1. Introduction

Understanding the nature of immune memory post-SARS-CoV-2 infection and vaccination is essential to control the COVID-19 public health crisis, which has caused millions of infections, deaths, and a tremendous economic and social burden globally. Concurrent to the ongoing COVID-19 vaccination campaigns, widespread and inclusive protective immunity screening against all emerging variants of SARS-CoV-2 will be critical to prevent future outbreaks, take effective public health measures, and ensure safe operation of communities. The presence of neutralizing antibodies correlates with protective immunity against symptomatic infection with SARS-CoV-2 [1,2], which implies that serology profiling can inform protective immunity status against COVID-19. Yet, the serological signatures of immune protection, duration, and temporal dynamics of anti-SARS-CoV-2 antibodies after infection and vaccination in different populations are diverse and still under investigation [3–5]. Specifically, patient demographics [6,7] and clinical parameters [8–10] introduce heterogeneities in the host immune response to SARS-CoV-2. Therefore, the performance of serology tests, such as antigen-specific multiplexed antibody profiling and quantification over a large dynamic detection range is critically important to generate accurate, reliable, and standardized datasets to evaluate immunity after infections and vaccinations from the immensely diverse human immune response.

Current serological assays for SARS-CoV-2 include enzyme-linked immunosorbent assays (ELISA) [11,12], chemiluminescent immunoassays [13,14], and lateral flow assays (LFA) [12,15]. The most widely used ELISA and chemiluminescent assay protocols require technical expertise, laboratory infrastructure, and laborious multi-step bioassays rendering them out-of-reach for disproportionately affected low-income populations and under-resourced clinical laboratories. On the other hand, low-cost and simple serological LFAs show promise as rapid screening tests; however, they cannot achieve multiplexed quantitative analysis, and independent evaluation demonstrate wide performance variations [16]. Notably, false-positive results are typical in serology tests that are based on a single antigen since serum samples may contain cross-reactive antibodies [17,18]. Altogether, the limitations of current serological test platforms motivate the development of LFA-like portable, simple, and rapid biosensors that can reliably provide results that are on par with clinical labs.

Elucidation of protective immunity after infection and vaccination requires a greater understanding of the humoral immune response, which cannot be achieved by conventional assays that simply screen antibody titers against a single viral antigen or viral lysates. Importantly, in the face of the rapidly developing COVID-19 pandemic, the biomedical community has been challenged by the lack of humoral immunity models after infection and vaccination [19,20]. Hence, there is a need for a mass-deployable, user-friendly, and inexpensive biosensor technology that can enable multiplexed and precise quantification of antibodies against multiple viral antigens to capture an accurate picture of the host immune response and identify protective immunity trends at the population level. Tests that do not require trained healthcare personnel or clinical lab facilities could also help alleviate the disproportionate burden of COVID-19 on low-income populations and essential frontline workers by periodically updating individuals on their protective immunity status and advising on protective measures and timely vaccine boosters. In response to the urgent diagnostic needs of the COVID-19 pandemic, new biosensor technologies have been proposed, in which labelling agents such as fluorophores and nanoparticles were used for antibody detection [21–23]. While promising, such molecular detection methods based on sandwich assays require complex labelling reagents, limiting their on-site implementation.

Optical biochemical sensors based on nanoengineered substrates are emerging diagnostic technologies, empowered by recent developments in nanotechnology and photonics [24–30]. Nanophotonic biosensors enable enhanced light-matter interactions via optical resonance phenomena; thus, facilitating specific and quantitative detection of target analytes directly as they bind to the capture molecules that are immobilized on the sensor surfaces. Their analytical detection principle is based on measuring changes in the spectral resonance properties induced by the refractive index variations in the resonator’s surrounding media; hence, they are named refractometric sensors. Since the detection mechanism does not require enzyme-based signal amplification or exogenous stains, simple and rapid bioassays can be performed using nanophotonic biosensors [31–33]. Moreover, robust, and reliable bioanalytical platforms can be developed when these sensors are implemented using low-cost and portable optical readers [33]. Specifically, microarray-based surface functionalization with multiple target-specific capture probes can enable a comprehensive analysis of a specimen by detecting multiple analytes simultaneously on the same platform [34]. Yet, the key criterion in the development of quantitative and multiplexed biosensors is the adaptation of critical sensor parameters, such as dynamic range, sensitivity, and limit-of-detection to each target’s application-relevant requirements. This challenge becomes particularly important in antibody quantification to infer immunity status as humoral immune response varies drastically among individuals.

