Choice of antigens and assay optimization
Most serological diagnostic tests for SARS-CoV-2 use S and N protein as the main antigens due to their highly immunogenic characteristic [11, 15]. In addition, RBD has been reported as the best predictor of SARS-CoV-2 infection [16, 17, 18]. For the assay described here, we used three different antigens: the trimeric S, produced in HEK293-3F6 cells (kindly provided by Dr. Leda Castilho, COPPE, UFRJ) ; the RBD (residues 319 to 541 of S protein), produced in HEK293T cells according to ; and the N-terminal domain of N protein (residues 44 to 180 of N protein; N-NTD), produced in E. coli as previously established by our group  (Fig. S1). We evaluated separately the reactivity of 3 immunoglobulins (IgG, IgA and IgM) to each of the antigens (S, RBD, N-NTD), so that the test provides 9 different reactivity results for each serum.
For assay development and optimization, we first determined the best conditions of plate coating with the viral antigens, as well as appropriate dilutions of sera and secondary antibodies. For these tests, we used serum samples from 23 individuals with positive PCR result for SARS-CoV-2, as well as 3 sera collected before SARS-CoV-2 pandemic (Table 1). Since PCR + samples reacted differently with each of the antigens, the 4 most reactive samples for a given antigen were chosen for the optimization of the respective test.
Antigen coating density
For an in-house ELISA, determining the ideal antigen concentration to coat the plates is particularly important when the antigens are also produced in-house and thus require production and cost optimisation. To define the best antigen coating density, we performed two-fold serial dilution of each antigen (S, RBD and N-NTD), ranging from 0.25 to 16 µg/ml, with 50 µl of antigen solution applied per well (12.5 to 800 ng antigen per well). For each antigen, tests were conducted using the 4 most appropriate PCR + and 3 pre-pandemic samples assayed for either IgG (Fig. 1A-C), IgA or IgM detection (Fig. S2). We observed no reactivity of pre-pandemic sera at any condition tested, while most PCR + samples showed a dose-dependent antigen recognition profile. Based on these results, we chose a concentration of 4 µg/ml for all antigens to perform the subsequent assays.
To evaluate the actual contribution of RBD to sera reactivity to S, we performed an assay coating the plates with S or RBD with the concentrations adjusted to maintain the same amount of RBD epitopes. Since S and RBD molecular masses are approximately 180 kDa and 26 kDa, respectively, we compared sera reactivity using 4 µg/ml S with 0.6 µg/ml RBD (RBD molar equivalent in the complete S) or 4 µg/ml RBD (maintaining the same protein mass) (Fig. S3). We observed a high reactivity to RBD even when used at a lower concentration. This result may be associated with a large amount of sample antibodies directed to the RBD, suggesting that the domain has a major contribution in the reactivity against the entire S. No significant changes were observed in the results of PCR + samples for IgG or IgA when 4 or 0.6 µg/ml antigen concentration were used, suggesting that using 0.6 µg/ml RBD is enough to ensure reliable results about IgG. However, for the IgA and IgM detection assay, a better separation of positive samples was seen when 4 µg/ml RBD was used.
To determine the best sera concentration in our assay, we performed two-fold serial dilution of the 4 most appropriate PCR + and 3 pre-pandemic sera samples, with dilutions ranging from 1:25 to 1:800. Assays were performed by using 4 µg/ml of the antigens to coat the plates, and assessing reactivity to either IgG (Fig. 1D-F), IgA or IgM (S4). We found that, especially for IgG reactivity against all the antigens (Fig. 1D-F) and for all antibodies tested against N-NTD (Figs. 1F and S4E-F), dilutions lower than 1:50 result in a substantial increase in the background of some negative samples, which may generate false positive results. For most cases of PCR + samples, the ideal dilution was between 1:50 and 1:100 (to reach values that allow a good separation of positives from negatives samples).
