Infection by SARS-CoV-2 virus (SARS-2) occurs mainly through inhalation of virus-laden respiratory droplets . When the virus reaches the lower lung airways, ACE2-expressing pneumocytes-II epithelial cells take up SARS-2 through interaction between host cell ACE2 receptors and the viral spike-protein (S-protein) [2-4]. Thus, neutralization is mainly achieved by antibodies (Abs) that interrupt the binding of ACE2 receptors and SARS-CoV-2 S-protein receptor binding domain (RBD) [3-5]. To fast-track the development of effective vaccines against SARS-CoV-2, surrogate high-throughput efficacy tests are urgently needed. Neutralization efficacy screening of vaccine candidates can be expedited by abstracting/distilling neutralization evaluation into a binding study between ACE2 and S-protein, or its RBD domain, in the presence or absence of vaccinated animal or human immune serum [6-8].
ELISA is a simple, high-throughput serological assay, in which antigens are coated onto disposable microtiter plates, and antigen-specific Abs from immune serum are allowed to bind to coated antigens. Binding is then detected by a chromogen-labelled secondary (2ry) Ab/probe [9-11]. ELISA is not only used to evaluate antigen-specific immune responses, it is also used to study binding affinity and kinetics, singularly or competitively, to extract meaningful binding values, e.g. the binding dissociation constant (KD) or binding extent [6-8]. Competitive ELISA (CELISA) demonstrates binding competition between two proteins that separately bind to the same location on an antigen [9-11]. However, when both competitor proteins are present, the protein with the more favorable binding characteristics and affinity binds preferentially and inhibits the competitor protein’s binding, as they sterically clash with each other.
Plaque reduction neutralization (PRNT) and virus neutralization (VNT) cell-based assays are considered to be the ‘gold standards’ for evaluating the neutralization efficacy against viruses [5, 6, 12]. Because these assays simulate an in vivo environment, with host cells and viruses present in biological media, they are capable of demonstrating the efficacy of different immune sera or antivirals, and their results correlate well with in vivo protection against infection challenge . However, cell-based assays are not feasible for high-throughput screening of large numbers of sera, as they are laborious, require special biosafety containment and are subject to long culturing and incubation times for host cells with live infectious viruses [14, 15]. Cell-based surrogate neutralization assays are also relative in their assessment of neutralization efficacy: outcomes may vary depending on assay set up, e.g. seeded cell counts, ACE2 expression levels, and virus dose, even when the same assay is used. Therefore, external standards, for example, known neutralizing Abs, are often used with each assay procedure to allow for comparison of oucomes. Despite variations, when proper controls and standard neutralizing Abs in cell-based assays are used, the assays generally correlate well with each other and with the protection afforded in infection challenge [5, 16, 17]. It is important to note that CELISA is not intended to replace PRNT or VNT assays, as they are more biorelevent, i.e. they more closely represent natural infection at the cellular level in complex biological environments. However, CELISA assays are well-suited to screen a large number of sera in a short time, can be fully-automated, and follow a simple, straightforward procedure, which does not require sterile conditions, special containment, or time consuming cell culturing. CELISA is a very efficient tool to fill the gap between initial screening of large numbers of immune sera from animals immunized with various vaccines and pinpoint highly effective vaccine candidates in a short time. Thus, we aimed to develop an assay that correlates with the well-established gold standard neutralization assessment assays for accurate, quantitative determination of neutralization efficacy/potential.
Comparison to Other Competitive ELISA Assays
There are several configurations of CELISA assays (Figure 1), reflecting the variation in reagents and protein concentrations and total applied quantities, order of addition, and procedures. In early CELISA [18-21] designs, wells were coated with RBD, followed by the addition of immune serum, but ACE2 as an RBD-binding competitor protein was not employed (Figure 1A). Instead, a known nAb was applied afterwards to determine whether the newly investigated serum, or purified Abs, competed for the same neutralizing epitope. However, protective antibodies (Abs) may independently target different neutralizing epitopes, thus, do not exhibit competitive binding to the known nAb. Further, while Abs that bind competitively to the same epitope sterically clash with each other, this does not prove inhibition of ACE2 binding, as critical ACE2-binding residues within the shared epitope may not be challenged by the tested Ab or serum. A neutralizing epitope was recently shown to overlap with an infection-enhancing epitope in a SARS-CoV pre-clinical study on macaques , thus steric clashing of antibodies against these epitopes does not guarantee ACE2-RBD binding neutralization.
