A Simple Agglutination System for Rapid Antigen Detection from Large Sample Volumes with Enhanced Sensitivity

Abstract

Lateral flow assays (LFAs) provide a simple and quick option for diagnosis and are widely adopted for point-of-care or at-home tests. However, their sensitivity is often limited. Most LFAs only allow 50μL samples while various sample types such as saliva could be collected in much larger volumes. Adapting LFAs to accommodate larger sample volumes can improve assay sensitivity by increasing the number of target analytes available for detection. Here, a simple agglutination system comprising biotinylated antibody (Ab) and streptavidin (SA) is presented. The Ab and SA agglutinate into large aggregates due to multiple biotins per Ab and multiple biotin binding sites per SA. Dynamic light scattering (DLS) measurements showed that the agglutinated aggregate could reach a diameter of over 0.5μm and over 1.5μm using poly-SA. Through both experiments and Monte Carlo modeling, we found that high valency and equivalent concentrations of the two aggregating components were critical for successful agglutination. The simple agglutination system enables antigen capture from large sample volumes with biotinylated Ab and a swift transition into aggregates that can be collected via filtration. Combining the agglutination system with conventional immunoassays, an agglutination assay is proposed that enables antigen detection from large sample volumes using an in-house 3D-printed device. As a proof-of-concept, we developed an agglutination assay targeting SARS-CoV-2 nucleocapsid antigen for COVID-19 diagnosis from saliva. The assay showed a 10-fold sensitivity enhancement when increasing sample volume from 50μL to 2mL, with a final limit of detection (LoD) of 10pg/mL (~250fM). The assay was further validated in negative saliva spiked with gamma-irradiated SARS-CoV-2 and showed an LoD of 250 genome copies per μL. The proposed agglutination assay can be easily developed from existing LFAs to facilitate the processing of large sample volumes for improved sensitivity.

Graphical Abstract

1. Introduction

Lateral flow assays (LFAs) have been widely used as point-of-care or self-diagnostic tests, due to their ease of use, rapid turnaround time, and low cost [1]. However, compared to nucleic acid amplification tests (NAATs) or enzyme-linked immunosorbent assays (ELISA), LFAs often result in lower sensitivity, thus limiting their applications for early diagnosis [1]. The sensitivity of most LFAs is inescapably limited by their ability to accommodate only 50–100μL of samples, while various non-invasive samples such as saliva, nasal swab elution, and urine often provide much larger volumes. Adapting LFAs to accommodate larger sample volumes can improve sensitivity by loading more target analytes. Various techniques have been explored to concentrate target analytes from samples, such as isotachophoresis [2], osmosis [3], immunomagnetic capture [4,5], electrokinetic separation [6,7], and phase extraction [8]. For example, Golden et al. reported concentrating the malaria antigen in plasma through the capture of stimuli-responsive polymer-antibody conjugates at porous membranes [9]. Although these methods have significantly improved assay sensitivity, they often require elaborate device fabrication, complicated manual operations, or long processing time, hindering their potential for use in point-of-care settings. Therefore, there is still a need for a simple yet effective system to concentrate target analytes from large sample volumes to improve assay sensitivity.

Agglutination refers to the assembly of two multivalent components into large aggregates via specific binding interactions under appropriate concentration ratios. Agglutination only requires mixing the two components and is effective in generating visible large aggregates within a short time. Since the first agglutination-based diagnostic for typhoid fever in 1896 [10], agglutination tests have expanded to different applications such as blood typing [11], influenza virus screening [12], and detection of antibodies or other protein biomarkers [13–15]. However, most agglutination tests use the target analyte as one of the agglutination agents, and the signal is from the agglutination itself. Since agglutination requires appropriate concentration ratios, target analytes need to be present at high concentrations, which limits the test sensitivity. In addition, running multiple dilutions of the sample or assay reagents is needed to find the agglutination zone, which increases the labor intensity. Moreover, red blood cells or latex beads are often used for agglutination to allow visible readout, which requires complicated conjugation methods and adds the complication of reagent stability and consistency [16].