Here we present an integrated biosensor platform for multiplexed and label-free SARS-CoV-2 antigen-specific antibody quantification from human sera to address COVID-19 pandemic-imposed urgent biomedical testing needs. We implement a single-step microarray bioassay on a nanoplasmonic sensor to simultaneously quantify antibodies against SARS-CoV-2 Spike (S) and nucleocapsid (N) proteins from small volume (50 µL) human sera. Using a hyperspectral imaging-based optical setup, we investigate the benefits and discuss the limitations of multispectral resonance interrogation in nanoplasmonic sensors. We experimentally study and optimize parameters for extracting detection signals from the multispectral image datasets to maintain high biosensing performance while keeping the optical instrumentation simple for point-of-care (POC) diagnostics applications. We present data processing strategies based on ensemble averaging and binary classification to broaden the dynamic detection range (0.1-100 µg/mL). We measure a cohort of COVID-19 active (N = 6), convalescent (N = 18) and naïve (N = 5) human serum samples and report that our results show high concordance with an electrochemiluminescence (ECL) assay [14] (Pearson’s R ∼0.89 and 0.87 for anti-N and anti-S, respectively). Moreover, our biosensor can discriminate COVID-19 convalescent sera from negative controls with 88% sensitivity and 100% specificity. We present the measured antibody levels in COVID-19 naïve, acute, and convalescent groups, which reveal substantial heterogeneities in the humoral immune responses to SARS-CoV-2 antigens among our sample cohort. Finally, we demonstrate a syringe-like optofluidic cartridge integrated with our nanoplasmonic sensor that can enable antibody quantification directly from fingerprick blood samples, leading the way for future POC serology tests.

2. Materials and methods

2.1 Au-NHA plasmonic chips

The Au-NHA chips were manufactured using high-throughput, nanofabrication techniques on wafer-scale. Firstly, Radio Corporation of America-cleaned 4-inch fused silica wafers were coated with Ti/Au (10/120 nm) using an e-beam evaporator (Alliance-Concept EVA 760, Cran Gevrier, France). Afterwards, the nanohole arrays (200-nm diameter and 600-nm period) were patterned using a 248 nm deep ultraviolet stepper (ASML PAS 5500/300 DUV, Veldhoven, Netherlands). Finally, after resist development, the nanohole arrays were transferred into the Ti/Au layer using an ion beam etching tool (Oxford Instruments PlasmaLab 300 IBE, Abingdon, UK). More details on the wafer-scale fabrication of the Au-NHA chips can be found in the materials and method section of [33]. The wafers were diced into 1 × 1 cm². Before dicing, the wafers were spin-coated (3000RPM for 60s) with a Shipley S813 photoresist layer (1500 nm) for protection. The dicing was done with a dicing machine (Disco, Tokyo, Japan) using a resin blade. Subsequently, the wafers were cleaned through a three-step cleaning process. Firstly, the chips were immersed in MICROPOSIT^TM remover 1165 (Rohm and Haas Electronic Materials, Massachusetts, USA) for three days; the solvent was changed daily. Secondly, the chips were exposed to Oxygen plasma (250W, 1 minute, 80 sccm), and finally, the first step of RCA cleaning was performed. All three steps were done to ensure a clean sensor surface.

