MultiSero: An Open-Source Multiplex-ELISA Platform for Measuring Antibody Responses to Infection

Abstract

A multiplexed enzyme-linked immunosorbent assay (ELISA) that simultaneously measures antibody binding to multiple antigens can extend the impact of serosurveillance studies, particularly if the assay approaches the simplicity, robustness, and accuracy of a conventional single-antigen ELISA. Here, we report on the development of multiSero, an open-source multiplex ELISA platform for measuring antibody responses to viral infection. Our assay consists of three parts: (1) an ELISA against an array of proteins in a 96-well format; (2) automated imaging of each well of the ELISA array using an open-source plate reader; and (3) automated measurement of optical densities for each protein within the array using an open-source analysis pipeline. We validated the platform by comparing antibody binding to Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) antigens in 217 human sera samples, showing high sensitivity (0.978), specificity (0.977), positive predictive value (0.978), and negative predictive value (0.977) for classifying seropositivity, a high correlation of multiSero determined antibody titers with commercially available SARS-CoV-2 antibody tests, and antigen-specific changes in antibody titer dynamics upon vaccination. The open-source format and accessibility of our multiSero platform can contribute to the adoption of multiplexed ELISA arrays for serosurveillance studies, for SARS-CoV-2 and other pathogens of significance.

Keywords: ELISA; SARS-CoV-2; multiplex; open-source; serology; serosurveillance.

Figure A1

Comparison between the Nautilus reader and a commercial plate reader. The same plate and wells were imaged using Nautilus (blue) and a SciREADER CL2 (red). Median background subtracted intensity (AU, left axis) and mean OD (OD, right axis) for RBD or S against an 8-point serial dilution of monoclonal antibody CR3022 is shown.

Figure A2

Evaluation of how comets affect measured ODs using duplicate ELISA-array wells. ODs from spots at the same locations in the array grid were compared across duplicate wells. (A) Schematic of example spot–spot comparison. The top-left non-fiducial spot in well A1 containing serum X was compared with the top-left non-fiducial spot in well F12 also containing serum X. (B) The data for one plate of duplicate sera are split according to the number of comets in the spot pairs: spot pairs in which one spot had a comet (orange); spot pairs in which both spots had comets (blue); and spot pairs in which neither spot had a comet (green). A y = x line is denoted on the plot (grey line).

Figure A3

Comparison of normalization methods for correcting biases and variances in ODs across plates. Two-dimensional distribution of duplicate OD values of antibody responses of RT-PCR positive sera to SARS-CoV-2 antigens. Spots with the same array location on duplicate plates are plotted against each other and then smoothed by the kernel density function to show the density of the data points (indicated by the brightness of the blue color; high brightness indicates low density). The spot OD values were normalized by dividing the mean of the reference spot ODs (anti-IgG Fc or fiducial spot) over each plate or well as indicated at the top. Duplicates with identical spot OD values will follow the function y = x (dashed line). Performance of different normalization schemes are quantified by the bias and variance of the normalized OD values across plates, which are defined by the mean of y − x and |y − x|, respectively. A total of 3 duplicates (6 plates) are shown. Normalization in this study provided only a small improvement; nevertheless, in cases where experiments are performed with different reagent lots or over a long period of time, xkappa-biotin (per plate) and xIgG (per plate) normalization methods should be evaluated.

Figure A4

Antigen-specific antibody responses classified by COVID-19 disease severity. multiSero OD values from sera obtained shortly after COVID-19 positivity, when vaccines were not available, are shown for asymptomatic (blue), symptomatic and not hospitalized (green), and symptomatic and hospitalized (red) individuals.

Figure 1

Overview of multiSero pipeline: (A) Samples containing antibodies derived from serum, saliva swabs, dried blood spots, or other sources are (B) overlaid onto an antigen array. Each spot in the array contains a concentration of a single protein or protein domain. The ELISA is performed and then the array is (C) imaged using the Nautilus plate reader or an alternative reader. (D) multiSero software is used to analyze the multi-antigen ELISA arrays. The software takes well images and a metadata file describing experimental conditions and imaging parameters as input (D1). The images are auto-cropped around the antigen array. Antigen spots are detected, and a grid is registered to the spots. Optical densities are computed from the spots that align with the registered grid. Optical density is computed as the −log of the ratio of spot intensity to the background intensity (D2). Sample antibody titers against each antigen in the array can be measured based on the ODs of control antibodies in the assay (D3).