Here, a new and simple agglutination system is presented using biotinylated antibody (Ab) and streptavidin (SA). The agglutination is facilitated by multiple biotins per Ab and four biotin binding sites per SA, generating large aggregates that can be concentrated by filtration (Scheme 1A). We characterized the agglutination process through both experiments and Monte Carlo modeling. Furthermore, a new format of agglutination assay is proposed by combining immunoassays and the simple agglutination system (Scheme 1B). Specifically, biotinylated capture Abs and gold-nanoparticle-labeled detection Abs will generate immunosandwich structures in the presence of the target antigen. Upon addition of SA, the immunosandwich complex will undergo agglutination and the aggregated structures can be collected by filtration and detected in an in-house 3D-printed device. In this format, while the selectivity of the Ab pair ensures the specificity towards target analytes, the agglutination is independent of the analyte, and the ratio of biotinylated capture Ab and SA can be well-controlled to ensure efficient agglutination without running multiple dilutions. As the new agglutination assay only requires standard biotinylation of Abs, it can be easily developed from existing LFAs. Since the proposed agglutination assay allows capturing antigens from large sample volumes, we hypothesize that the assay can achieve a better limit of detection (LoD) compared to conventional LFAs by detecting more analytes.

Scheme 1. Agglutination of biotinylated Ab and SA. (B) Workflow of the proposed new agglutination assay.

(A)

The coronavirus disease 2019 (COVID-19) pandemic has caused over 6 million deaths globally since its outbreak. Despite the development of effective vaccines, as of February 8, 2023, there are still over 200,000 daily new cases globally, according to the World Health Organization (WHO) [17]. Therefore, diagnosis remains critical in controlling the spread of infection. Rapid COVID-19 antigen tests are widely adopted to detect SARS-CoV-2 nucleocapsid antigen from nasal swab samples. In addition to nasal swab detection, saliva testing has been drawing attention as a minimally invasive alternative sample type. In addition, most studies using RT-PCR have found similar sensitivities using saliva samples compared to nasal swabs [18]. Notably, up to 5mL of saliva can be collected from the average person by spitting, making it suitable for increasing the sample volume to improve sensitivity [19]. However, existing saliva-based LFAs miss the opportunity to use large sample volumes and reportedly have diagnostic sensitivity less than 50% [20,21]. As a proof-of-concept, we developed an agglutination assay detecting SARS-CoV-2 nucleocapsid antigen for COVID-19 diagnosis from saliva. We compared the assay performance with the conventional LFA format and further validated the assay in saliva matrix.

2. Materials and Methods

2.1. Antibody Biotinylation

Biotin-dPEG₄-TFP ester (Quanta Biodesign, Cat. No. 10009) was dissolved at 33mM in DMSO upon arrival and stored at −20°C. Anti-SARS-CoV-2 nucleocapsid monoclonal antibody (Exonbio, Cat. No.: NP_11A7) was diluted to 6.67μM in 500mM sodium bicarbonate buffer (pH 9.6) and spiked with the biotin linker stock. The final molar ratio of biotin linker to antibody was at 20, 50, or 80 to achieve different biotinylation levels. The reaction was incubated at 4°C overnight. The biotinylated antibody was purified into 1xPBS using Zeba Micro Spin Desalting Columns 7K MWCO (Thermo Scientific, Cat. No.: 89877) following the manufacturer’s protocol. The biotinylation level on the antibody was quantified with HABA assay (Sigma-Aldrich, Cat. No.: H2153) following the manufacturer’s protocol.

2.2. Agglutination Characterization

33.3nM biotinylated antibody was mixed with different concentrations of streptavidin (Sigma-Aldrich, Cat. No: 189730) or polystreptavidin R (Antibodies-Online, Cat. No.: ABIN4370319) in PBS and incubated at room temperature for 15 minutes. Particle size was then measured via dynamic light scattering.