Eventually, ACE2-coated plates were employed, and a mixture of RBD and neutralizing serum was added to the plates [17, 18] (Figure 1B). The advantage of this configuration is that it requires fewer processing steps. However, it was difficult to use controls, as RBD-neutralizing Ab binding inhibited ACE2-RBD binding. Thus, bound, or excess unbound RBDs, were washed away with the serum before 2ry Abs were added. Therefore, serum interaction with RBD, whether as non-specifically bound or excess unbound molecules, could not be detected or confirmed by controls. This assay strategy also requires manipulation of serum/neutralizing Ab concentration, serum to RBD ratio, and mixing with RBD before it is added to the ACE2-coated plates. Thus, neutralization (%) can be easily evaluated only at very few high serum concentration values; as RBD and sera have to be mixed first before addition to the wells. In addition, nAb/RBD mixing time is critical, as binding kinetics influence the extent/percentage of RBD-nAb binding over the time [23, 24]; as such the mixing/incubation time requires further investigation in assay strategy B. In contrast, standard ELISA procedures guarantee antigen-specific binding to reach true equilibrium, even for viscous sera, after 1-2 hours of incubation. Lastly, in He Y. et al, the concentration of purified Abs that were employed in this assay strategy were high (50 µg/mL). This would have affected binding affinity and might even have allowed for the binding of weakly neutralizing Abs .
Recently, a more complicated RBD/S-protein coating was employed. Rabbit anti-histidine Abs (1 µg/mL) was first coated on microtiter plates, then His6-RBD or His6-S-protein (10 µg/mL) solution was added to the wells in a similar manner to sandwich ELISA (Figure 1C) . Sera or purified Abs were subsequently added to neutralize indirectly-coated RBD or S-protein. Human IgG-Fc-conjugated ACE2 (hFc-ACE2) was applied, followed by 2ry Abs and substrate. This strategy is similar to our approach (Figure 1D) [8, 25], although the initial anti-His coating step is not included in our strategy since RBD or S-protein binds well to ACE2 when coated directly onto the plate , however, no additional controls were employed, in strategy (C), to determine the amount of indirectly bound S-protein. Assays that use anti-His coating require extra time and effort due to the addition of competitor ACE2 protein, which is generally longer than standard ELISA. This protocol used an unknown quantity of S-protein, which was indirectly coated on the plates, and suboptimal amounts of ACE2 (0.1 µg/mL). Low quantities of ACE2 (below saturation) could result in the presence of unbound S-protein, i.e. neither bound by ACE2 nor by neutralizing Abs. This would influence total optical density (OD) readings, because neutralization and ACE2 binding could occur at the same time. Therefore, in our assay strategy we used enough ACE2 to reach saturation to ensure the assay’s sensitivity for neutralization detection.
Currently, the most common CELISA strategy [6, 7], which has also been reported by our group , involves allowing RBD- or S-protein-coated plates to interact with immune sera, i.e. become neutralized. Thereafter, hFc-ACE2 (or biotinylated-ACE2, for testing human sera) is added, followed by anti-hFc 2ry Ab-HRP (or streptavidin-HRP for human serum) and substrate. This strategy involves sequential addition of binding proteins (serum antibodies, 2ry antibody and ACE2), thus facilitates full use of controls. This enables the detection of non-specific interactions, e.g. between RBD and serum proteins, such as complement proteins and detects cross-reactivity of 2ry Abs with primary/murine Abs or residual serum proteins, e.g. by including blank ACE2 control, . In the reported tests, 100 ng of RBD (50 µl of 1 µg/mL solution) was used for plate coating; however, only 50 ng of ACE2 (100 µl of 1 µg/mL solution) was added. Despite that, a 4- to 5-fold increase in ACE2 quantity was required to achieve equimolar saturation of coated RBD (Figure 1D). Through the development of our assay, we found that 50 ng of directly coated RBD provided a sufficient OD450 signal, even when it was partially bound to ACE2. The resulting signal was within the linear range of Beer-Lambert’s Law, and protein sparing as well, because 200-220 ng of ACE2 is needed, per well, to saturate coated-RBD and result in an OD450 signal (plateau) (Figure 2).
