Bead-based multiplex detection of dengue biomarkers in a portable imaging device

: Dengue is one of the most rapidly spreading mosquito-borne viral diseases in the world. Diﬀerential diagnosis is a crucial step for the management of the disease and its epidemiology. Point-of-care testing of blood-borne dengue biomarkers provides an advantageous approach in many health care settings, and the ability to follow more than one biomarker at once could signiﬁcantly improve the management of the disease. Bead-based multiplex technologies (suspension array) can measure multiple biomarker targets simultaneously by using recognition molecules immobilized on microsphere beads. The overarching objective of our work is to develop a portable detection device for the simultaneous measurement of multiple biomarkers important in dengue diagnosis, monitoring and treatment. Here, we present a bead-based assay for the detection of one of the four serotypes of dengue virus non-structural protein (DENV-NS1) as well as its cognate human IgG. In this system, the ﬂuorescent microspheres containing the classiﬁcation ﬂuorophore and detection ﬂuorophore are imaged through a microﬂuidic chip using an inﬁnity-corrected microscope system. Calibration curves were plotted for median ﬂuorescence intensity against known concentrations of DENV-NS1 protein and anti-NS1 human IgG. The limit of quantitation was 7.8 ng/mL and 15.6 ng/mL, respectively. The results of this study demonstrate the feasibility of the multiplex detection of dengue biomarkers and present its analytical performance parameters. The proposed imaging device holds potential for point-of-care testing of biomarkers on a highly portable system, and it may facilitate the diagnosis and prevention of dengue as well as other infectious diseases.


Introduction
Dengue fever is a major public health problem in tropical and sub-tropical areas. It is the fastest spreading mosquito-borne viral disease, and it affects an estimated 390 million people in more than 120 countries every year [1][2][3]. Dengue fever is caused by one of four antigenically distinct serotypes of dengue virus (DENV), which is a Flavivirus like Zika virus, West Nile virus, and yellow fever virus.
In response to the virus, infected individuals produce immunoglobins (Igs, such as IgG or IgM) against DENV envelope glycoproteins, commonly resulting in the elimination of the virus. However, in a small number of cases it evolves into a severe illness called dengue haemorrhagic Point-of-care testing (POCT) holds great potential for bridging the gap between healthcare providers and patients by bringing diagnostic tests in portable and cost-effective devices right to the bedside [12]. POCT provides actionable diagnostic information at the patient's bedside without immediate need for the primary healthcare facility to have access to a medical laboratory [13][14][15][16]. Qualitative POCT platforms such as lateral flow tests are widely used, however, this approach is restricted to a positive/negative test result without the capacity to provide quantitative results. Recent technological advances, such as microfluidic sample handling, plasmon enhanced optical sensors and smartphone-based systems, enabled the development of a new generation of rapid and sensitive detection systems of biomarkers [17][18][19][20][21][22].
Compared with a single analyte assay, multiplexing can improve the throughput of biomarker detection, reduce the cost per assay by consuming less reagents/labor, and produce parallel testing results faster [23]. Bead-based multiplex immunoassays typically require the measurement of fluorescent emission using flow cytometers [24][25][26]. Flow cytometers can measure up to a million cells, or microspheres, in seconds and a wide range of parameters can be obtained including forward scattering, side scattering and fluorescent intensity of multiple colors. A number of lab-on-a-chip flow cytometers have been reported to measure the scattering, fluorescence and electrical impedance of different sized beads [27]. Our previous work demonstrated the capacity of POCT for counting CD4 cells and measuring C-reactive proteins [28,29]. The success of these studies warranted the further development of a universal system for multiplex immunoassays.
The objective of this work was to develop a system capable of simultaneously detecting and quantifying DENV-NS1 protein and DENV-specific IgGs. A multiplexed and portable imaging device was prototyped using an assembly of microfluidic chips, objective lens, filters, prism, camera, etc. Monochrome images were captured to generate the fluorescence intensity values for bead set classification and biomarker detection. These data show that the system has the potential to be developed into a commercial device and fill the gap between POCT needs and the low accessibility of centralized laboratory resources.