2.3. Monte Carlo Simulation

See Supplementary Information for model details. In all simulations, the Ab number was 10,000, and the reaction extent p was set to 0.4. The valency of SA and poly-SA was set to 4 and 100, respectively. The number of SA or poly-SA and the number of biotins per Ab were varied for different simulation conditions.

2.4. Experimental Estimation of Reaction Extent

HABA (TCI America, Cat. No: H0586) was dissolved in 0.01M NaOH to make a 10mM HABA stock. 0.3mM HABA in 1xPBS was spiked into standard dilutions of SA and absorbance at 500nm was measured to generate a standard curve. Agglutinated reactions were spiked with 0.3mM HABA to quantify the remaining free SA concentrations.

2.5. Filtration Efficiency of Agglutinated Aggregate

Samples of 20ug mL⁻¹ Ab only, 48μg mL⁻¹ poly-SA only, and a mix of them both were incubated for 15 minutes. Whatman Grade 934-AH glass microfiber filter (Cytiva, Cat. No.: 1827–025) was put inside a syringe filter holder (Sartorius, Cat. No.: 16517-E). 10mL of 10ug mL⁻¹ BSA in PBS flowed through the filter to passivate the filter and reduce unspecific binding, followed by 10mL PBS, to wash the filter and minimize BSA in the final flowthrough. 500μL sample was immediately added to the syringe filter, and the flowthrough was collected. Total protein concentration in the remaining non-filtered sample and the filtered flowthrough was measured using the Micro BCA Protein Assay Kit (Thermo Scientific, Cat No.: 23235) following the manufacturer’s protocol (test tube procedure), except that only 75μL sample was used. The BCA assay was added to a 96-well plate to measure A562 using BioTek Synergy Neo reader.

2.6. SARS-CoV-2 Nucleocapsid LFA

2mg mL⁻¹ Polystreptavidin R was deposited onto nitrocellulose (Sartorius, Cat. No. 1UN95ER100020NT) as the capture line using a Biodot ZX1010 Dispense System. After drying overnight, the nitrocellulose was pre-blocked by soaking in PBS with 1% casein for 2 hours at room temperature followed by a 10-minute soaking in 1xPBST buffer. After drying overnight, the nitrocellulose was then assembled with absorbent pads on backing cards and cut into 4mm-wide lateral flow strips.

Anti-SARS-CoV-2 nucleocapsid monoclonal Ab (Bioss, Cat. No.: bsm-41411M) conjugated with gold nanoparticles (custom conjugation provided by the vendor) was used as the detection antibody. The recombinant SARS-CoV-2 nucleocapsid protein (AcroBiosystems, Cat No.: NUN-C5227–1mg) in PBS with 1% BSA and 1% Triton X-100 was spiked with 1μg mL⁻¹ biotinylated capture Ab and 5μg mL⁻¹ detection Ab, and pulse-vortexed for 15 seconds for mixing. After incubating for 15 minutes, 50μL sample was added to the lateral flow strip, followed by 50μL PBST as wash buffer.

2.7. SARS-CoV-2 Nucleocapsid Agglutination Assay

The assay device was printed with a 3D printer (Original Prusa i3 MK3S+). The Whatman Grade 934-AH glass microfiber filter was pre-blocked by soaking in 1xPBS with 1% casein for 2 hours at room temperature, followed by 10 minutes in 1xPBST buffer. The filters were then dried overnight and cut into quarters. The filter was placed under the sample port of the device, followed by 5 layers of absorbent pads (Cytiva, Cat. No., 8115–2250). Two springs were used to hold the layers together to ensure close contact.

For 50μL agglutination assays, the reaction was prepared the same as in the LFA using the same Ab concentrations. After a 15-minute incubation, 0.5μL of 240μg mL⁻¹ poly-SA was spiked into the reaction and pulse-vortexed for 15 seconds for mixing. After another 15 minutes, the sample was added to the agglutination device, followed by 50μL PBST wash. For 2mL agglutination assays, all the procedures were the same except that 0.5μg mL⁻¹ biotinylated capture Ab and 0.2μg mL⁻¹ detection Ab were used. 500μL PBST was used as the washing buffer. The diameters of the sample port in the agglutination device for 50μL and 2mL assays were 1.5mm and 2.5mm, respectively.