All reported assay strategies have assumed linear calculations, as the signal OD value is divided by a blank and then subtracted from 100% to yield percent neutralization. This assumes linear RBD-ACE2 binding or competition, as well as full ACE2 saturation at low serum concentration  or in the absence of serum . However, all of these assumptions are incorrect (Figure 2). While Abe et al.  employed a different calculation approach, they used the area under the curve (AUC) displacement as an independent neutralization assessment parameter in their CELISA as an alternative relative neutralization measurement.
While potently neutralizing sera can maintain neutralization at dilutions of up to 1/200 – 1/1000, we found significant positive interference (>10% OD450 of non-neutralizing control) in assays with high serum concentrations (>2%) if the serum was not heat-treated (not shown). The correlation between neutralization results of CELISA and pseudovirion cell entry assay was R2=0.76, but this dropped to 0.6 between CELISA and PRNT assays (VNT equivalent), potentially due to serum-associated interference . Ultimately, none of the reported assay methods or protocols employed heat inactivation of the sera, even at high concertation’s, nor did they employ calibration curves to evaluate diluted immune serum signals. We found that these two steps were essential for the accurate determination of the neutralization efficacy of immune sera.
CELISA Assay Development
Several critical aspects of CELISA assay development were not sufficiently considered in previous assay designs. These include, a) the ACE2 solution concentration (> KD) employed, b) the total amount of ACE2 applied (to saturate bound RBD), c) the use of a non-linear calibration curve for calculations (considering the inaccuracies introduced from a linear assumption), d) maximizing the sensitivity of the OD signal through adjustment of the RBD quantity while maintaining the optimum RBD/ACE2 ratio, e) ensuring assay specificity by eliminating non-specific binding interference of high serum concentrations to added ACE2 or 2ry Abs, and f) ensuring that the conventional validation parameters conformed to standard assay specifications by evaluating the accuracy, precision, quality of non-linear fit, specificity, and sensitivity (i.e. detection and quantification limits). In order to ensure that the binding equilibrium is shifted towards ACE2 binding of RBD, unless nAbs are present, the concentration of added ACE2 solution must exceed the KD of ACE2-RBD binding for a given coated amount of RBD. Otherwise, even in presence of ACE2, minimal binding would be expected, or the equilibrium could shift towards RBD binding to a weakly neutralizing competitor Ab, instead of ACE2. The minimum ACE2 concentration required for favorable binding equilibrium to coated RBD at 50 ng/100 µL per well was KD=10 ng/100 µL (0.083 nM) for Fc-ACE2. However, this still does not ensure the saturation of all coated RBD with ACE2. Therefore, the concentration should exceed the KD and the binding extent should be near saturation, i.e. all coated and accessible RBDs (26.6 kDa) should bind to Fc-ACE2 (120 kDa). Thus, an equimolar ratio per well of both proteins may be employed (RBD/ACE2 ratio of 0.23). Ultimately, Fc-ACE2 total mass per well should be 4- to 5-fold the amount of coated RBD to ensure maximum coverage/saturation. We employed this ratio in our binding curve (Figure 2), as this was the maximum amount of ACE2 that resulted in a significant increase in the OD450 signal.
When using lower amounts of RBD, which has commonly been done, accessible, unbound RBDs are available in the presence of a suboptimal amount of ACE2, resulting in simultaneous, non-competitive Ab-neutralization and ACE2/RBD binding. However, due to the non-linearity of binding approaching saturation, changes in OD readings are very minor (at the plateau), while the amount of ACE2 still increases. Therefore, this may significantly influence readings near the saturation point. To solve this issue, standard neutralizing Abs with known neutralization efficacies (N50 or IC50) must be employed, or a saturating amount ACE2 used. Otherwise, significant differences between the OD readings (signal replicates) from different amounts of bound ACE2 may be observed, reducing the accuracy of the assay. Binding equilibria are non-linear and follow a sigmoidal pattern (Figure 2), even in the absence of a competitor probe, e.g. neutralizing Abs. Therefore, accounting for neutralization by the missing or unbound ACE2 using only the difference between OD readings and the blank (absent serum) plate wells assumes a linear relationship between bound ACE2 (%) and OD readings (Figure 2), which is grossly inaccurate. This linear relationship assumption in the calculation is, unfortunately, common across all reported forms of the CELISA assay, and this is likely one of the main reasons for the technique’s limited correlation with cell-based assays.