Bead coupling/assay
Beads encoded with different APC concentrations act as unique barcodes. The beads with high APC concentration (bead set A) were coupled with the anti-DENV-NS1 specific synthetic antibody to capture DENV-NS1 protein, while the beads with low APC concentration (bead set B) were coupled with DENV-NS1 to capture anti-DENV-NS1 IgG antibodies. Thus, as illustrated in Fig. 2, bead set A detects the DENV-NS1 protein, while bead set B detects specific human IgGs. Carboxyl-modified beads (1 million beads) were covalently coupled with each capture protein via a two-step coupling approach [31] using magnetic separation between each step. After washing with deionized water, beads were first activated with EDC (10 mM) and sulfo-NHS (20 mM) in MES buffer (50 mM, pH 6.0) for 15 minutes. After activation, beads were washed with 200 µL MES buffer twice. The beads were then mixed with 2.4 µg capture protein dissolved in 200 µL MES buffer. The mixture was incubated and mixed end-to-end for 1 hour at room temperature on a rotator. The reaction was terminated by adding 5 µL Tris buffer (100 mM, pH 7.4). After incubation for 5 minutes, beads were washed with PBSBT (PBS, pH 7.4, 0.1% BSA (w/v), 0.05% Tween-20 (w/v)) twice. The capture beads were then resuspended in PBS to carry out the immunoassay as follows. A population of 50,000 capture beads was added to the individual wells of a 96-well plate (EPX-44444-000, Thermofisher Scientific, MA, USA) and washed twice with PBST (PBS, pH 7.4; 0.05% Tween-20 (w/v)). Two-fold dilution series of DENV-NS1 standard samples were prepared in PBSB (PBS, pH 7.4; 0.1% BSA (w/v)) with concentrations ranging from 1.0 µg/mL to 0.001 µg/ml. The DENV-NS1 standards were mixed with capture beads in the wells and incubated for 1 hour on a plate vortex at 500 rpm at room temperature. After two washing steps with PBST using a magnetic separator, PE conjugated detection antibody was added into the beads for 30 min incubation at room temperature. The beads were then washed twice and then resuspended in PBS for detection. During the assay development process, flow cytometry was used to characterize and optimize the performance of the bead assay. After the initial characterization, beads were introduced in microfluidic channels and tested in the portable imaging device. Calibration curves of various concentrations of DENV-NS1 and IgG were plotted against the median PE fluorescent intensity. Simulated multiplex samples of DENV-NS1 and specific IgG were tested to demonstrate the characteristics of the multiplex detection.

Imaging device prototype
The optical detection system consists of a customized infinity-corrected microscope to image beads in a microfluidic chip ( Fig. 3(A)). Two laser sources, with wavelengths of 532 nm and 635 nm, excite the microfluidic channel from an oblique angle of 40.7°± 0.3°. The microfluidic chip is contained in a 3D printed cartridge, and the latter is fixed on a manual XYZ translation stage for motion control. One of the key limitations of a microfluid-based cytometer is the potentially lower throughput, therefore limiting the number of cells or particles that can be analyzed. [32,33] To increase sample throughput we used wide (800 µm) and shallow (20 µm) microfluidic detection channels. This configuration produces a ribbon-like fluid sample that increases throughput while limiting the chance of multiple beads flowing through the same location. We adopted a prism approach to separate the two fluorescent bands. A dual bandpass filter and a prism were placed between the objective lens and tube lens ( Fig. 3(A)). A dual bandpass filter was used to narrow the wavelength range of the emission bandwidth of each fluorophore and reduce the spill-over between the APC and PE fluorescence bands. Figure 3(B) shows an example of the image pairs formed after the dual band filter and the prism. Figure 3(C) shows an example of the intensity profile of the image pair. The intensity of the classification fluorescence (red) represents the identity of the bead set, indicating which biomarker is tested. The intensity of the detection fluorescence (yellow) represents the quantity of biomarkers and its intensity value correlates with the concentration of a biomarker in the fluid sample.