2.8. Quantification of Inactivated SARS-CoV-2 Stock Concentration with RT-qPCR

Gamma-irradiated SARS-CoV-2 was obtained from BEI Resources (Cat. No.: NR-52287). RNA was extracted using QIAmp Viral RNA kits (Qiagen, Cat. No.: 52904) and added to the FDA-EUA-approved Smart Detect^™ SARS-CoV-2 rRT-PCR kit (InBios International, Inc.) for quantification following the manufacturer’s protocol. The PCR reactions were run on a Bio-Rad CFX96 real-time detection system. Synthetic SARS-CoV-2 RNA control from Twist Biosciences (Cat. No.: MT007544.1) was used as the standard.

2.9. Agglutination Assay with Saliva Matrix

Pooled saliva was collected from volunteers with no reported COVID-19 symptoms and stored in a −80°C freezer before use. Sample collection was approved by the University of Washington IRB (Study STUDY00010543). After thawing, 2mL saliva was spiked with recombinant SARS-CoV-2 nucleocapsid antigen or inactivated virus and mixed with 0.5mL lysis buffer, resulting in a final buffer condition of 0.1M sodium phosphate (pH=8), 10mM NALC, 1% Triton X-100 and 1% BSA. The sample was incubated for 15 minutes and then filtered with a 0.45μm syringe filter (PALL, Cat. No.: 4654). The sample was then spiked with capture Ab and detection Ab, followed by poly-SA, in accordance with the agglutination assay protocol.

2.10. Signal Quantification

All strip images were scanned using an Epson V700 photo scanner. The signal intensities were quantified in ImageJ and mean intensity with the background signal subtracted was reported. For LFAs, a rectangle 73 pixels wide and 25 pixels long was centered over the test line as the region of interest (ROI). Rectangles of the same size 50 pixels above and below the ROI were selected as the background. The averaged mean intensity from the two background regions was used for background subtraction. In cases where lines were not visible, the distance between the test lines and the bottom of the strips was used to identify the test line region. For the 50μL agglutination assay results, the procedure was the same except a circle with a diameter of 24 pixels was used instead of a rectangle. For 2mL agglutination assay, a circle with a diameter of 65 pixels was used, and two regions 100 pixels away from the ROI on the filter, either in the x or y direction, were selected as the background.

3. Results and Discussion

3.1. SA-Biotin-Based Agglutination System

The agglutination system comprised of biotinylated Ab and SA. Commercial biotin-dPEG₄-TFP (tetrafluorophenyl) ester was used for the biotinylation of anti-SARS-CoV-2 nucleocapsid Ab. By tuning the biotin linker:Ab molar ratios during conjugation, Abs with 8, 5, and 3 biotins per Ab were generated. Since SA exhibits 4 biotin binding sites, we hypothesized that agglutination could occur between SA and biotinylated Abs. As shown in Figure 1A, when the 8-biotin Ab was mixed with increasing concentrations of SA, the particle diameter measured using dynamic light scattering (DLS) changed from several nanometers to over 0.5 micrometers before returning to the nanometer level. Similar trends were observed in other agglutination systems, such as the Heidelberger-Kendall curve describing the extent of antigen-antibody agglutination as a function of their concentration ratios [22]. The sharp increase of the particle size to near-micron-level diameters suggested that the SA-biotin-based agglutination system was effective in generating large aggregates from protein monomers. As expected, the agglutination peak appeared when the molar ratio of the biotins and SA binding sites was near unity, suggesting that equal concentrations facilitated the agglutination. No agglutination peak was observed for 5-biotin and 3-biotin Abs, and particle size remained at the nanometer level (Supplementary Figure S1). This suggested that higher valency was critical for agglutination.

Figure 1. Agglutination characterization.