OD readings are relative for every measurement, as they depend on the substrate-enzyme type, properties, concentration, and reaction time. Thus, a calibration curve or external standard is required for each measurement. This has also, unfortunately, not been employed in any of the reported CELISA assays, as they falsely assumed that the highest reading corresponded to 100% bound ACE2, and that values less than this maximum represented unbound ACE2, i.e. full neutralization of immobilized RBD in a linear fashion. This is incorrect due to a number of factors, including positive interference from non-specific binding. Therefore, a calibration curve must be established to quantify the amount of bound or free ACE2 as an external standard for the determination of neutralization (%).
The total amount of coated RBD should be high enough to obtain a clear OD readout signal, thus increasing test sensitivity. However, it should not be high enough to exceed the Beer-Lambert cut-off of 0.9 absorbance unit in order to avoid convoluting the calculations further with an additional non-linear component. The immune serum must not interact with any of the ingredients, so as not to impart significant interference. This last point is very important, as we found that serum non-specifically binds to ACE2 and/or 2ry Abs at concentrations exceeding 2-5%, even with extensive plate washing. We found in the early development of the assay that OD readings from sera wells at various serum concentrations (RBD+5-20% serum+ACE2+2ry Abs) could be twice as high as maximum OD readings from blank serum wells in the calibration curve (100% RBD+ACE2+2ry Ab) that should have maximum theoretical OD signal. This shows that serum complement activation at high concentrations non-specifically traps ACE2. Similarly, irrelevant serum from a PBS-immunized group of mice (n=5) at high concentration (2-20%) was also found to result in a low but significantly high OD reading interference in blank ACE2 control (RBD+2-20% serum+2ry Ab) compared to background OD readings, despite the absence of antigen-specific Abs and ACE2. This shows that complement activation could also trap 2ry Abs. Therefore, as commonly used in biotechnology techniques, heat inactivation of the serum was conducted to eliminate this non-specific binding/trapping effect. Following heat inactivation, the background interference was minimal (<5% of total OD), even at the highest employed serum concentrations (1/10 dilution), thus increasing assay specificity and sensitivity to neutralization detection and accurate quantification. Another alternative technique involves purifying and isolating RBD-specific IgGs to evaluate them directly in the assay. While purified Abs are easier to evaluate, serum testing requires these precautions and a number of controls to ensure that no significant interference or nonspecific interactions are present. Moreover, conventional validation parameters and compendial limits are also important to investigate, even if their boundaries are often less strict for immunological and biological techniques because of their inherent variability. Important validation parameters include accuracy, specificity, precision, sensitivity, and non-linear calibration curve fit quality. Finally, serum assay results should correlate and conform to gold standard cell-based neutralization assays to prove validity and usefulness as an alternative high-throughput screening technique for neutralization efficacy determination.
The relationship between binding affinity/extent and added total substrate/binder protein concentration kinetics follows a non-linear sigmoidal pattern “exponential rise to plateau”. Thus, the simplest mathematical function to represent binding is the Michaelis-Menten adsorption equation. Alternatively, the more flexible four-parameter logistic sigmoidal function, which results in a better fit with more complex binding affinity curves, could also be used:y=[(A1-A2)/(1+(x/xo )p ]+A2, where A1 and A2 are the minimum and maximum measured/found OD signals that correspond to the amount of bound ACE2, p is the power exponent representing the growth rate of the signal as the binder protein concentration increases, and xo is the x-axis ACE2 concentration value at 50% binding, i.e. half-height of the maximum OD signal (Figure 2).