Image data processing
Conventional flow cytometers that use a photo multiplier tube (PMT) generate a photocurrent pulse when particles pass through the detection window. The 'pulse' is then amplified and converted from analog to digital values and the area, or height, value of the pulse is acquired for analysis. Here, 2048 × 2048 pixel images were captured by the camera. The images were analyzed using a bespoke algorithm for background subtraction, region recognition and intensity calculation.
To generate a lookup table for the pixels of the bead region of interest, the original 2D grayscale images were segmented from their background using an adaptive threshold, obtained from local first-order statistics. The local mean intensity in the neighbourhood of each pixel was used to differentiate the region of interest (ROI) pixels from total number of pixels in the image. Locally adaptive threshold uses a sensitivity factor between 0-1 range to specify sensitivity towards thresholding more pixels as foreground. The grayscale image was then converted to a binary image. The binary image was further modified to fill any empty pixels lying in the ROI, removing any connected component between bead regions. A 'regionprops' function was utilized to determine the exact number of beads in the image and their position. A pixel list of all the pixels lying in the bead region was generated and the centroid of each bead was calculated. Different approaches were evaluated to determine the 'classification' and 'detection' median fluorescence intensity (MFI) values, including measuring the height of the ROI, measuring integrated values of all ROI pixels and measuring the median grayscale value of a region around the ROI centroid [34,35]. Finally, a classification MFI was generated from integrated values of all ROI pixels. A detection MFI was generated from the median intensity value of a reduced ROI around the centroid.
After collecting the intensity values for the APC and PE channels, the first step was to define the classification (APC) fluorescence intensity range of the measured two bead sets. The second step was to statistically analyze the detection (PE) fluorescence intensity and report the MFI for the purpose of generating calibration curves and calculating sample concentrations. A logistic regression was used for curve fitting. The lower limit of quantitation (LLOQ) is defined as the lowest standard concentration showing a ≤ 20% relative error between expected concentration and interpolated concentration using logistic fitting.

Bead coupling/assay
Bead set A was coupled with a capture antibody specific to DENV-NS1, bead set B was coupled with a capture protein specific to anti-DENV-NS1 IgG. The coupling efficiency was confirmed by performing a bead-based enzyme-linked immunosorbent assay using serial dilutions of the target protein. It was found that Tween-20 interferes with the coupling reaction and reduced the surface coverage of the capture proteins significantly. Tween-20 was therefore avoided in the activation and coupling reactions and the washing steps. However, after bead coupling, we found that the PBST suspension buffer had a higher bead count recovery than PBS because Tween-20 reduced the attachment of beads to the tube surface. Coupled beads were then analyzed by measuring forward scattering, side scattering, and APC fluorescence intensity in a flow cytometer before and after coupling. Forward scattering, side scattering and APC intensity had less than a 3.0% change after the coupling reaction. This indicated that the bead diameter, surface morphology and internal APC concentration remained consistent after coupling.
The number of beads required for imaging in the microfluidic channel was estimated based on the dimension of the channel and imaging frame. For example, approximately 25,000 beads were required in a 10 µL sample, according to the dimension of the microfluidic channel (W × D × L: 0.80 × 0.020 × 58.5 mm), an assumed imaging frame (0.5 × 0.5 mm) and an assumed concentration of 10 beads per frame. Accounting for the possible loss of beads along the immunoassay and transfer, 50,000 coupled beads were used to perform the immunoassay. Twenty percent of the bead population was analyzed by flow cytometry as a reference and the rest of sample was analyzed in the imaging device. Four-parameter logistic (4PL) regression curves were plotted for varying concentrations of DENV-NS1 and IgG proteins ranging from 0.001 µg/mL to 1.0 µg/mL. The percentage error of the expected concentration and the interpolated concentration (from 4PL curves using flow cytometer results) was used to evaluate the performance of the bead assay itself. Flow cytometry results showed that the percentage error was less than 5% for concentration levels from 0.004 to 1.0 µg/mL. This indicated that the current immunoassay could accurately measure the quantity of DENV-NS1 and IgG in the range of 0.004 to 1.0 µg/mL using flow cytometry. Further investigation of the proposed imaging device was performed and is discussed in the following sections.