(A) Results of 8-biotin Ab agglutinating with SA at different concentration ratios. Left vertical axis shows DLS measurements of the aggregate size. Right vertical axis shows the median values of the modeled aggregate size, where the modeled size refers to the total number of Ab and SA within the aggregate. (B) DLS measurements of the peak sizes of Ab with different biotinylation levels agglutinating with SA or poly-SA. In both panels, the whiskers of the DLS measurements extend to 2% and 98% of the distribution, and the boxes cover the interquartile intervals while the middle lines represent the median values.

To explore key design parameters of the SA-biotin-based agglutination system, we sought to model the agglutination process. Numerous models have been developed on different agglutination systems. However, most models focused on the agglutination kinetics and temporal evolution, which involved parameters such as binding rate constants that needed to be experimentally determined [22–24]. Goldberg et al. modeled the equilibrium state of antibody-antigen agglutination by introducing the reaction extent (p; the fraction of antigen sites that have reacted), but the model could only be applied when p was below a critical reaction extent [25,26]. Here, we adapted the Goldberg model’s assumptions in conjugation with a Monte Carlo approach to simulate the process. A detailed description of the model can be found in Supplementary Information. Briefly, the model inputs included the number of Ab and SA, the number of biotins per Ab, the number of biotin binding sites per SA, and the reaction extent p. We chose the fraction of SA sites that were reacted during agglutination to represent p. To experimentally estimate p, we used the HABA reagent to measure the remaining free SA sites after agglutination, resulting in p around 0.4 when the molar ratio of biotins and SA sites ranged from 0.5 to 2 (Supplementary Figure S2 and Table S1). Hence, p was set to 0.4 in the model. For the model output, the number of total Ab and SA within an aggregate was used to represent the aggregate size. Figure 1A presented the modeled size distribution when the Ab number was kept at 10,000 and the SA number was varied for different concentration ratios. With 8-biotin Ab, the modeled agglutination peak appeared when the concentration ratio was at 0.9, 1.0, and 1.1, where all the Ab and SA were essentially agglutinated into one large aggregate. The slightly larger aggregate size at higher concentration ratios was due to a higher input number of SA. Whereas for 5-biotin Ab and 3-biotin Ab, the agglutination peak was not observed in the model (Supplementary Figure S1). The modeled results aligned well with the DLS results, which confirmed that agglutination peaks were most likely to appear when the molar ratio of reacting sites was near unity, and higher valency of the components was critical for agglutination.

Since higher valency is expected to promote agglutination, we further tested using commercial poly-streptavidin (poly-SA) for agglutination. Poly-SA consists of multiple SAs crosslinked together via polymers. Although the exact valency and molar concentration were unknown, the estimated poly-SA molecular weight was 30–300 times that of single SA, suggesting a much higher valency. We mixed different concentrations of poly-SA with biotinylated Abs and empirically identified the agglutination peak using DLS. As shown in Figure 1B, the peak sizes with poly-SA were all larger than those using SA. Additionally, the poly-SA agglutination occurred with both 5-biotin and 3-biotin Abs, which did not aggregate with SA, and the median sizes of the agglutination peaks were 1.23μm and 0.65μm, respectively. The same effects were observed using our model simulation when increasing the SA valency to 100 (Supplementary Figure S3). Moreover, the measured median size of the agglutination peak using 8-biotin Ab and poly-SA reached 1.68μm. Therefore, we selected 8-biotin Ab and poly-SA to move forward with the development of the agglutination assay as it had the largest agglutination size.

