samples such as UTM solution or saliva may have an influence on the response. The substrate interference needs to be further confirmed for real applications.

6. Conclusion

We developed a cost-effective gold nanostructure-based SPR imaging platform system for direct detection of dielectric nanoparticles. The platform shows sensitive, fast, and reliable diagnostic capability for detecting virus-like particles. A commercial transmission-type scanner, self-referencing two-color analysis, and high throughput analysis significantly enhance sensing stability, sensitivity, and reliability. The results show that the detectable refractive index change of the current 25-well sensing platform was 2.96 ×10⁻⁵ RIU, SARS-CoV-2 VLPs detection with a LOD down to 1 pg/mL. The detection ranges are from 1pg/mL to 1ng/mL with ≦30-minute detection time and 80 uL sample volume. These results also show that this sensing platform can perform high-throughput detections of virus-like particles [38]. The 25-well biological analysis was successfully conducted with two commercial transparency scanners, Perfection V370 Photo and Perfection V800 Photo, Epson. Since the Perfection V800 Photo scanner could scan an A4-sized area, the number of sensing wells can be increased to 96, 384 and 1536 for more high-throughput applications.