S-protein’s RBD has been shown to be SARS-CoV-1’s Achilles’ heel [8]. Due to the homology between the S-proteins as well as the RBDs of SARS-CoV-1 and SARS-CoV-2 (Fig. 1a), this may also hold true for the virus driving the current pandemic. This is why we decided to establish an ELISA protocol, which uses the SARS-CoV-2 RBD and stabilized ectodomain (StabS [9]) as antigens (Fig. 1b). To promote robust production of the two recombinant proteins, coding sequences were codon-optimized and the autologous S-protein signal peptide was replaced by a minimal version of the tissue plasminogen activator (tPA) signal peptide developed in our lab. RBD yielded 70 mg/l supernatant for the RBD and 14 mg/l supernatant for the StabS protein. Purity and homogeneity were verified by reducing SDS-PAGE and size exclusion chromatography (Fig. 1c, d).
A number of SARS-CoV-2 serology ELISA protocols have been published recently [10,11,12,13,14,15]. In our protocol, the antigens were directly absorbed to the plate’s plastic surface. To control protein integrity, we used our ELISA protocol in combination with the structure-dependent monoclonal anti-RBD antibody CR3022 [16,17,18]. High affinity binding as reflected by a KD of 0.27 nM for the RBD protein and 0.48 nM for StabS as well as a strong absorption (OD450–630 > 1 at saturation after 4 min of development time) in both cases demonstrated sufficient amounts of well-folded protein (Fig. 1e). We used CR3022 at saturation concentration to test the stability of RBD-coated plates, which had been coated up to 8 days earlier. No performance loss was detected during this period when plates were stored at room temperature in PBS-T (Fig. S2).
We used our ELISA to screen for anti-S and anti-RBD antibodies in sera from COVID-19 patients and controls. In a first step, we quantified anti-RBD responses and anti-S responses in 22 sera, which displayed differential ELISA signals (Fig. S3).
We found that antibody responses against StabS correlated very well with antibody response to RBD as determined by correlation of OD450–630 at 1:100 dilution, area under the curve (AUC), effective concentration at 50% signal (EC50) and titer (Fig. S4). Moreover, comparable signal strengths in these analyses suggest that RBD is a key immunogenic determinant of anti-S responses (Fig. S3). Therefore, anti-RBD antibody levels represent a valuable surrogate for testing anti-S-directed antibody responses.
High-throughput titration of sera for the purpose of determining end point titers or calculating EC50 values—in particular in seroepidemiological surveillance studies on broader population scale—is a time- and material-consuming procedure. Thus, we tested whether OD450–630 at 1:100 dilution of sera correlate with standard serum characteristics such as area under the curve (AUC), effective concentration at 50% signal (EC50) and titer. This analysis revealed that 1:100 dilution of sera reflects total serum antibody responses (Fig. S5). Therefore, we used single-point measurements for all further assays.
Next, we wanted to establish cutoff values for anti-RBD-directed IgG, IgA and IgM antibody responses following the requirements for validation of diagnostic assays in clinical virology [19]. For that purpose, we measured 190 SARS-CoV-2 naïve sera that had been collected before the current SARS-CoV-2 pandemic (Fig. S6). As recently proposed by Okba et al. [11], we used the mean of the background signals plus six standard deviations (SD) to define the cutoff value. Using these parameters, we determined the specificity of our assay by measuring 1000 independent SARS-CoV-2 infection naïve sera. We obtained a false positive rate of 7 out of 1000 sera, corresponding to a specificity of 99.3% (Fig. 2).
To further verify the specificity of our ELISA, we tested whether sera from patients, that suffered from infection with seasonal CoVs, cross-react with our assay. For that purpose, we retrieved in a total of 43 sera from patients that had PCR-proven seasonal corona virus infection from our diagnostic repository. In a commercial line blot assay, 34 sera out of 43 scored positive for seasonal CoV-specific antibodies (Table S2). We used these 34 sera to further validate our RBD ELISA. Of utmost importance, none of these seasonal coronavirus sera showed any cross-reactivity in our ELISA (Fig. 2).