3.2. Filtration of Agglutinated Aggregates

We next tested the separation of the agglutinated aggregate from the solution by filtration. After evaluating different filters, we selected glass microfiber filters with a particle retention size of 1.5μm due to their high loading capacity and relevant cutoff size. To test filtration efficiency, agglutinated samples were pushed through the selected filter in a syringe filter holder, and total protein concentrations before and after filtration were measured with the bicinchoninic acid (BCA) assay. The results were compared with the PBS control and non-agglutinated samples (biotin-Ab or poly-SA only). As shown in Figure 2, for agglutinated aggregates, the filtration flowthrough had non-detectable protein concentrations, showing that simple filtration was effective in capturing the agglutinated aggregates from the solution. The filtered flowthrough for biotin-Ab-only and poly-SA-only samples contained 51% and 63% of total proteins, respectively, as quantified with standard curves of the same biotin-Ab and poly-SA (Supplementary Figure S4 and S5). The protein loss likely resulted from non-specific binding to the filter. The results showed that agglutination enabled effective protein capture by filtration.

Figure 2. Filtration of the agglutinated aggregates.

BCA assay measurements are plotted for agglutinated samples before and after filtration and compared to PBS control.

3.3. Agglutination Assay Development and Performance

Based on the simple agglutination system and effective separation of agglutinated aggregates via filtration, we designed an agglutination assay that can accommodate large sample volumes. Similar to conventional immunoassays, biotinylated capture Ab and AuNP-labelled detection Ab were added to the sample to form sandwich structures with the antigen (Supplementary Figure S6). The sample was pulse-vortexed for 15 seconds to ensure thorough mixing. After incubation, poly-SA was spiked into the sample solution to trigger agglutination with the biotinylated capture Ab. The whole sample was then transferred to a 3D-printed filtration device (Figure 3) where the aggregate was captured on the filter surface together with the immunosandwich structure, generating signals for the detection of the antigen. To minimize background signals caused by nonspecific absorption of detection Ab onto the filter paper, filters were pre-blocked with 1% casein, and the running buffer also included 1% BSA.

Figure 3. Filtration device.

(A) Schematics of the filtration device design. (B) Pictures of the device. A five-cent coin is used for scale.

To evaluate the performance of the agglutination assay at different sample volumes, we selected the detection limit of the target concentration as the comparison standard since it is clinically the most relevant. We first developed in-house LFAs using the same antibody pair to provide a benchmark limit of detection (LoD). Representative strip images and quantified band intensities are shown in Figures 4A and 4B. With 50μL samples, the LoD was 50pg mL⁻¹ as determined by the quantified line intensity, similar to other published SARS-CoV-2 nucleocapsid LFAs [27–29]. We then tested the agglutination assay using 50μL samples with the same Ab concentrations. The positives showed a circle pattern as expected, and the intensity maintained relatively uniform across the region based on the line plots (Supplementary Figure S8). The LoD was 100pg mL⁻¹, slightly higher than the LoD of LFAs. The higher LoD was likely due to higher background noise caused by the rough surface of the filter paper compared to the smooth nitrocellulose used in LFAs. Nevertheless, the comparable level of LoD validated the feasibility of the proposed new format of the agglutination assay.

Figure 4. Agglutination assay performance.

The assay workflow is shown in Supplementary Figure S6. (A) Representative scanned images of the 50μL LFAs, 50μL agglutination assays, and 2mL agglutination assays. A uniform contrast is added for display purposes. Raw images without added contrast from all replicates (N=3) are shown in Supplementary Figure S9. (B) The quantified intensity of the 50μL LFA and 50μL agglutination assay results. (C) The quantified intensity of the 2mL agglutination assay results. All intensities are quantified using ImageJ. The dotted horizontal lines represent the cut-off value for LoD determination and SD represents the standard deviation. All LoD determination in the three assay formats followed the same method described in the reference [32]. The solid dots represent the individual value for each replicate (N=3), and the curves and shades represent the best-fit lines with linear regression and their 95% confidence bands.

