Going beyond clinical routine in SARS-CoV-2 antibody testing - A multiplex corona virus antibody test for the evaluation of cross-reactivity to endemic coronavirus antigens

Given the importance of the humoral immune response to SARS-CoV-2 as a global benchmark for immunity, a detailed analysis is needed to (i) monitor seroconversion in the general population, (ii) understand manifestation and progression of the disease, and (iii) predict the outcome of vaccine development. Currently available serological assays utilize single analyte technologies such as ELISA to measure antibodies against SARS-CoV-2 antigens including spike (S) or nucleocapsid (N) protein. To measure individual antibody (IgG and IgA) responses against SARS-CoV-2 and the endemic human coronaviruses (hCoVs) NL63, 229E, OC43, and HKU1, we developed a multiplexed immunoassay (CoVi-plex), for which we included S and N proteins of these coronaviruses in an expanded antigen panel. Compared to commercial in vitro diagnostic (IVD) tests our CoVi-plex achieved the highest sensitivity and specificity when analyzing 310 SARS-CoV-2 infected and 866 uninfected individuals. Simultaneously we see high IgG responses against hCoVs throughout all samples, whereas no consistent cross reactive IgG response patterns can be defined. In summary, our CoVi-plex is highly suited to monitor vaccination studies and will facilitate epidemiologic screenings for the humoral immunity toward pandemic as well as endemic coronaviruses.