2.2 Protein, assay, and microarray patterning

SARS-CoV-2 Nucleocapsid (N) and Spike (S) proteins (Sinobiological, Beijing, People’s Republic of China) were prepared through dilution from 1 mg/mL stock solution to 150 µg/mL with 1x phosphate buffer saline (PBS), and Milli-Q Water, respectively; both solvents contain 0.5% trehalose (m/v) and 0.005% Tween20 (v/v). The PBS was used according to instructions from the manufacturer and has a pH of 7.4, similar to human blood. For control rabbit anti-bovine IgG (Thermo Fisher Scientific, Massachusetts, USA) was diluted to 200 µg/mL using 1x PBS and 0.005% Tween20 (v/v). The microarray patterning of 150 pL droplets was performed using an iTWO-300 spotter (M2-Automation, Berlin, Germany) for both target proteins and negative control. The droplet size on the chip was 100 µm diameter and spotted at 200 µm period. Specific patterns were performed to help with orientation while imaging the protein in the subsequent imaging steps. After microarray patterning was finished, the chips were incubated for 2 hours to ensure the binding of the proteins on the gold. During incubation, the chamber humidity was set at 70%, thus reducing the chamber dew point to 17^°C. The chips were placed on a cooling table set at 16^°C to avoid rapid evaporation of the droplet. After the initial incubation, the non-patterned areas of the chips were blocked through immersion of the chips in 1% (v/v) bovine serum albumin (BSA) in 1x PBS solution for 20 minutes. Subsequently, the chips were washed in 1x PBS with 1% (v/v) Tween20 solution for 5 minutes under constant agitation to remove unattached proteins. The patterned chips were then imaged using the optical setup, incubated with the investigated serum, washed with 1x PBS under agitation for 5 minutes, and imaged again.

2.3 Optical setup and data processing

A broadband laser SuperK FIANIUM15 (NKT Photonics, Birkerod, Denmark) coupled to an LLTF filter (Photon etc., Montreal, Canada) was used as a tunable narrowband light source with a full-width half maximum (FWHM) of ∼ 0.25 nm. The wavelength scanning was performed between 600 and 700 nm with 1 nm increments; this covers the plasmonic resonance peak of Au-NHAs, which is typically at ∼ 650 nm. The Au-NHAs were imaged in transmission with a 10x magnification objective on the optical light path of a TE-200 inverted microscope (Nikon, Tokyo, Japan) using a Prime BSI Express CMOS camera (Teledyne Photometrics, Arizona, USA). The camera exposure time was set to 25 ms, and the acquisition was made using a custom-made MATLAB (MathWorks, Massachusetts, USA) user interface, where the camera and filter functionalities were integrated inside one application using the app designer feature. After the acquisition of the desired part of the Au-NHA, the chip was removed, and another acquisition was performed with a neutral density (ND) filter of optical density (OD) 3 in lieu of the chip for determining the spectra of the source. The data processing was done using various Python libraries in Jupyter Notebook and is as follows. Firstly, the image pixel values of the Au-NHA were divided pixel-wise by the source dataset to eliminate the spectral shape of the light source in the transmission spectrum. Secondly, the spatially aligned hyperspectral dataset of before and after incubation with the studied serum were subsequently subtracted from each other. Thirdly, the resulting intensity changes of each pixel at particular wavelengths ΔIavg(λp) were then extracted from manually selected six circular areas with radii of 25 pixels. Finally, the ensemble average Δ(ΔIavg(λp)) was extracted, and binary classification methods were implemented on the obtained ΔIavg(λp) distributions of the pixels to extract quantitative molecular detection information. The Δ(ΔIavg(λp)) at N and S protein spots were further subtracted with Δ(ΔIavg(λp)) of control to remove the effect of random bindings. For Δ(ΔIavg(λp)), the values were multiplied with -1 to give positive correlations instead of the negative values from the raw signal (as shown in Fig. 3(b)) for convenience.

2.4 Calibration curves and sera measurements

For calibration curves of N-protein, one convalescent serum sample with a known high N-protein concentration was analyzed by diluting the serum 1 (no dilution), 2, 5, 10, 20, 50, 100, 150, 200, 500, and 1000 times. For the calibration curve of S-protein, measurements of 0.2, 0.5, 1, 5, 10, 50, 100 µg/mL anti-S antibodies (Sinobiological, Beijing, People’s Republic of China) diluted in 1x PBS solution were performed. The ensemble average for each dilution was calculated, and the data points were fitted with four parameters logistic function. Measurements of anti-N antibodies (Sinobiological, Beijing, People’s Republic of China) in 1x PBS solution were also performed with concentrations of 5 and 10 µg/mL, from whose values the range of the curve could be obtained: 0 to 100 µg/mL.