We next explored the ability of this agglutination assay to accommodate large sample volumes. We selected a 2mL sample volume as it is reported to be a feasible volume for saliva collection [30,31]. When the same assay conditions for 50μL were applied to 2mL samples, the whole workflow took over an hour, and the signals from negative controls were extremely high. We hypothesized that the filtration region was clogged due to a large number of agglutinated aggregates that trapped the AuNPs and generated high background signals. To solve the issue, the diameter of the filtration region was increased from 1.5mm to 2.5mm by readjusting the device’s sample port diameter. Antibody concentrations were also decreased to reduce background signals. The final flow time was about 5 minutes for 2mL samples, and the negative controls remained clean. Furthermore, the LoD of the assay with 2mL samples decreased to 10pg mL⁻¹ (~250fM), showing a 10-fold enhancement in detection compared to using 50μL samples. The LoD improvement of approximately 10-fold fell short of the 40-fold improvement that might be predicted based solely on the larger number of antigens in the 40-fold sample volume. This may have been due to a reduction of antibody binding efficiency at the reduced antibody concentration in the 2mL sample test and/or increased signal dispersion of the larger flow-through area in the 2mL sample test.

The biggest challenge we encountered with increased sample volumes was high background signals. Besides reducing antibody concentrations, an alternative assay format is to add the AuNP-labeled detection Ab in the end, after the antigen capture, agglutination, and filtration, to avoid the unspecific binding of the detection Ab to the large aggregates or other proteins in the sample. However, the agglutinated aggregates may hide the captured antigen from the detection Ab due to steric hindrance. The fast flow time may also limit the binding between the detection Ab and antigen. On the other hand, in the current format, both antibodies bind the antigens in a homogenous solution, which allows binding all available analytes with fast kinetics. Therefore, the alternative format was not tested. Regardless, while multiple aspects of the assay, such as incubation time, can be further optimized to improve the sensitivity, the relative comparison of the three formats validated the feasibility of the proposed agglutination assay and its ability to improve assay LoD with larger sample volumes.

3.4. Assay Validation in Saliva Matrix

We used pooled saliva collected from healthy donors to validate the assay compatibility with the saliva matrix. Saliva is known to be viscous due to its high mucin content [33,34]. In initial attempts, saliva was simply diluted and added to the agglutination assay. However, the flow time was over two hours and background signals were high, likely due to filter clogging from mucin aggregates or other large particles in saliva. N-Acetyl-L-cysteine (NALC) in alkaline phosphate buffer is reported to reduce the viscosity of mucoprotein solutions mediated through the sulfhydryl group reducing disulfide bonds [35]. We found that incubating saliva with NALC buffer helped reduce the viscosity and homogenize the sample. An additional prefiltration step with common 0.45μm syringe filters further helped remove any remaining large components in saliva such as food particles and microorganisms. In the final assay, 2mL saliva was mixed with 0.5mL lysis buffer with a final concentration of 10mM NALC and 1% Triton X-100 for reducing saliva viscosity and lysing the SARS-CoV-2 virus, respectively. After incubating for 15 minutes, the sample was prefiltered with syringe filters and then transferred to the agglutination assay (Supplementary Figure S7).

We first evaluated the assay LoD using negative saliva spiked with SARS-CoV-2 nucleocapsid antigen. As shown in Figure 5A, the assay was able to detect concentrations as low as 10pg mL⁻¹ antigen, the same as the LoD in the buffer, suggesting that the assay was compatible with the saliva matrix. We further tested the assay with negative saliva spiked with gamma-irradiated SARS-CoV-2. The stock concentration of the inactivated virus was quantified using a commercial SARS-CoV-2 RT-qPCR kit (Supplementary Figure S10). The agglutination assay was able to detect 250 genome copies per μL of inactivated virus (Figure 5B). The lower variance of the signal intensities with inactivated viruses may suggest that the antibody pair binds more strongly towards native antigens versus the recombinant ones. One SARS-CoV-2 virus is reported to contain about 1800 copies of nucleocapsid protein [36]. The LoD of 250 genome copies per μL corresponds to 35pg mL⁻¹ nucleocapsid antigen, which is on par with the observed 10pg mL⁻¹ LoD. The slightly higher LoD with inactivated viruses could result from incomplete lysis during the assay or antigen degradation during the inactivation process of the stock. Nevertheless, the WHO target product profiles (TPP) for point-of-care COVID-19 tests set the acceptable LoD at 1000 genome copies per μL [37], which is higher than the 2mL agglutination assay LoD. The LoD of 10pg mL⁻¹ nucleocapsid antigen and 250 genome copies per μL of virus in saliva was well within the range of the reported viral load in saliva samples [38,39].