To verify precision and reproducibility of our ELISA [19], we used a minimal panel of seasonal (N = 5), naïve (N = 15), weakly IgG reactive (N = 10) and strongly IgG reactive (N = 10) sera. For weakly IgG reactive sera, the relative standard deviation (σrel) was 3.98% for IgG, 1.72% for IgM and 7.15% for IgA; for strongly reactive IgG positive sera σrel was determined to be 0.16% for IgG, 0.22% for IgM, and 2.73% for IgA. Combined inter-assay and inter-operator variabilities for weakly IgG reactive sera was determined as a σrel of 10.26% and for strong IgG reactive sera σrel was calculated to be 4.12%.
To determine the sensitivity of our assay, we analyzed sera from patients suffering from COVID-19 (N = 144), assuming that after symptomatic SARS-CoV-2 infection eventually all subjects develop antibodies. We quantified the anti-RBD IgG responses, which correlated well (R2 = 0.8812, Spearmen’s ρ = 0.917, p value < 0.0001) with anti-SARS-CoV-2 responses measured using a commercial IgG ELISA (EUROIMMUN, Fig. S7) that has been validated recently [20]. Next, we determined the IgM, IgG and IgA levels at different time frames after the first detection of SARS-CoV-2 RNA by RT-qPCR in our cohort (proven infection; Fig. 3a–c). At > 10 days after proven infection, these assays displayed sensitivities of 92% for IgA, 96% for IgG and 98% for IgM. However, we found that the antibody responses were already remarkably elevated at early time points after proven infection and, in concordance with previous findings, did not display a steady increase over time [21]. We speculated that infection preceded virus testing by several days. For 41% of the subjects (N = 59), the time point of symptom onset was available and we could calculate the average period between symptom onset and detection of SARS-CoV-2 RNA by RT-qPCR (5.7 days; Fig. S8). This suggests that the majority of patients who were subjected to PCR testing had already suffered from COVID-19 for 5.7 days. Therefore, an analysis of anti-RBD responses in relation to days after symptom onset should result in lower anti-RBD responses at early time points and a steady increase of antibody levels over time. Indeed, in the subgroup of patients where the time point of symptom onset was available, we detected very low antibody responses early after symptom onset and a steady increase in all anti-RBD antibody isotypes (Fig. 3d–f).
Neutralizing antibodies correlate with protection against several pathogens. Since we used the RBD of SARS-CoV-2, we expected the serum reactivity measured in our ELISA to correlate with virus neutralization due to sterical hindrance of RBD’s binding to its receptor ACE2 (Fig. 1b). In a first step, we isolated ten SARS-CoV-2 strains from respiratory specimen of COVID-19 patients. Of three different cell lines, kidney epithelial cells (Vero) supported virus propagation more efficiently than hepatocarcinoma-derived Huh-7 cells and lung carcinoma-derived A549 cells (Fig. 4a). For this reason, Vero cells were selected for neutralization experiments, using a highly replicative and cytopathic SARS-CoV-2 isolate (strain CA).
Prior to infection of Vero cells, the virus was incubated with serial dilutions of the 22 sera, which displayed differential ELISA titrations (Fig. S3). After removal of the inoculum at 12–24 h post-infection, viral loads were determined in cell culture supernatants at 48 h post-infection using SARS-CoV-2 RT-qPCR. All sera which were negative in the ELISA (N = 6) did not inhibit virus entry and replication (Fig. 4b). In contrast, sera with borderline or positive ELISA values reduced SARS-CoV-2 viral loads by two (N = 2) to six log10 (N = 14). Notably, the reduction of viral loads showed a varying pattern, from a sudden to a more gradual inhibition of viral replication. Neutralization capacity was more pronounced in sera with high ELISA values. Most importantly, neutralization titers (given as IC50 values) correlated strongly with anti-RBD (R2 = 0.8943, Spearmen’s ρ = 0.965, p value < 0.0001) and anti-StabS (R2 = 0.9057, Spearmen’s ρ = 0.964, p value < 0.0001) antibody levels (Fig. 5a, b and Fig. S9a–f).