Device prototyping
Because of the wavelength difference between the fluorescence bands used, 565-589 nm (center wavelength 577 nm) and 665-715 nm (center wavelength 690 nm), a 15°prism was used to separate the two bands with a 0.078°angular deviation. Considering the angle and distance from the prism to the camera sensor, a separation of 55.7 pixels should be achieved at the imaging sensor. The 565-589 nm band presented a 15-pixel dispersion on the image plane, and the 665-715 nm band caused 19.5 pixels of dispersion. However, due to differing levels of focus being achieved throughout the channel, the pixel gap between the two bands was smaller in practice. As shown in Fig. 3(B), the separation between the two images in a bead pair ranged from 20-25 pixels (based on peak width at 5% peak height) and this was dependent on the intensity of both bands.
The red laser was utilized to excite the internal classification fluorescent dye (APC), and the green laser was used to excite the labelled detection fluorescent dye (PE). The excitation power was calibrated with a power meter to supply a consistent power level of 30 mW for the green laser and 45 mW for the red laser. Because of the Gaussian distribution of the beam intensity, it remained a challenge to maintain a uniform beam profile across the entire frame. After characterizing the beam size and intensity distribution, an elliptical area 480 × 720 pixels (minor axis × major axis) was used for further analysis. Within this area, the beam intensity varied by less than 10%. This area corresponded to a 240 × 360 µm (minor axis × major axis) elliptical area on the microfluidic channel.
In order to maximize the sensor area used to image the flow channel, while ensuring a sufficient depth of field (DOF), a 10x/0.25 numerical aperture (NA) objective lens was chosen. The DOF for this lens was smaller than the depth of the channel. However, as the intensity of the entire bead was measured, the focus was sufficient throughout the channel using this objective. While using a higher resolution 20x/0.4 NA objective lens (Olympus RMS20X) was tested to improve image resolution, that lens had a smaller DOF and the beads became highly out of focus when they were imaged at the edges of the channel. Therefore, the 10x/0.25 NA objective lens was used for further experiments to match with the width of the microfluidic channel and to generate the highest quality images of the fluorescent beads.
Two technical approaches were utilized to capture images of microspheres, the static-chip approach, and the static-fluid approach. The static-chip approach requires the pumping of fluids in microfluidic channels and the microfluidic chip assembly does not move. This approach captures images through a fixed area of the microfluidic channel. Therefore, once the imaging plane is in focus, it does not require further refocusing and will image the fluorescent bead at a higher speed. However, the flow velocity of microspheres in fluids must be carefully controlled to maintain the quality of images. Because consistent bead motion is required while capturing images, the effects of viscous force, inertial force and pressure change must be characterized and compensated for different fluid samples. Experiments were performed to image the flowing beads in microfluidic channels, but it was observed that the static-chip approach required a significant amount of high bead concentration sample to generate sufficient bead images for further analysis. Therefore, another approach, the static-fluid approach was attempted since it requires the fluid to be fixed/motionless in the channel when the camera sensor captures images. Because the precise control of bead flow is not required, this approach could accommodate various fluid sample types without the need to optimize the fluidic control system. To capture enough image frames, a longer microfluidic channel may be required, and the microfluidic chip assembly may be motorized to acquire multiple images. To test the feasibility of the static-fluid approach, we utilized a microfluidic chip with long channels (length 60 mm) and multiple images were captured by manually moving the microfluidic chip along the channel. The captured image sets were then used for intensity analysis.

Bead classification
The classification capability of the proposed device was characterized using an array of bead sets with four levels of classification intensity. Flow cytometry was first performed to evaluate the distribution of classification intensities. The coefficient of variation of each bead set fluorescence intensity was in the range of 7.7-10.8%. The same sample was measured through the imaging device and the coefficient of variation was in the range of 12.0-23.1%. This demonstrated the capability of the imaging device for measuring four bead sets at once, although the higher coefficient of variation compared with the flow cytometer is attributed to laser intensity fluctuation across the frame and clumping of multiple beads to form doublets or triplets. Different laser excitation power levels (15, 30, 45 mW) were tested to effectively separate the four intensity peaks. An incident power 45 mW provided the best performance for classifying the tested APC intensities. Two bead sets were used to perform the bead coupling and immunoassay as described in section 4.1.