2.5 Cartridge and bovine blood

The cartridge was designed using Solidworks (Dassault Systèmes, Vélizy-Villacoublay, France) and was printed with plastic resin material using stereolithography technique using service from Midwest Prototyping (Wisconsin, USA). Bovine blood was obtained from the Dairy Cattle Center at the University of Wisconsin and was filtered using the cartridge and blood filter sheet (Cytvia, Massachusetts, USA).

2.6 Clinical samples

Human studies were performed according to the Declaration of Helsinki and were approved by the University of Wisconsin Institutional Review Board. All subjects provided written informed consent. Blood samples were collected by venipuncture into vacutainers that either allow the blood to clot (to prepare sera) or contain anticoagulants to prevent clotting (to prepare plasma). Blood was then centrifuged, and the supernatant sera or plasma was collected and then stored at -80 degrees. Serum was collected from COVID-19 naïve subjects prior to the pandemic. All subjects with COVID-19 had disease confirmed by SARS-CoV-2 PCR test. Serum and plasma were collected from COVID-19 acute subjects 0-21 (mean 6.2) days after a positive SARS-CoV-2 PCR test. COVID-19 convalescent sera were collected 5 or 12 weeks post-symptom resolution as determined by questionnaire as first described in Amjadi et al. [14].

3. Results

Our multiplexed and label-free SARS-CoV-2 antigen-specific antibody detection principle is illustrated in Fig. 1. We use nanoplasmonic sensor chips that are uniformly patterned with periodic nanohole arrays (NHAs) (period = 600 nm, hole diameter = 200 nm) into thin (120 nm) Au films, which are deposited on glass substrates. Among the myriad of nanophotonic devices proposed to date, plasmonic Au-NHAs are frequently preferred for refractometric label-free biosensing applications due to their excellent sensitivity to minute refractive index changes induced by target analyte binding to the ligand-functionalized sensors. The Au-NHAs exhibit extraordinary optical transmission (EOT) that are supported by the surface plasmon resonance phenomenon [35,36]. Surface plasmons in Au-NHAs can be excited by normal incident light, which enables their easy integration with optofluidic platforms and wide-field imaging-based optical interrogation. The spectral properties of the asymmetric Fano-type resonance profile [37] depend on the material and geometry of the NHAs (i.e., hole diameter, periodicity, metal, and substrate material), as well as the refractive indices of the surrounding media. While the latter makes NHA plasmonic devices highly sensitive and label-free refractometric sensors, the strict dependence of optical resonance properties on small variations in their nanoarchitecture makes them susceptible to manufacturing parameter divergences. The precision and accuracy of optical resonance properties within and in-between nanoplasmonic sensor chips is key to their successful implementation in real-world biosensor applications. Specifically, the plasmonic resonance uniformity within a nanopatterned chip enables reliable detection of a multitude of biomarkers from different spatial regions. Moreover, the coherence among chips from different manufacturing batches is critical for the overall analytical sensitivity and specificity of the biosensor platform. We use Au-NHA sensor chips that are fabricated using cost-efficient wafer-scale deep ultraviolet lithography and ion beam etching techniques [33]. Our approach yields sensor chips with robustly uniform plasmonic resonance properties (see S1), which enabled us to rapidly develop an antibody detecting biosensor platform to screen immunity status and address one of the most critical testing needs in the COVID-19 pandemic management.

 