To evaluate quantitatively the assay sensitivity, we broadened the sera dilution range (10-fold serial dilutions) using three different groups of positive samples: (1) 23 PCR + sera collected before vaccination has started (PCR+); (2) 4 sera from individuals vaccinated with two doses of ChAdOx1 nCoV-19 vaccine who were not reactive before vaccination (Vaccinated – previously non-reactive); and (3) 4 sera from individuals vaccinated with two doses of ChAdOx1 nCoV-19 vaccine who were reactive before vaccination (Vaccinated – previously S IgG reactive). As negative controls, we used 10 sera samples collected before pandemic. Sera titration revealed that reactive samples may show very different antibody titers (Fig. 2). Calculations of the area under the curves (AUC) showed that sera from vaccinated individuals who tested positive before vaccination presented significantly higher titers when compared to sera from vaccinated individuals who were non-reactive when receiving the first vaccine dose. It is also clear the high antibody titer variability among the pre-vaccination PCR + sera (Fig. 2A and C). This variability would be explained by the timing, severity of symptoms and longevity of the humoral response against SARS-CoV-2, as those sera were collected in different time points after individuals were infected, or by differences in the response intensity within a population. Additionally, individuals with lower titers before vaccination had significantly greater increases in antibody titers when compared to individuals with high titers before vaccination (Fig. 2C, colored symbols). Altogether, the results show the importance of sample dilution for quantitative analysis, highlighting that dilutions between 1:50 and 1:100 were suitable for differentiating negative and positive samples without background increase and false positives.
Dilution of the detection antibodies
The dilution of the secondary antibody must also be adjusted to avoid high backgrounds in the assay. Here we evaluated the best conditions for detection antibodies for three immunoglobulins, IgG, IgA and IgM, using S as antigen (Fig. 1G-I). Dilutions of less than 1:10,000 of anti-IgG or anti-IgA resulted in high backgrounds, clearly observed for the pre-pandemic samples (Figs. 1G and H, respectively). On the other hand, dilutions above 1:20,000 impaired the detection IgA positive samples. IgM reactivity remained detected throughout the dilution range tested (Fig. 1I). Based on the results, we chose a 1:10,000 dilution for all three detection antibodies.
Incubation periods for each assay step
To optimize the time spent in the complete assay, we set the duration for each step of the protocol: incubation of sample (Fig. 3A-C), secondary antibody (Fig. 3D-F), and chromogenic substrate (Fig. 3G-I). Sera reactivity reached a plateau from 1 hour of incubation under all conditions analyzed. However, differences amongst IgG positive samples were detected between 10 and 30 minutes of reaction. These differences were reduced or lost when the samples were incubated for 2 hours, when samples exceed the linearity limit of the assay (Fig. 3A-C). A similar profile was observed for the incubation with the secondary antibody. For IgG and IgA, it is possible to differentiate positive and negative samples earlier than for IgM, for which 1 hour incubation was the most adequate for better separation between the negative and positive groups (Fig. 3D-F). Regarding the incubation time with the chromogenic substrate TMB, more pronounced differences were observed depending on the class of antibodies to be detected. For IgG, 4 minutes of reaction was enough to separate positive from negative samples, while for IgA and IgM, 8 minutes incubation was needed to obtain a proper separation of negative and positive groups (Fig. 3G-I).
We also set the conditions for plate blocking. Plate incubation for different time periods with 3% BSA solution revealed that a 10-minutes incubation was enough to completely reduce the background of the assay (Fig S5A). A comparison between blocking with either 3% BSA or 3% milk powder solution showed no significant difference (Fig S5B).
When sera samples potentially containing infectious agents are handled, heat inactivation at 56°C for 30 minutes is recommended to reduce the risk. Thus, to ensure assay reproducibility, it is important to evaluate temperature stability of the serum antibodies. In addition, sample collection can often be frozen and thawed for many tests. To assess serum antibodies stability to temperature variations, we performed the assay using non-inactivated and inactivated samples (Fig. S6A), as well as comparing the results obtained using fresh samples or samples subjected to 10 freeze and thaw cycles (Fig S6B). We observed no significant differences between the reactivity of samples tested, suggesting that the antibodies present in the serum are stable under these conditions.