The amount of bound ACE2 can only be determined via constructing and interpolating an ACE2-RBD binding calibration curve. Employing ACE2 concentrations below saturation will reduce test sensitivity to detect neutralization. While applying excess ACE2 beyond saturation, will not change signal, at the plateau (Figure 2), thus wasting expensive ACE2 protein and could be disrupting to the binding equilibria. Therefore, the ideal amount of ACE2 is at the beginning of the RBD/ACE2 binding plateau, where the differences in OD readings beyond which becomes non-significant statistically, e.g. 250 ng/100 µL concentration point in figure 2. This can be calculated experimentally or provided from the supplier’s RBD certificate of analysis effective 50% binding concentration/molar ratio to RBD value, if available. Finally, the OD-reading should be adjusted to lie within a linear absorbance range of 0.1-to-0.9, in accordance with Beer-Lambert’s Law; as greater values will impart non-linearity to the response and unnecessarily complicate the calculations.
In our procedure, we took into account all eight of the issues mentioned above. We conducted our CELISA assays using the configuration presented in Figure 1D and Figure 3 on several immune murine sera to evaluate neutralization efficacy. We interrogated the procedure using conventional assay method development validation parameters, such as non-linear calibration curve fit and significance, precision, accuracy, specificity, and sensitivity in terms of the limit of quantification and detection of neutralization. Lastly, we conducted a standard cell-based neutralization efficacy assay (VNT assay) to compare and correlate the neutralization efficacies of both tests.
Evaluation of Conventional Validation Parameters
We adopted rigorous compendial analytical validation parameters to qualify the neutralization detection and quantification capabilities of this assay. Non-linear calibration curve fit quality is important to confidently determine and interpolate the neutralization values of each well at each serial serum dilution, as demonstrated in procedure. Goodness of fit was evaluated by correlation coefficient value (R2>0.95). Further, standard deviations between each two successive mean bound ACE2 amounts in the calibration curves (n=5) was statistically significant (p < 0.05, paired student t-test). This ensures that there is no overlap between two successive neutralization values. Our procedural set-up confirms significant differences and excellent fitting of logistic function. Our 4P logistic fit of the calibration curve gave correlation coefficient (R2) values that were often in the range of 0.98 to 0.99. Precision or repeatability evaluation was conducted on four different serum samples with various neutralization efficiencies through determination of the 50% neutralization (N50) and percent neutralization extent (N%) values. N% is the percent reduction/neutralization of ACE2-RBD binding at a given serum dilution, and N50 is the serum dilution (or antibody concentration) responsible for reducing/neutralizing ACE2-RBD binding to 50%. The N50 and N% for each serum (at different dilutions) had coefficients of variance below < 5%. Since the calibration curve should simulate neutralization conditions to different extents (N%), we immobilized different amounts of RBD on ELISA plates. This is because when neutralization occurs, the RBDs become inaccessible to ACE2, i.e. less RBDs are available, so we deliberately reduced the amount of coated RBDs.
Accuracy was determined by adding irrelevant serum (which does not bind to RBD) to the calibration curve with various coated-RBD concentrations, ranging from 2.5 to 50 ng/100 µL/well. These corresponded to 95% to 0% of neutralization, respectively. The blank (subtracted) calibration curve OD450 readings (in the presence of irrelevant murine serum) remained within ± 5% of the labelled bound ACE2 (%) standard value (in absence of the serum), without statistically significant differences (student t-test, p > 0.05, n=4). Specificity was evaluated by testing the neutralization potential of a non-neutralizing irrelevant serum sample and determining whether it had significant neutralizing response at any dilution. The non-neutralizing serum had no detectable neutralization at the highest concentration and the level of interference was below 2%, which was attributed to acceptable random error. The sensitivity of the assay was evaluated by its capacity for quantitative detection of neutralizing sera. The neutralization detection limit was calculated from the mean background OD value plus three standard deviations (OD450=0.056), which corresponded to 0.3% bound ACE2. The quantitation limit was the mean background OD value plus 10 standard deviations (OD450=0.083), which corresponded to 7% bound ACE2. These represent very sensitive detection and quantification limits (Figure 1D); as the lowest OD signals result from the lowest amount of bound hFc-ACE2 (high neutralization extent), thus, in turn, have the lowest amount of bound 2ry Abs, which are responsible for the color change.
Virus Neutralization Assay: Orthogonal Validation
Twenty-five individual murine immune sera were tested (n=25) for neutralization using both VNT and CELISA assays. Mice were immunized with various SARS-CoV-2 vaccines or a PBS negative control. Group mean neutralization extent (%) was calculated using both assays by averaging the neutralization efficacy (%) for each serum dilution (1/20, 1/40, or 1/80) (Figure 4A). CELISAs were additionally conducted with and without heat treatment (Figure 4B-C). Finally, serum neutralization profile and dilution value corresponding to N50 were determined by both assays and compared (Figure 4D). The calculation protocol for the CELISA technique is provided in the detailed protocol.