Figure 2

The Nautilus, an open-source 96-well ELISA-array plate reader based on the Squid platform. (A) An early version of Nautilus, used to acquire data shown in Figures 3–6, Figure A1, Figure A2 and Figure A3. (B) A smaller form-factor Nautilus unit that incorporates an optional motorized focus adjustment. (C) CAD of the current Nautilus unit, used to acquire data shown in Figure 7.

Figure 3

ELISA array for SARS-CoV-2 and its analysis with multiSero software. (A) Image of a single well of a 96-well plate and its corresponding array layout. Antigens included in the array are color coded as RBD (green), S (ochre), N (pink), anti-kappa-biotin (fiducials, black), anti-human IgG-Fc (mauve), and GFP-foldon (grey). (B) Grid registration for spot quantification. The center points of all spots in the cropped image (left panel) are detected first. A coordinate transformation that registers the initial guess for the fiducials (middle panel) with the detected fiducial spots is calculated. A grid containing all spot locations in the antigen layout is then transformed onto the image using the coordinate transformation (right panel). (C) Spot detection in the presence of comets. The original well image (left panel) shows spots that exhibit comets (white arrows). The output of Scienion spot detection and analysis (middle panel) is overlayed onto the original image, showing spots that were successfully measured (light green) or that were missed (black). The size of the spot indicates the area analyzed by Scienion. The output of multiSero spot detection and analysis (right panel) shows the final grid registration (dark green) used for spot measurement. The size of the grid registration spots are unrelated to the area analyzed.

Figure 4

Antigen-specific antibody responses in COVID-19-positive and negative cohorts. multiSero OD values from sera obtained shortly after COVID-19 positivity, when vaccines were not available (grey dots), are shown for RBD (250 picograms (pg)), S (62.5 pg), N (50 pg), anti-IgG-Fc, GFP-Foldon, and anti-kappa-biotin. COVID-19-positive (“Positive”, 93 samples) and negative (“Negative”, 87 samples) mean OD values are plotted, along with the cohort-wide mean OD and standard deviation (black lines). The sera that showed the highest (green dots) and lowest (red dots) binding to RBD, in the COVID-19-positive cohort, are also shown for all other antigens. For anti-kappa-biotin (fiducial, secondary antibody), the single spot highlighted (red dot) corresponds to the sample that gave low antibody binding across all other antigens. All sera were assayed at a 1/200 dilution.

Figure 5

Specificity and sensitivity of antigen-specific antibody responses. (A) ROC curves (black line) for classifiers that use RBD, S, or N antibody responses. Confidence intervals of 95% of the area under the ROC curve (AUC) are reported below each curve. (B) Specificity (red line) and sensitivity (blue line) as a function of multiSero OD values. The dotted lines represent a cut-off value equal to 3 standard deviations above the COVID-19-negative mean for that antigen. Data for Figure 5 are identical to those used for Figure 4 (93 COVID-19-positive samples and 87 COVID-19-negative samples).

Figure 6

Comparison of multiSero to commercial SARS-CoV-2 antibody assays. (A) Heatmap of Spearman’s rank correlation between multiSero ODs and commercial SARS-CoV-2 antibody tests. Values inside the cells correspond to the Spearman’s rank correlation coefficient. Commercial assays (y-axis) are abbreviated as: Abbott ARCHITECT SARS-CoV-2 IgG (N-Abbott), Roche Elecsys anti-SARS-CoV-2 total (N-Roche), Ortho Clinical Diagnostics VITROS anti-SARS-CoV-2 total (S-Ortho Ig), Ortho Clinical Diagnostics VITROS anti-SARS-CoV-2 IgG (S-Ortho IgG), DiaSorin LIAISON SARS-CoV-2 S1/S2 IgG (S-DiaSorin), and Monogram PhenoSense Assay (Neut-Monogram). (B) Scatter plots of multiSero OD values and S-Ortho IgG or N-Abbott calibrator result index (S/C) values. Data shown are from the paired multiSero and commercial assay outlined in panel (A).

Figure 7

Comparison of antibody responses over time in vaccinated and not vaccinated cohorts. multiSero OD values from sera obtained shortly after COVID-19 positivity, when vaccines were not available (blue dots, “pre-vaccine availability”), and matched sera at a later time point when vaccines were available (red dots, “post-vaccine availability”) are shown for vaccinated (top panel) and not vaccinated (bottom panel) individuals. Grey lines connect matched individual sera. Black points denote matched post-availability minus pre-availability OD values, with black lines showing the mean of the difference and standard deviation. The dashed black horizonal line corresponds to zero OD difference. All sera were assayed at a 1/25,600 dilution, and the amount of each printed protein (RBD, S, N, anti-IgG-Fc, and GFP-foldon) for this analysis is identical to that of Figure 4.