Effects of pH variations
The pH of the solutions used in the assay is also an important factor to ensure reproducibility. To assess the effect of pH variation, we quantified the reactivity of IgG against S protein at pH 6.6, 7.4 and 8.0 (Fig. S6C). We found that at mild acidic pH (pH 6.6), 11 of 12 positive sample tested negative, with a drastic reduction in their absorbance. At weak basic pH (pH 8.0), only 1 of 12 positive sample tested negative, although the remaining 11 showed an important reduction in their absorbance. Therefore, it is crucial to carefully adjust the antigen dilution buffer to the optimal pH for the assays (pH 7.4).
Evaluation of the performance of the assays by determination of the cut-off
Determination of the cut-off values is an important step to be taken when the assay is used for diagnosis purposes. We first determined the sensitivity (true positive rate) and specificity (true negative rate) of the assays using as negative group, 42 sera samples collected before 2019 (pre-pandemic sera) and, as positive group, 23 sera collected from individuals who tested positive for SARS-CoV-2 infection between June 2020 and April 2021 (PCR + sera). Two different approaches were used to establish the cut-off value (Figs. 4, S7). In the first approach, it was used the three standard deviations from the mean, which is a common cut-off in practice for identifying outliers in a Gaussian or Gaussian-like distribution of ELISA results based on OD ranges  In this heuristic approach, the cut-off was calculated as the mean + 3 SD of the absorbance values of the 42 sera pre-pandemic samples. Thus, PCR + samples with an OD higher than the mean of the negative controls + 3 SD were considered true positive and inversely, pre-pandemic samples with an OD higher than the mean of the negative controls + 3 SD were considered true false positive. Using this method, we identified 2 false positive samples for IgG reactivity to S and for IgA reactivity to N-NTD, corresponding to a specificity of 95.2% for the reactivity of both immunoglobulins. For IgM reactivity to the 3 antigens and for IgA reactivity to RBD, we identified 1 false positive, corresponding to a specificity of 97.6%. For IgA reactivity to S and IgG reactivity to N and RBD, no false positives were found, corresponding to a specificity of 100%.
In the second approach we performed a ROC analysis to determine the assay accuracy. In order to determine the performance of the assays, the ROC analysis was performed using OD data of all ELISA assays (Fig. 4, Table 2, and suppl. Fig S7). Figure 4B, C, D shows the ROC analysis of the optimized in-house ELISA for IgG reactivity to S, RBD, or N-NTD. The area under the curve (AUC) graphs (above) were generated to graphically represent the performance of the assays and the distribution graphs (below) represent the dispersion of the PCR-positive (solid lines) and pre-pandemic (dotted lines) serum samples relatively to the optimal cut-off for each ELISA (vertical dotted lines). The area under the curve was high for all three assays, with highest value for the S-IgG ELISA (range 1–1). The Youden method that integrates both sensitivity and specificity of the test allows selecting the best cut-off values that were estimated in a range of 0.583 to 1.073 units of absorbance (Table 2). These calculated cut-off values determine the classification of the samples (positive and negative predicted values and likelihood ratios) for subsequent analyses (Table 2). If the cut-off values calculated based on two different approaches were very similar for IgG reactivity to S and RBD, they are very different regarding N-NTD (respectively, 0.53 and 1.51 units of absorbance for ROC and mean + 3 SD analyses). This apparent discrepancy is mainly due to the high dispersion of the pre-pandemic results for IgG reactivity to N-NTD (two small peaks in the distribution plot Fig. 4D), which very likely results from false positive samples (non-specific reactivity and/or cross-reactivity to N-NTD). In addition, the low sensitivity results obtained for N-NTD, especially in the three sigma approach, may be explained by the absence of antibodies against this protein in the PCR + samples used in the assay. Similar results of high sensitivity and specificity were obtained for IgA reactivity to S and RBD, while poor specificity and sensitivity was observed against N-NTD (Fig. S7). Finally, IgM reactivity to S was found to be very sensitive and specific, in contrast to reactivity to RBD, which was neither specific nor sensitive (Fig S7).