Figure 5. Agglutination assay in saliva matrix.

(A) The quantified intensity of agglutination assay detecting recombinant SARS-CoV-2 nucleocapsid antigen spiked into saliva. (B) The quantified intensity of agglutination assay detecting inactivated SARS-CoV-2 virus spiked into saliva. The acceptable LoD from the WHO target product profile (TPP) is 1000 copies μL⁻¹ [37]. The dotted horizontal lines represent the cut-off value for LoD determination and SD represents the standard deviation. Scanned raw images for all replicates can be found in Supplementary Figures S11 and S12.

In summary, we started with testing in the buffer and demonstrated that the new agglutination assay showed comparable detection limit to LFAs when using the same 50μL sample volume, and a 10-fold lower LoD when the sample volume was increased to 2mL. Besides the ability to accommodate large sample volumes and improve LoD, the run time for the agglutination assay was also fast (about 5 minutes for 2mL samples) due to the vertical flow. Moreover, the agglutination assay could be easily converted from existing LFAs by biotinylating the capture antibody to enable agglutination with poly-SA. We further validated the assay with saliva matrix. The LoD was the same as in the buffer and well within the clinical range. Although the assay required an additional prefiltration step, multiple commercially available saliva collection devices incorporate filtration during sample collection [40,41]. Our results suggest that an agglutination assay using 2mL saliva shows promises for sensitive SARS-CoV-2 antigen detection from saliva samples.

There are a few limitations that need to be addressed in future work. During assay development, we encountered high background signals when scaling up the sample volume, which may be caused by clogged filters. In addition to optimizing antibody concentrations and the filtration region diameter, future work could explore increasing the aggregate size with even higher valency or including multiple binding pairs besides SA and biotin, such as nucleic acid hybridizations. This may enable the use of filters with larger pore sizes and higher loading capacity. Additionally, the current assay workflow includes multiple user steps, and the filter needs to be removed from the device to observe the assay result, adding to the complexity for the point of care. The issues can potentially be resolved with improved device design, such as a spinning disk system that can automate assay steps and drive the filtration with centrifugation [42], or the detection region could be covered by a transparent film to allow direct observation of the assay results. Alternatively, the device could include a detachable part that can be simply removed after the assay to expose the filter region. Nevertheless, the proposed agglutination assay is simple to develop from existing LFAs and can be easily scaled up to accommodate different sample volumes.

4. Conclusions

We have developed a simple agglutination system comprising biotinylated Ab and SA or poly-SA. The agglutinated aggregates can reach sizes over 1 micron and be effectively separated from the solution via simple filtration. Moreover, with both experimental measurements and Monte Carlo modeling, we verified that higher valency and equal concentrations of agglutinating pairs facilitated agglutination, which can be applied to the development of other agglutination systems. Further, we developed an agglutination assay that can accommodate large sample volumes. Using the COVID-19 antigen test as a proof-of-concept, the assay showed a 10-fold enhancement in sensitivity with 2mL samples compared to 50μL samples. The final assay was able to detect down to 10pg mL⁻¹ SARS-CoV-2 nucleocapsid antigen and 250 genome copies per μL of inactivated virus spiked into saliva. The new agglutination assay can be simply developed from existing LFAs by biotinylating the capture Ab to enable agglutination. Although further simplification in the assay workflow is needed, we anticipate that the proposed agglutination assay can be applied to improve LFA sensitivity for different applications where large sample volumes are available.

• Biotinylated antibody can agglutinate with streptavidin into large aggregates.

• A COVID-19 agglutination assay able to process large sample volumes is developed.

• The sensitivity of the assay is enhanced with increased sample volumes.

• The assay can detect 10pg/mL SARS-CoV-2 nucleocapsid antigen spiked into saliva.