Fig. 1. Imaging-based nanoplasmonic biosensor for multiplexed anti-SARS-CoV-2 antibody detection. (a) Schematic of a hyperspectral wide-field optical imaging platform showing a microarray-functionalized plasmonic Au nanohole array (Au-NHA) biosensor. Nanoplasmonic chip is illuminated with a tunable light source and imaged with a CMOS camera to acquire high-resolution spectral information from each camera pixel. (b) Hyperspectral datacube enables flexible probing wavelength (${{\boldsymbol \lambda }_{\boldsymbol p}}$) selection for optimized intensity interrogation of the plasmonic resonance peak shifts that are induced by antibody binding to the viral antigens immobilized in a microarray on the sensor surface. Note that spectral information is not necessary, as detection information can be retrieved by probing wavelength (${{\boldsymbol \lambda }_{\boldsymbol p}}$) in the form of intensity contrast, $\Delta {\boldsymbol I}({{\boldsymbol \lambda }_{\boldsymbol p}}$), which enables simpler optical reader setups. (c) Single-step label-free biorecognition assay to quantify multiple antigen-specific antibodies simultaneously. Plasmonic sensor chips are functionalized with SARS-CoV-2 Nucleocapsid (N), Spike (S) proteins, and anti-bovine antibodies for negative control in a microarray pattern. (d) A large-area nanoplasmonic sensor chip functionalized with high-throughput microarray printing technology. (e) A single image showing a representative field-of-view that can capture 49 detection spots at once. (f) Schematic and photographs demonstrate the plasmonic sensor integrated fluidic cartridge that can take low-volume blood samples, filter blood, and transfer plasma into biosensing chamber rapidly using a low-cost plunger system, which is critical for point-of-care deployment.

Download Full Size | PDF

Figure 1(a) depicts the hyperspectral imaging-based optical interrogation of Au-NHA biosensors. We use a supercontinuum laser source coupled to a continuously tunable bandpass filter for multi-wavelength excitation and a CMOS camera for image acquisition on a transmitted light brightfield optical setup (See S2). Each hyperspectral data cube maps a 1.3 × 1.3 mm² field of view (FoV) sensor area on a 4 megapixels detector array with a diffraction-limited spot size of 1.4 µm. Narrow bandwidth (∼0.2 nm) illumination with high-resolution wavelength tunability (Δλ=1 nm) allows for spectral analysis of plasmonic resonances with high spectral resolution. Figure 1(b) shows the measured spectra of a typical EOT peak before and after the plasmonic resonance redshift (Δλpeak), induced by antibody binding on antigen functionalized microarray spots. The magnitude of the shift depends on the refractive index change in top media (Sensitivity = 615 nm/RIU), which in turn correlates with antibody concentration. Measuring Δλpeak requires sophisticated instrumentation, such as spectrometers or tunable light sources, potentially increasing the complexity of the optical readers. Alternatively, imaging-based intensity interrogation facilitates simple and low-cost optical platforms. Herein, the analyte concentration is directly extracted from intensity contrast ΔI(λp) in a single image acquired with a narrowband light source at a probing wavelength, λp, which is spectrally tuned to the resonance peak. The intensity contrast ΔI(λp) originates from the shift of the resonance peak and light extinction caused by the protein biofilm when target binding occurs. In this work, we extensively study the intensity interrogation technique in a biosensor development context and report imaging-based methods that yield results on par with hyperspectral detection.

In our experiments, Au-NHA sensor chips (area = 1 × 1 cm²) were functionalized with SARS-CoV-2 N and S proteins and rabbit anti-bovine antibodies for negative control using a high-throughput microarray dispenser (Fig.1d). The N and S proteins are two of the major structural proteins of SARS-CoV-2. Specifically, SARS-CoV-2 binds to the human ACE2 receptor via the receptor-binding domain, which is a fragment of S protein. Thus, S and N are considered main immunogens, and antibodies against S and N-proteins are frequently used in serological tests to predict immune status against SARS-CoV-2 infections [38–40]. Following the microarray formation, the chips are rinsed to remove excess biomolecules and blocked using bovine serum albumin (BSA) to reduce non-specific binding. Fig.1c shows typical image data collected before and after incubating the chips with human serum samples. Fig.1e shows a single FoV image that contains detection data from 49 microarray spots. In fact, the sensor could contain up to a couple of thousands of microarray elements; therefore, multiple image data sets can be acquired from a single chip, enabling high-throughput antigen-specific antibody profiling from small-volume samples.

An important obstacle that hinders the field deployment of serology tests is the need for blood withdrawal and pre-processing for serum or plasma extraction. Here, we propose a sensor-integrated fluidic cartridge that can take low-volume blood samples (∼100 µL), which can potentially be acquired by a fingerprick (Fig. 1(f)). Moreover, our design can filter blood and transfer plasma into a measurement chamber where the plasmonic biosensor is integrated. Importantly, our approach does not need any sophisticated instrumentation, such as pressure pumps, but rather it uses a low-cost and user-friendly plunger system, which is critical for field-deployment of our biosensor platform. We report proof of concept antibody detection results from bovine blood using our fluidic cartridge integrated plasmonic biosensor in the Supplement 1 (S3).