Calibration curves
The imaging device was used to measure standard samples with varying concentrations of DENV-NS1 and IgG. Calibration curves were plotted by analyzing the detection intensity of bead images (Fig. 3(B)). As shown in Fig. 4 and 5, the expected correlation between the median fluorescence intensity and the concentration was observed. 4PL regression was used to fit the detection fluorescence intensity with the expected standard concentrations. The DENV-NS1 and IgG assays were able to measure a relatively low concentration in both samples, with the lower limit of quantitation at 7.8 ng/mL and 15.6 ng/mL respectively. These concentration levels are significantly lower than clinically relevant concentrations of DENV-NS1 and IgG proteins, and they are comparable to the assay range of conventional ELISA methods [9,11,36]. Comparably, bead intensity for the IgG assay had a higher standard deviation than for the DENV-NS1 assay, presumably resulting from the less uniform coverage of the capture protein. This leads to a higher standard deviation of detection intensity of beads either in flow cytometry or in imaging device measurements. To improve the uniformity of the capture reagent coverage on beads, additional coupling protocols should be investigated, especially with the mixing/handling procedures and chemical compositions (buffer composition, pH). It was observed that end-to-end mixing of the coupling reaction is important for achieving a more uniform distribution of surface protein coverage, presumably due to reduced aggregation and absorption of beads on the centrifuge tube wall. Surfactants (e.g. Tween-20) could potentially alleviate this issue although it may interfere with the efficiency of coupling reaction. Therefore, further investigation may be required to develop the optimal protocols for the coupling reaction.

Multiplex sample
Effective quantitation of multiple biomarkers could help diagnose the phase of dengue infection. To demonstrate the concept of multiplex detection, bead-based assays for DENV-NS1 and IgG were performed separately in microcentrifuge tubes. The bead samples from these two assays were mixed in different combinations to simulate an actual sample measurement [1,3,6,37]. Shown in Fig. 6 are scatter plots for the individual events detected using the imaging device. Simulated multiplex samples of the DENV-NS1 assay and the IgG assay at various concentration levels were measured. Cut-off values of classification intensity were applied to the DENV-NS1 assay (280-360) and the IgG assay (145-225). Some events were detected between the classified bead sets or beyond a normal intensity range. This is potentially due to the aggregation of beads to form doublets, triplets, etc. For a typical multiplex assay in flow cytometry, at least 35-50 events are required for each bead set to generate statistically significant measurements [38,39]. The current study adopted the similar strategy of maintaining at least 35 events per single assay. It was considered that the two bead samples may agglutinate with each other because of the complementary antibody affinity. However, this was not observed in the performed bead detection experiments. Theoretically, agglutination happens preferably among much smaller microspheres (<1,000 nm). Additionally, clumps of larger microspheres may not be strong enough to withstand the hydrodynamic forces trying to break them apart, and it will take considerably longer for all the larger microspheres to form clumps [40].
According to previous research, circulating DENV-NS1 ranges from 7 to 280 ng/mL in dengue-positive patients' sera [11]. Another study showed that free DENV-NS1 levels in plasma correlated with viremia levels and were higher in patients with dengue hemorrhagic fever than in those with dengue fever. An elevated serum DENV-NS1 level (>=600 ng/mL) within 72 h of illness onset identified patients at risk for developing dengue hemorrhagic fever [41]. Therefore, the detection range in our detection system has the potential to accommodate the DENV-NS1 sensitivity requirements for patient samples. Quantitative ELISA is usually performed using a designated positive serum calibrator from infected patients [42]. Optical density ratios between the designated calibrator and the specimen are used as 'quantitative' values, but this does not give a molar or mass concentration amount for any pathogen or epitope-specific IgG. Here, in contrast, a specific IgG is used as the calibrator. This will require testing a set of positive calibrator and negative control samples to properly calibrate the concentration of specific IgG standards with the antibody titer in clinical samples. Previous research showed that epitope-specific IgG concentrations ranged from 0.4 to 21 µg/mL for a bacterial infection [43]. Our detection system can quantify specific IgG as low as 0.0156 µg/mL, and therefore, it will enable sensitive detection of specific IgG. Its analytical performance, such as linearity and interference, needs to be further studied, however, our detection systems provides a new tool to investigate the antibody titers of positive calibrator sera and it holds strong potential to be developed and deployed for clinical validation.