VNT assays were conducted as described previously [26, 27]. Briefly, overlay medium, block buffer, and polysorbate/phosphate saline (wash buffer) were prepared. The primary probes (mouse sera) were heat inactivated. Vero E6 cells were cultured in DMEM medium. SARS-2 virus isolate (QLD02, GISAID accession EPI_ISL_407896) was used for this assay. Vero E6 (5 x 104) cells were seeded in 96-well plates with DMEM medium, and incubated overnight at 37°C and 5% CO2. After incubation, the medium was removed, heat inactivated mouse serum was serially diluted 5-fold, and viral inoculum (~260FFU/well) was incubated with serially diluted sera for 1 hour at 37°C. In a similar process to the mouse serum, final concentration of 10 µg/ml of the nAb, S309 , was serially diluted 5-fold, then incubated with similar amount of SARS-CoV-2 virus. The mixture (50 µl) then were added to each well of the cell plates to infect the cells. The overlay medium was added onto the cells, and the plates were re-incubated for 14 hours at 37°C and 5% CO2 prior to fixing the cells with 80% acetone. The plates were dried, blocked using blocking buffer for 1 hour at room temperature. The block buffer contain milk diluent sera (KPL, Seracare) and 0.1% Tween in PBS. Plates were then probed with anti-spike antibody (CR3022)  and followed by IR dye®-conjugated 2ry Abs (both diluted in blocking buffer) were added to each of the seeded cell wells. The plates were read using an Odyssey Infrared Imaging System infrared high-resolution scanner LI-COR CLX. Spots denoting the number of infected Vero E6 cells were counted using the procedure below.
VNT assay plate spots (signaling viral cell entry; Supporting Figure 2) were counted using ImageJ Fiji software version 1.53e. The plate images were cropped, and the color threshold was adjusted to the default settings: black and white (B&W), in RGB color space with the parameters: Red=33, Green=74, and Blue=49, against a dark background. The diluted serum wells across all plates were processed using the same parameters, including immune serum plates and the naïve serum plate. After threshold adjustment, we conducted particle counts for every well. The counts were divided by virus only (blank serum) positive control numbers to yield N% at a given dilution. For individual sera that exceeded 50% neutralization values at 1/20 or 1/40 dilutions, the reciprocal serum dilution that corresponded to 50% neutralization was interpolated to yield N50 values.
Correlation between VNT assay neutralization and CELISA using untreated individual mouse serum (R2=0.54) was much weaker compared to that of heat-inactivated individual sera (R2=0.975, n=25) or combined group mean neutralization (R2=0.98, n=10; Figure 4A-C). The overall weak-to-moderately neutralizing murine serum neutralization 50% efficacy values (N50) between both CELISA and VNT assays correlated very well (R2=0.94, n=6), with the slope approaching unity (Figure 4D).
Our novel assay strategy is not limited to the detection of RBD-bound nAbs. N-terminus domain-bound nAbs can be determined by coating the plates with 100 µL of 3.32 µg/mL proline-substituted S-protein instead of RBD (21.23 kDa, on amino acid Mwt basis). This is an equimolar concentration to our assay RBD concentration, as S-protein monomer (141 kDa, on amino acid Mwt basis) contains a single RBD, while the trimer (423 kDa, on amino acid Mwt basis) contains three RBDs. An initial quick binding affinity titration may be required to establish the optimum S-protein coating concentration that would yield a suitable OD reading range for detection and quantification. A similar 6.6-fold increase in coating solution concentration using full-length S-protein, as substitute for RBD, was suggested in a recently published CELISA assay to yield comparable results to RBD-coated plates . However, the limitation in all currently employed CELISA assays, including ours, is its lack of detection of priming process inhibitory Abs compared to cell-based assays. These Abs prevent furin and cathepsin-L enzymes from priming spike protein; however, they do not prevent RBD/ACE2 binding, so a different assay is required to evaluate this neutralization mechanism.