After establishing the measurement and data acquisition workflow, we investigated different data processing methods to identify the suitable metrics for antibody quantification. We acquired hyperspectral image cubes before and after measuring numerous samples with varying antibody concentrations to generate titration curves from 100 ng/mL to 100 µg/mL. Fig. 2(a) depicts images of representative microarray spots acquired from different antibody titers. In correlation with increasing antibody concentrations, we observe an increase in the contrast of antigen-immobilized spots. Importantly, negative control spots functionalized with non-specific antibodies do not show any significant contrast with respect to the BSA blocked background despite being incubated in a high-concentration (100 µg/mL) anti-S antibody solution, emphasizing the antigen-specific antibody detection in our bioassay. In our raw image data pre-processing protocol, we first registered images acquired before and after sample incubation (t = 1 h) and identified the ΔI(λp) distribution at the single image-pixel level (∼2000 pixels/spot). Fig. 2(a) shows ΔI(λp) pixel value histograms for different concentrations.

To extract quantitative information, we implemented both binary classification and ensemble averaging methods on the contrast distribution data. Figure 2(b) shows a titration curve where the detection signal is the difference between the mean contrast values, Δ(ΔIavg(λp)), measured from negative control and N-protein spots (N = 6). Error bars correspond to the standard deviation from six different microarray spots measurements. Using a baseline that is calculated using a criterion of three standard deviations above the mean value of control measurements, the limit of detection (LOD) achieved by the ensemble averaging method is identified as 5 µg/mL. It has been previously reported [24] that the quantitative data from low concentration samples can be better captured using the binary classification method when compared to conventional ensemble averaging, which is susceptible to background noise. Thus, we re-processed the same dataset and plotted the titration curve, this time using the area under curve (AUC) parameter from the receiver operating characteristic (ROC) curves (shown in the inset of Fig. 2(a)). The titration curve shown in Fig. 2(c) reveals that the LOD can be advanced to 100 ng/mL simply by employing a binary classification method enabled by our digital imaging-based approach, which yields spatially rich detection information. Analytical platforms with a wide dynamic detection range are crucial for multiplexed detection of different biomarkers with varying clinically relevant concentration ranges. Therefore, our multimodal data processing approach is essential in multiplexed detection of antibodies as we can extend the dynamic quantification range by simply changing the data processing method, enabling the quantification of anti-SARS-CoV-2 antibodies from broad populations.

 

Fig. 2. Binary classification vs. ensemble averaging for quantitative data processing. (a) Images show individual S-protein-immobilized spots before and after detection of anti-S antibodies at various concentrations. A negative control spot is shown before and after incubation with a 100 µg/mL concentration of anti-S antibody solution. Before and after intensity differences at each spot are determined, and the probability density distributions of these differences at the pixel level are shown in histogram plots. The higher the concentration of the antibodies, the further the distribution average shifts from zero. The color-coded receiver operating characteristic curves of different concentrations that are calculated based on binary classification are shown in the inset. (b) Antibody detection calibration curve calculated based on ensemble-averaged intensity differences between the negative control spots and various antibody concentrations denoted as Δ(ΔIavg(λp)) in the inset. (c) The same dataset presented in (b) processed using binary classification and antibody detection calibration curve is shown in terms of area under curve values for each antibody concentration.