Discussion
One important aspect for a field deployable diagnostic device is the shelf-life of the reagents. This is influenced by two factors: the stability of the coupled beads and the longevity of the pre-packaged, microfluidic device containing the assay reagents. First, the stability of the coupled beads will depend on the nature of the coupling reaction, such as covalent bond coverage and non-specific absorption coverage. According to specifications of typical off-the-shelf bead-based assay reagents, the coupled beads could last for 12 months, or longer when they are stored under proper conditions (e.g. 2-8°C in buffer solutions containing a stabilizer and preservatives) [44]. Second, if the reagents can be freeze-dried, or air-dried and immobilized in the microfluidic cartridge, the prefabricated testing cartridge may not need to be stored in a refrigerator. Roomtemperature storage may be feasible and therefore, its distribution facilitated. Ultimately, the stability of the beads and testing cartridges require the tracking of their analytical performance over time and these experiments are currently underway.
The repeatability of the presented biosensing method can be approached by analyzing the coefficient of variation of obtained analyte concentration between repeated image capturing. The CV% is determined to be 22.6% for the NS1 and 33.9% for the specific IgG at a respective concentration level of 0.01 µg/mL. It should be noted that this repeatability value can be further improved by optimizing the bead coupling protocol and applying a more uniform laser illumination source. Another factor that may contribute to the repeatability of the measurement is the photostability of the reporter fluorophore, PE. Also, labeling the beads with more durable reporter fluorophore (e.g. Quantum dots) may in theory improve the measurement precision. Overall, we have demonstrated the feasibility of the biosensing method and we have devised a strategy to further improve its performance.
In this study, APC is used for classification fluorescence, and PE is used for detection fluorescence. The selection of these two dyes is based on the availability of off-the-shelf fluorescent microspheres and fluorescent labels for flow cytometry applications. PE is widely used in flow cytometry detection because it has a very high quantum yield and therefore permits higher sensitivity. However, it was observed that PE brightness decreases within seconds when illuminated at a power of 50 mW, due to its susceptibility to photobleaching. This is not a concern for flow cytometry because individual beads pass through the detection window within microseconds [45]. But it emphasizes the need for the imaging device to interrogate the beads and capture images within a very short period (seconds) to counteract the photobleaching of PE. Therefore, the camera must be set to a short exposure time. Alternatively, other fluorophores could be considered. For example, quantum dot (QD) 565 was tested in the current set-up, and the brightness of QD565 was relatively low compared to PE, because of the low excitation efficiency of 532 nm for QD565. By replacing the 532 nm laser with a 405 nm laser, the excitation efficiency of quantum dots may be significantly improved.
The proposed device demonstrates the potential for measuring multiple biomarkers in a miniaturized setting. Oblique illumination was adopted in the current prototype for simplicity. Epifluorescence could would arguably be better for improving sensitivity by preventing the diffuse stray light reaching the detector. Image processing algorithms were successfully applied for the classification and detection of fluorescent beads. Further developments may involve intensity normalization using blank control beads and classification beads. Doublet beads may be present in the samples, requiring an event screening using morphology factors in addition to the intensity-based classification, akin to the singlet gating operation (forward scattering/side scattering) used in flow cytometer. The biosensing system presented is applicable to the detection of large protein molecules (e.g. > 10 kD) and the assay system should be capable of detecting other virus antigens based on the principle of sandwich immunoassay. Therefore, it holds great potential to be developed into a point-of-care testing device for the detection of dengue and other infectious diseases.

Conclusions
This proof-of-concept study was performed to demonstrate the characteristics of a new imaging device built to detect one of the serotypes of DENV-NS1 and the corresponding specific IgG antibody in a multiplex bead assay. Fluorescence intensity was calibrated with known concentration of the target DENV-NS1 and IgG. The lower limit of quantitation was shown to meet the reference range requirements for clinical samples and further investigation is needed to validate the analytical performance in clinical positive and negative specimens. This imaging device makes an important contribution to the development of a fully functional POCT device for the diagnosis of dengue fever and other emerging infectious diseases.

Funding
IC-IMPACTS (the India-Canada Centre for Innovative Multidisciplinary Partnerships to Accelerate Community Transformation and Sustainability).