Introduction
Given the importance of the humoral immune response to SARS-CoV-2 as a global benchmark for immunity, a detailed analysis is needed to (i) monitor seroconversion in the general population 1,2 , (ii) understand manifestation and progression of the disease 3 , and (iii) predict the outcome of vaccine development 4,5 . Currently available serological assays utilize single analyte technologies such as ELISA to measure antibodies against SARS-CoV-2 antigens including spike (S) or nucleocapsid (N) protein 1, [5][6][7] . To measure individual antibody (IgG and IgA) responses against SARS-CoV-2 and the endemic human coronaviruses (hCoVs) NL63, 229E, OC43, and HKU1, we developed a multiplexed immunoassay (CoVi-plex), for which we included S and N proteins of these coronaviruses in an expanded antigen panel. Compared to commercial in vitro diagnostic (IVD) tests our CoVi-plex achieved the highest sensitivity and specificity when analyzing 310 SARS-CoV-2 infected and 866 uninfected individuals.
Simultaneously we see high IgG responses against hCoVs throughout all samples, whereas no consistent cross reactive IgG response patterns can be defined. In summary, our CoVi-plex is highly suited to monitor vaccination studies and will facilitate epidemiologic screenings for the humoral immunity toward pandemic as well as endemic coronaviruses.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint Next, we used an extended sample set with 310 SARS-CoV-2-infected and 866 uninfected donors for clinical validation of CoVi-plex. A simplified overview of this set is shown in Fig. 2a; a complete breakdown is displayed in Extended Data Table 2. A direct comparison revealed that Spike Trimer and RBD were the best predictors of SARS-CoV-2 infection. Signal cut-offs were defined for both, IgG and IgA detection, based on ROC analysis with focus on maximum specificity. Additionally, we defined a cut-off for overall IgG and IgA positivity for which both individual cut-offs for Spike Trimer and RBD had to be met (Fig. 2b). As shown above, cut-offs based on IgG were shown to be more sensitive and specific than those based on IgA. With the IgG overall cut-off, we reached a specificity of 100 % (Fig. 2c), which would not have been possible for either of the antigens individually, while still retaining acceptable sensitivity. To identify samples with an early immune response, we simultaneously measured IgA response.
With CoVi-plex, we identified eight IgA-positive samples that showed no IgG response (Fig.   2d). Two of these were uninfected and falsely classified as positive. For four of the remaining six infected samples, details regarding the time between the onset of symptoms and sample drawing were available (2, 6, 7, and 15 days). We hypothesized that IgA in these samples can be used to measure an early onset of antibody response. Thus, we classified samples with strong IgA positivity -signal to cut-off (S/CO) > 2 for Spike Trimer and RBD -as "positive", irrespective of their detected IgG response. With this combined IgG + IgA classification, we reached an optimal sensitivity of 90 % while retaining a specificity of 100 %.
Further analyzing the Ig response towards both subdomains of the spike, S1 and S2, we achieved no additional sensitivity for the classifier (Fig. 2e). Interestingly, RBD, as a part of S1, showed much fewer uninfected samples with increased IgG response compared to S1.
For S2 even more uninfected samples had increased signals, pointing to potential crossreactivity in this domain of the spike protein (Fig. 2e). We further complemented our assay with the N and N-NTD proteins. Although these antigens were successfully used in singleanalyte assays 11 , we observed a high cross reactivity in uninfected samples for both (Fig. 2f).
Across the entire data set, only one sample showed a distinct immune response to N and N-NTD, but not to all spike derived antigens.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint Longitudinal samples from five hospitalized patients were subjected to a small-scale time course analysis of IgG and IgA immune responses (Fig. 3a). Levels of both Ig classes strongly increased within the first ten days after the onset of symptoms. While IgG levels appeared constant over roughly two months, IgA levels started to decline between day 10 and 20 after the onset of symptoms where samples were available. These effects were consistent for the majority of SARS-CoV-2 antigens. Furthermore, we found that patients´ hospitalization, as a measure of disease severity (Fig. 3b), seemed to correlate with an increased humoral immune response, especially in IgA. Furthermore, there is indication for a trend for increasing age as well (Fig. 3c). However, it should be considered that patients of higher age also had a higher rate of hospitalization in our study population.
In order to explore cross-reactivity of hCoVs with SARS-CoV-2, we included S1, N, and N-NTD antigens from human α-(NL63 and 229E) and β-hCoVs (OC43 and HKU1) in our CoVi-plex panel (Extended Data Fig. 1). The immune response towards all hCoV antigens was more dependent on coronavirus clade than on N or S1 antigen. However, within the clades of α-hCoVs and β-hCoVs, types of antigens were more dominant than the virus subtype, as demonstrated by rank correlation analysis and hierarchical clustering. Interestingly, IgG response against α-hCoVs clustered more closely to SARS-CoV-2 than to β-hCoVs (Fig. 4a,   Extended Data Fig. 4a). Overall, we identified a considerable immune response to hCoV antigens throughout the whole sample set with no notable differences between samples from SARS-CoV-2-infected and uninfected donors in IgG or IgA for S1 (Fig. 4b), N (Fig. 4c), or N- Fig. 4b).
We therefore used the IgG signal relative to the average response per antigen for further analyses, which allowed comparison among all hCoV antigens on one scale. For those uninfected samples, which showed an IgG cross reactivity towards Spike Trimer (Spike Trimer false positives), we observed partially increased responses towards hCoV antigens. Those samples, which did not show an immune response after SARS-CoV-2 infection (false negatives, as determined by CoVi-plex, combined IgG + IgA) were closer to the baseline ( Fig.   4d-e, Extended Data Fig. 4c). This indicates that cross-reactivity with hCoVs causes some of . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint the observed SARS-CoV-2 immune response in samples taken from individuals not exposed to SARS-CoV-2.
To investigate the correlation of hCoV and SARS-CoV-2 immune response further, we grouped samples into high and low responders for α-hCoVs and β-hCoVs. High responders had relative IgG signals > 0 for N and S1 antigens of both hCoV subtypes within the clade, while lowresponders had < 0, respectively. Samples with SARS-CoV-2 immune response (as determined by CoVi-plex, combined IgG + IgA classification) were significantly overrepresented within the group of α-hCoV high responders (p = 3.78e-03, Fisher's exact test, two sided), while being significantly underrepresented within the group of α-hCoV and β-hCoV low responders (p = 1.14e-03 and p = 1.56e-02, respectively, Fisher's exact test, two sided) (Fig. 4f). These results showed that while there were no discernible global effects for single antigens, there is a correlation between the SARS-CoV-2 immune response with high hCoV responses, especially towards α-hCoVs.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint

