The study was approved by the ethics committee of the Rostock University Medical Center under file number A 2020-0086. All donors filed their written informed consent for participation. Apart from sex, age, sampling dates, SARS-CoV-2-related PCR, and serological data, no other clinical parameters or comorbidities were recorded. COVID-19 patient and pre-COVID cohorts were recruited between September 2020 and January 2021, before the advent of routine vaccination for the general population (15). In detail, pre-COVIDs were recruited from the local test center that they visited for the exclusion of COVID-19. Inclusion in our study required a negative PCR result for SARS-CoV-2. Patients with COVID-19 were recruited from the Division of Tropical Medicine and Infectious Diseases at Rostock University Medical Center within the first two weeks of infection. Infection was confirmed by a SARS-CoV-2-specific diagnostic PCR. Blood samples were collected via venipuncture in the presence of EDTA as an anti-coagulant and plasma was harvested after centrifugation. Plasma aliquots were stored at -80°C. Subjects for the vaccine cohorts were enrolled from clinical, scientific, and administrative staff members of the Rostock University Medical Center (AZD1222, BNT162b2, and NVX-CoV2373) and from the local population at municipal vaccination centers (NVX-CoV2373). Vaccinations followed the recommendations of the German Standing Committee on Vaccination (STIKO). The AZD1222 “A” cohort received two doses of AZD1222 at a mean interval of 12 weeks. Boost immunizations with an mRNA-based vaccine (either BNT162b2 or mRNA-1273) were administered at a mean interval of 39 weeks after priming. The BNT162b2 “B” cohort received two doses of BNT162b2 at a mean interval of 29 days. Boost immunizations with either BNT162b2 or mRNA.1273 were administered at a mean interval of 44 weeks after priming. The “N” cohort received two doses of NVX-CoV2373 at the recommended interval of 21 days. Inclusion into any of the vaccine groups at any time point required negative test results for anti-nucleocapsid antibodies. Blood samples were collected between July 2021 and October 2022.

Antibodies of type IgG against SARS-CoV-2 Spike RBD domain or nucleocapsid protein were quantified using commercial anti-SARS-CoV-2S and anti-SARS-CoV-2N Elecsys^® immunoassays (Roche Diagnostics, Mannheim, Germany). The assays were done by the central laboratory of the Institute of Clinical Chemistry and Laboratory Medicine of the Rostock University Medical Center. Antibody reactivity is referred to as the international WHO standard for SARS-CoV-2 immunoglobulins and is expressed as binding antibody units (BAU) (16). A commercial surrogate virus neutralization assay (sVNT, GeneScript) was used to determine the neutralization capacities of plasma samples as described previously (17).

We employed a commercially available peptide microarray slide (JPT Peptide Technologies GmbH, product RT-WCPV-S-V04-2) carrying 21 identical microarrays that include 318 peptides derived from SARS-CoV-2 Wuhan-Hu-1 wildtype Spike protein sequences (P0DTC2; Uniprot release 2021_02). The annotated peptide content of a microarray can be found as part of the data record in Supplementary Table S1 . The peptides are 15mers with 11 amino acid overlaps and are covalently attached to the glass slide via an N-terminal spacer. Because of technical reasons, they carry an additional glycinate residue at their C-terminus. A 96-well frame (JPT Peptide Technologies GmbH) that carries up to four slides separates the subarrays into individual wells. To get replicate measurements, each sample was applied to three subarrays. Subarrays that were only probed with secondary antibodies served as a baseline for later quantitation. Incubations of the microarray slide were done on a plate shaker (Thermo Fisher) at 30°C. The staining workflow commenced with the blocking step using SuperBlock/TBS (Thermo Fisher) with 0.5% Tween 20 for 1 hour and two rinses with washing buffer (TBS, 50mM TRIS/HCl pH 7.4, 150mM NaCl plus 0.1% Tween 20). Next, the subarrays were probed with plasma pools of different sample groups diluted in the blocking buffer for 2 hours. Each pool was made up of 10 plasma samples in which each sample was diluted 1:100. Subarrays were rinsed five times with washing buffer and then incubated for 1 hour with a secondary antibody Fab preparation of goat anti-human IgG conjugated with Alexa Fluor 647 (Thermo Fisher Z25408), diluted to 0.8 µg/ml in blocking buffer. After five rinses in washing buffer and two rinses in distilled water, the microarray slide was dried by brief centrifugation and then scanned in a fluorescence scanner (Agilent G2505C) at 5µm per pixel resolution and with 16-bit sample depth. Image analysis, including spot detection, outlier correction, and filtering for reactive peptides over background, was done as described in detail in a previous study (18). Briefly, GenePix Pro version 6.1 (Molecular Devices) was used to determine the median foreground fluorescence of each spot. Then, the fluorescence intensity of the three replicates was averaged. The list of peptides with specific signals over background was derived by using the MAID approach with a signal threshold of 400 that considers the criteria signal intensity as well as fold change to the buffer control. The selection of peptides for further study was done manually using the following criteria: We first built groups of all peptides with signals over background according to their reactivity with plasma samples from the cohorts. These included groups “vaccinees and patients”, “vaccinees NOT patients”, “AZD1222 only” and “BNT162b2 only”, “patient only”, “pre-pandemic only” and “pre-pandemic & patient”. We then removed all peptides present only in Spike protein sequences from SARS-CoV-2 variants. Within each of the peptide reactivity groups, we chose the peptides with the highest reactivity. This could be the maximum for single cohort groups or the sum of signals when more than one cohort contributed to a selection group. When peptides were overlapping, we selected the one with the highest signal.