Download Full Size | PDF

Next, we investigated the effects of illumination wavelength, λp, on our imaging-based nanoplasmonic biosensor platform. Specifically, we compared the hyperspectral illumination method, where the λp is dynamically selected from a hyperspectral data cube based on the maximum contrast criterion against fixed single λp and fixed dual λp illumination configurations. Figure 3(a) depicts typical resonance peak spectra recorded before and after analyte binding in addition to their difference, which is the λp dependent intensity contrast function, Δ(ΔIavg(λp)). In the bottom plot of Fig. 3(a), Δ(ΔIavg(λp)) is shown for various analyte concentrations. Notably, these experimental data indicate that the optimal λp, which generates the maximum contrast, depends on the plasmonic resonance peak properties and analyte concentration. We observed a redshift in the optimal λp as analyte concentration increases. An ideal hyperspectral imaging-based optical setup gives access to dynamic λp selection; however, this comes at the cost of sophisticated instrumentation, which is usually not compatible with the POC tests. To address this challenge, we studied the single fixed λp = 650 nm and fixed double λp1 = 650 & λp2 = 657 nm illumination configurations, which can easily be implemented using low-cost off-the-shelf LED light sources or filters that are suitably chosen to match the resonance spectra. These illumination methods and the titration curves that are computed based on each method using the same experimental data are shown in Fig. 3(b). We compare the sensitivities achieved by two individual fixed single λp (Fig. 3(c)) and fixed dual λp configurations (Fig. 3(d)) against hyperspectral measurements and find that the fixed dual λp measurements highly correlate with the hyperspectral counterparts (Pearson’s R = 0.99). Our experimental findings underscore the vigor of the imaging-based intensity interrogation method enabling reliable, yet low-cost and portable biosensors without the need for complex optical instrumentation. Accordingly, all the results in the following sections were obtained using the dual wavelength method.

 

Fig. 3. Hyperspectral, dual, and single wavelength illumination-based intensity interrogation. (a) Plasmonic Au-NHA resonance peaks are acquired using hyperspectral imaging setup via averaged transmission signals at N-protein spots before and after incubation with anti-N antibodies. Green line shows Δ(ΔI(λp)), which is wavelength-dependent intensity contrast due to resonance peak redshift induced by antibody binding (top). Intensity contrast function, Δ(ΔI(λp)), for various anti-N antibody concentrations (bottom). (b) Antibody detection calibration curves are extracted at different illumination wavelength (λp) configurations. Hyperspectral single λp dynamically selects λp that generates maximum contrast ΔI(λp) for each antibody concentration, which requires hyperspectral imaging with tunable illumination source (top). Fixed dual (middle) and fixed single (bottom) methods probe intensity contrast variations at predetermined wavelengths simplifying the optical reader requirements. Imaging-based plasmonic biosensing performance comparison of hyperspectral vs. single (c) and dual (d) wavelength illumination configurations.

Download Full Size | PDF

Currently, quantitative serology often relies on ELISA, immunofluorescence, and chemiluminescence methods, which usually achieve quantitative results over a dynamic range by measuring multiple dilutions of the sample and correlating data to calibrators and standard curves. Since our approach directly measures the antibody presence, rather than label induced signals, it can robustly provide concentration values in weight/volume units over a wide dynamic range. To establish calibration curves that can be used to report meaningful quantitative results, we worked with a convalescent serum sample with high antibody levels reported by an ECL assay [14]. We first prepared a set of dilutions (see materials and methods) of this serum sample and measured them using our biosensor. Then, we spiked three different concentrations of commercial anti-N antibodies in diluted (1000x) sera in 1x phosphate buffer saline (PBS) solution and used the measurements from these known concentrations to convert dilutions into concentration units. Thus, we established calibration curves for anti-N and anti-S antibodies (Fig. 4) to be used in human sera evaluations.

 

Fig. 4. Calibration curves for SARS-CoV-2 anti-N (a) and anti-S1 (b) antibody detection using dual-wavelength acquisition configuration which are used in the quantitative analysis of human sera.

Download Full Size | PDF

We validated our biosensor platform on a cohort of human samples including COVID-19 convalescent sera (n = 18) collected at 5, and 12 weeks post symptom resolution, plasma samples collected from subjects with active or very recent COVID-19 (acute samples, n = 6), and COVID-19 naïve sera (n = 5) collected before the pandemic. First, we benchmarked our anti-N and anti-S antibody results from convalescent sera (n = 14) against values previously reported for immunoglobulin G (IgG) binding to N and S proteins using an ECL assay [14]. Even though our bioassay measures antigen-specific total immunoglobulins (Ig), we identified high concordance for both anti-N (R = 0.89) and anti-S (R = 0.87) Ig results between our biosensor and the ECL assay (Fig. 5(a)-(b)). Notably, our biosensor achieves a large dynamic range in a single label-free measurement directly from sera without the need of multiple dilution measurements.