Discussion
We demonstrated that CoVi-plex, a novel multiplex assay, is highly suitable to classify seroconversion in SARS-CoV-2-infected individuals. With a combined cut-off using SARS-CoV-2 trimeric full-length spike protein and RBD, we were able to eliminate false positive responses and achieved a sensitivity of 90% with a specificity of 100% for 310 samples from SARS-CoV-2-infected and for 866 samples from uninfected individuals. We found that detection of IgG more accurately reflected infection compared to IgA, although both were highly specific. However, by simultaneously monitoring IgA, we additionally were able to detect an early immune response in some patients. The CoVi-plex approach allows the easy addition of SARS-CoV-2-specific antigens, here six in total, which provides an additional level of confidence in patient classification. Thus, for example, we noticed that the spike S1 domain showed fewer false positive responses compared to the S2 domain. Interestingly, Ng et al. 12 reported reactivity towards SARS-CoV-2 S2 from sera of patients with recent seasonal hCoV infection. These sera prevented infection with SARS-CoV-2 pseudotypes in a neutralization assay. Additionally, we found that spike non-responders also did not show a response to nucleocapsid, which has been described as strongest inducer of antibody responses 11,13 ; and not vice versa.
In our comparison to commercially available IVD tests, we classified fewer samples as false negative using our CoVi-plex approach. For 10% of all infected samples, we could not detect a SARS-CoV-2 specific immune response, which is in line with previous findings 3,14,15 . Those non-responders may be able to limit viral replication by innate immune mechanisms or cellular immunity is dominant in mediating viral clearance 16,17 .
Expanding our CoVi-plex approach to the endemic hCoVs NL63, 229E, OC43, and HKU1 revealed a clear IgG immune response for all tested samples. Furthermore, we did not observe a difference for the samples from proven hCoV-infected individuals, compared to other samples. Due to the general lack of availability of samples from hCoV-naïve individuals, it was difficult to analyze hCoV-mediated cross-reactivity. Nevertheless, our multiplexed readout indicates a correlation between the SARS-CoV-2 immune response and high hCoV responses.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint Currently, we are identifying population groups which were highly exposed and showed different susceptibility to SARS-CoV-2 infection, e.g. the "Ischgl-study group" (unpublished data) 18 , in order to elucidate potential cross protection derived from immune responses towards endemic hCoVs in more detail. Alternatively, CoVi-plex studies analyzing hCoV signatures in samples from individuals before and after SARS-CoV-2 infection would help to get insight into a potential cross protection.
A multiplex setup such as CoVi-plex is especially suited to vaccination studies, since the flexibility and broad antigen coverage allows to efficiently map vaccine immune responses to an immunoglobulin isotype and subtype level for the target pathogen and related species 19 .
Interestingly, previous SARS-CoV-1 vaccine studies clearly indicated that a detailed characterization of vaccine-induced antibody responses is mandatory for efficient coronavirus vaccine development 20,21 .
In summary, we have established and validated CoVi-plex, a robust, high-content-enabled, and antigen-saving multiplex assay. This assay is suitable for comprehensive characterization of SARS-CoV-2 infection on the humoral immune response and for epidemiological screenings to accurately measure SARS-CoV-2 seroprevalence in large cohort studies. It further provides the unique opportunity to assess and correlate immunity for both endemic and pathogenic coronaviruses. Finally, the multiplex nature of CoVi-plex can deliver urgently needed data on the outcome of SARS-CoV-2 vaccination.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 17, 2020. . . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 17, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Generation of expression constructs for production of viral antigens
The cDNAs encoding the full-length nucleocapsid proteins of SARS-CoV-2, hCoV-OC43, hCoV-NL63, hCoV-229E, and hCoV-HKU1 (GenBank accession numbers QHD43423. The cDNA encoding the S1 domain (aa 1 -681) of the SARS-CoV-2 Spike protein was obtained by PCR amplification using the forward primer S1_CoV2-for 5´-CTT CTG GCG TGT GAC CGG -3´ and reverse primer S1_CoV2-rev 5´ -GTT GCG GCC GCT TAG TGG TGG TGG TGG TGG TGG GGG CTG TTT GTC TGT GTC TG -3´ and the full length SARS-CoV-2 SPIKE cDNA as template and cloned into the XbaI/NotI-digested backbone of the pCAGGS vector, thereby adding a C-terminal His6-Tag.