Validation of the in-house ELISA assays
In order to validate our in-house ELISA, we applied the complete assay to 3 different groups of samples: (1) sera collected from 23 individuals who tested positive in a PCR test for diagnosis of SARS-CoV-2 infection (PCR + samples); (2) 206 sera samples collected from 103 individuals (2 samples from each individual, collected in October-November 2020 and February-April 2021) who were not diagnosed for SARS-CoV-2 infection and before receiving any type of vaccine for SARS-CoV-2 (Non-diagnosed individuals); and (3) 26 sera samples collected from 13 individuals vaccinated against SARS-CoV-2 (Vaccinated individuals), being 2 samples from each individual, the first before vaccination and the second 20 days after the first dose of ChAdOx-1 nCoV-19 (7 individuals) or CoronaVac (6 individuals) vaccines (Table 1). Among these 13 individuals, 8 were not reactive for SARS-CoV-2 antibodies and 5 tested positive in a PCR test for diagnosis of SARS-CoV-2 infection before vaccination. All results obtained were grouped as a heatmap graph (Fig. 5).
PCR + samples
Applying the test to the PCR + samples, we found 100% IgG reactivity to S, 78% to RBD and 52% to N-NTD (Fig. 5A). The detection of IgG against the 3 antigens was observed in 47% of the samples. The 4 sera that were negative for IgG against RBD had the lowest IgG titer against S, highlighting the importance of RBD epitopes to the serological reactivity to S. IgA and IgM reactivity to S were found in 80 and 60% of the samples, respectively. Among the samples reactive to RBD, 72% were positive for IgA and 44% for IgM, while for N-NTD, 58% of the reactive samples were positive for IgA and only 8% for IgM.
The analysis of the serological response of the first sample collected (November 2020) of all the non-diagnosed individuals showed that 12.6% of them were positive for S IgG, 11.6% for RBD and 11.6% for N-NTD IgG. Only 4.8% of the samples were positive for all antigens (Fig. 5C). Demographic analysis of these individuals revealed that 39% (40 out of 103 individuals) reported some flu-like symptoms before November 2020 (Table 1). From this sub-group, 17% showed IgG reactivity to S and N-NTD and 11% to RBD (Fig. 5C). Among the other 63 asymptomatic individuals, only 8% showed IgG reactivity to S and RBD, while 14% were positive to N-NTD. In addition, it is important to note that 7 individuals (4 from the 63 asymptomatic individuals – 9.3% – and 3 from the 40 symptomatic – 5%) showed serologic response only to N-NTD IgG, which can be interpreted as a non-specific reactivity and/or a cross-reaction to N-NTD of other coronaviruses or even to proteins from other viruses that are endemic in Brazil . Analyses of samples collected from those individuals (non-diagnosed in November 2020; results shown in Fig. 5E) until 5 months later, to April 2021, revealed an important seroconversion, with a 100% increase in S and RBD IgG incidence and a 16% increase in N-NTD IgG detection (Fig. 5D). Differently from the PCR + samples, only 30% of the positive sera for IgG against any of the 3 antigens were also positive for IgA, and only 20% for S IgM and 8% for RDB and N-NTD IgM. Very few samples showed only IgA or IgM reactivity.
As the vaccination progress, seroconversion from SARS-CoV-2 infection overlaps with the serological response to vaccines. To illustrate the potential differences that can be detected by assessing serological responses at a time when a substantial part of the population is vaccinated, we performed our in-house using serum samples collected from individuals who tested positive for SARS-CoV-2 infection or from non-reactive individuals before and after receiving the first dose of CoronaVac or ChAdOx-1 nCoV-19 vaccines (Fig. 5B). CoronaVac is an inactivated virus vaccine, so that the individual is immunized by the contact with all the viral proteins . In contrast, ChAdOx-1 nCoV-19 vaccine consists of an adenoviral vector containing the full-length SARS-CoV-2 S protein . As expected, all sera from vaccinated individuals were positive for S IgG, but, for these analyzed samples, IgG reactivity to RBD was more evident in PCR + individuals after immunization for both vaccines. Despite containing N protein, CoronaVac did not induce a significant N-NTD IgG response in the tested samples. ChAdOx-1 nCoV-19, but not CoronaVac, increased the S IgA titer in the subjects previously diagnosed with SARS-CoV-2 infection and induced this response in individuals who were previously not-reactive. Only 1 PCR + individual showed IgA reactivity to RDB after receiving ChAdOx-1 nCoV-19. No induction of IgM response was seen for any of the three antigens after immunization with any of the vaccines.