 

Fig. 5. Validation of nanoplasmonic biosensor using COVID-19 patient samples. Correlation of electrochemiluminescence (ECL) assay and nanoplasmonic biosensor measured anti-SARS-CoV-2 N (a), and S (b) antibody levels from COVID-19 convalescent patient sera (n = 14). Anti-N (c) and anti-S (d) antibody measurements using the nanoplasmonic biosensor from COVID-19 naïve (n = 5), acute (n = 6), and convalescent 5 weeks (n = 6) and 12 weeks (n = 12) post symptom resolution (*p < 0.005 and **p < 0.00005 by ANOVA).

Download Full Size | PDF

Next, we present our results from the COVID-19 naïve, acute, and convalescent patient groups (Fig. 5(c)-(d)). Antibody binding to both N and S proteins was higher in COVID-19 acute and convalescent samples than in negative controls with p < 0.005 for N and p < 0.00005 (by ANOVA) for S proteins. In our cohort, S protein discriminates between COVID-19 convalescent and negative control sera (sensitivity= 87.5% and specificity= 100%) better than N protein (sensitivity= 95% and specificity= 60%), which aligns well with previously reported large cohort clinical studies [40]. Our results show that some negative control sera exhibit positive values in N protein binding, suggesting some degree of antibody cross-reactivity, which has been reported previously [40,41]. Some COVID-19 convalescent individuals’ low anti-S antibody levels measured 12 weeks after symptoms are likely due to low humoral response potentially in mildly symptomatic cases given relatively stable anti-spike antibodies over time [14,42]. This study implies that our biosensor platform has the potential to be used in large clinical studies to elucidate the development and persistence of anti-virus antibodies after infection or vaccination.

4. Conclusions

We demonstrated an integrated biosensor platform for multiplexed and label-free quantification of anti-SARS-CoV-2 antibodies from human sera. Leveraging the high refractometric sensitivity of nanoplasmonic Au-NHA photonic devices, we investigated multiple optical interrogation techniques, including hyperspectral, double, and single wavelength imaging approaches in transmission mode. Our experimental findings demonstrate that antibody detection achieved by simple bright field images acquired at two different illumination wavelengths, which are strategically chosen considering the nanoplasmonic resonance spectral properties, can yield similar performance to hyperspectral imaging (R = 0.98). This critical development eliminates the need for complex optical instrumentation, such as spectrometers and frequency sweeping light sources, which are unsuitable for portable point-of-care deployment of the platform. Furthermore, we showed a data processing workflow based on ensemble averaging and binary classification to increase the dynamic detection range (0.1-100 µg/mL), which is necessary for quantitative antibody detection from broad sample cohorts. To test our platform on real human samples, we simultaneously quantified antibodies against two important SARS-CoV-2 immunogens (S and N proteins) using only a small volume of human sera (50 µL). We report results from a cohort of COVID-19 acute (N = 6), convalescent (N = 18) and naïve (N = 5) human samples. The results demonstrated a high concordance with an electrochemiluminescence (ECL) assay [14] (Pearson’s R ∼0.89 and 0.87 for anti-N and anti-S, respectively), and could discriminate COVID-19 convalescent sera from naive controls with 88% sensitivity and 100% specificity. Moreover, our biosensor was able to capture the substantial heterogeneities in the humoral immune responses to SARS-CoV-2 antigens within our sample cohort of COVID-19 naïve, acute, and convalescent groups. Finally, we demonstrated a syringe-like optofluidic cartridge integrated with our nanoplasmonic sensor that can enable antibody quantification directly from fingerprick blood samples, leading the way for future on-site serology tests. Altogether, our results indicate that our multiplexed biosensor technology can fulfill analytical sensitivity and specificity requirements on a simple, portable, and cost-effective serological test platform. Our technology has the potential to be deployed for large epidemiological studies monitoring the humoral response to COVID-19 infections and vaccinations, which are critical to win our fight against the COVID-19 pandemic and to increase our communities’ readiness for future public health crises.