Protein expression and purification
All purified proteins were analyzed via standard SDS-PAGE followed by staining with InstantBlue Coomassie stain (Expedeon) and immunoblotting using an anti-His antibody (Penta-His Antibody, #34660, Qiagen) in combination with a donkey-anti-mouse antibody labeled with AlexaFluor647 (Invitrogen) on a Typhoon Trio (GE-Healthcare, Freiburg, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
To further confirm correct expression, integrity, and purity, proteins were analysed by mass spectrometry. To control the production reproducibility of the antigens, potential aggregation and melting temperatures of the proteins were investigated by nano differential scanning fluorimetry (nanoDSF) using a Prometheus (Nanotemper, Munich, Germany).

Commercial antigens
Two commercial antigens were used to complement the in-house-produced antigen panel.
Bead stocks were vortexed thoroughly and sonicated for 15 seconds. Subsequently, 83 µL of 0.065% (v/v) Triton X-100 and 1 mL of bead stock containing 12.5 x 10 7 beads of one single bead population were pipetted into each well. The beads were then washed twice with 500 µL of activation buffer (100 mM Na2HPO4, pH 6.2, 0.005% (v/v) Triton X-100) and beads were activated for 20 min in 300 µL of activation mix containing 5 mg/mL EDC and 5 mg/mL sNHS in activation buffer. Following activation, the beads were washed twice with 500 µL of coupling buffer (500 mM MES, pH 5.0, 0.005% (v/v) Triton X-100) and the antigens were added to the activated beads and incubated for 2 h at 21 °C to immobilize the antigens on the surface.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Data analysis
Data analysis and visualization was performed with R Studio (Version 1.2.5001, using R version 3.6.1) using the Median Fluorescent Intensity (MFI). Statistical analysis was performed using R package "stats" from the base repository. Mann-Whitney U test was used to determine difference between signal distributions from different sample groups. Hierarchical clustering was performed to group antigens by response from the entire sample set. Fishers' exact test was used to calculate significance of overlap between sample groups.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Quality control
In order to test the repeatability of CoVi-Plex three quality control samples (QCs) were processed in duplicate on each test plate (n = 17) during the sample screening and inter-assay variance was assessed for each antigen in the multiplex. For intra-assay variance, 24 replicates for each of the three QC samples were analyzed on one plate. Achieved results are presented in Extended Data Table 1   is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020.

Competing interests
The authors declare no competing interests.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint Extended Data Fig.4 a, Correlation of IgA response for the entire sample set (n=1176) is visualized as heatmap based on Spearman's ρ coefficient; dendrogram on the right side displays antigens after hierarchical clustering was performed. b, Immune response (IgG and IgA) towards hCoV N-NTD proteins are presented as Box-Whisker plots of sample MFI on a logarithmic scale for SARS-CoV-2-infected (red, n=310) and uninfected (blue, n=866) individuals. Outliers determined by 1.5 times IQR of log-transformed data are depicted as circles. c, Relative levels of IgG-specific immune response towards hCoV N-NTD proteins are presented as Box-Whisker plots / stripchart overlays of log-transformed and per-antigen scaled and centred MFI for the sample subsets of Spike Trimer false positives (blue, n=17) and combined IgG + IgA false negatives (red, n=31). Table 1 Intra-and inter-assay variance were determined by repeated measurement of QC samples and blank sample as replicates on one plate and in duplicates over 17 plates, respectively.

Extended Data
Standard deviation relative to mean (%CV) is given for each antigen. A limit of detection (LOD) was calculated from 24 blank sample replicates on the same plate as the mean MFI + 3 times standard deviation. Table 2 Complete overview of study sample set. Samples are divided into columns by age groups and . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . IgA (MFI) HKU1 S1 Color Key . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint Extended Data - Table 1 Spike Trimer RBD S1 S2 N N-NTD S1 N N-NTD S1 N N-NTD S1 N N-NTD S1 N N-NTD  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 17, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 17, 2020. . https://doi.org/10.1101/2020.07.17.20156000 doi: medRxiv preprint