Peptides of 15 amino acid length with N-terminal biotin, spacer, and an additional glycinate at their C-terminus were obtained from a commercial source (JPT Peptide Technologies GmbH, BioTides™). Prior to first use, peptides were dissolved at 0.2mM in dimethylsulfoxide and stored at -80°C. Further dilutions were done in TBS (50mM TRIS/HCl, 150mM NaCl). Polystyrene 96-well plates (NUNC Maxisorp, Thermo Fisher) were coated with 100µl of a solution of 5µg/ml Neutravidin (Thermo Fisher) in 0.1M carbonate buffer pH 9.6 at 4°C for 18 hours. After rinsing twice with TBS, wells were loaded with 50 pmoles per well biotinylated peptide in 100µl TBS for 1 hour at 37°C. Wells were rinsed twice with PBS and then blocked with 300µl Superblock/TBS (Thermo Fisher) at 37°C for 1 hour, followed by another two rinses with TBS. The plasma samples were diluted 1:200 in antibody dilution buffer (Superblock/TBS plus 0.5% Tween20 and 10µg/ml Neutravidin, the latter to pre-adsorb potential Neutravidin-reacting immunoglobulins) and incubated in 100µl volumes per well at room temperature for 2 hours. Each well was rinsed once, washed twice (5 min incubations), and rinsed again with washing buffer (TBS, 0.1% bovine serum albumin, 0.5% Tween 20). Bound primary antibodies from the plasma samples were probed with goat anti-human IgG antibodies conjugated to peroxidase (Jackson Dianova 109-035-088) at room temperature for 1 hour. Following the washing steps as described after the plasma sample incubation, the substrate reaction with TMB (BioLegend #421101) was initiated until the OD at 620nm (blue) of a chosen peptide/plasma sample used on each ELISA plate reached 0.5, usually approximately 20 minutes. Then the reaction was stopped with 2N sulfuric acid and the OD was read at 450nm (yellow). Plasma-specific peptide reactivity was calculated by subtracting the OD 450nm of a well without peptide from the raw OD 450nm of the well with peptide probed with the same plasma sample. If this difference was negative, the value was set to zero. Preliminary testing in independent triplicate experiments and using only a selection of nine peptides and 19 plasma samples resulted in overall robust ELISA data with high Pearson correlation (median values 0.94; minimum 0.73). For this reason, we decided to determine the whole dataset by single measurements.

We used packages of the R software environment (www.r-project.org) as well as the methods and tools implemented in JASP v 0.16.4 (Sept 29, 2022; https://jasp-stats.org) to draw data illustrations and to statistically evaluate them as indicated in the figure legends and in the text. Initial graphs were imported into Corel Draw X8 for post-processing to get the final figure layout and to add labeling. The data of multiple groups were compared with the Kruskal-Wallis test and, if a significance level of p < 0.05 was reached, the data was subjected to Dunn’s post-hoc analysis with Benjamini-Hochberg correction for multiple testing. The longitudinal series were evaluated using the Friedman test followed by a Conover post-hoc analysis when initial p-values were smaller than 0.05. Correlations of anti-peptide signal intensities with values of the commercial anti-RBD ELISA as well as the neutralization assay are based on Spearman’s rank correlation coefficient (rho and p-value; https://www.wessa.net/rwasp_spearman.wasp). To compile known B-cell epitopes, we searched the Immune Epitope Database and Analysis Resource (www.iedb.org (19); accessed on 2023-01-23) for hits within Spike protein of SARS-CoV-2 Wuhan-Hu-1 (Uniprot P0DTC2). In particular, we focused on sequence S-657 using the search options “linear peptide”, “substring” “B cell assay: any” and “Host: human”. In addition, we screened publications with the subject “B-cell epitope mapping” of SARS-CoV-2 Spike protein and extracted the reported immunodominant hits. The display of 3D protein structures was done with the help of PyMol 2.5.0 (Schrödinger, LLC).

In the first approach, we compared linear B-cell epitopes recognized by IgG-type antibodies in the plasma of SARS-CoV2 vaccinees who received either vector-based AZD1222 (n=20) or mRNA-based BNT162b2 (n=20). The samples from the vaccinated cohorts were compared to a control group of 10 pre-COVID plasma samples and 10 samples from COVID-19 patients with active disease. The groups were matched for age and sex ( Table 1 ). To obtain an overview of the various reactivity profiles, we used pools of plasma representing the sample groups and stained a commercially available peptide microarray carrying overlapping 15mer peptides that cover the whole Spike protein sequence from wildtype Wuhan-Hu-1 ( Supplementary Figure S1 ). The quantitative evaluation of the peptide microarray images revealed that 115 out of 318 wildtype peptides were specifically recognized by at least one of our plasma pools. The summary of all peptides and heatmaps illustrating the reactivity landscape are documented in Supplementary Table S1 and Supplementary Figure S2 , respectively.

We filtered the data for hits in SARS-CoV-2 Spike wildtype peptides that showed either high reactivity or potentially interesting patterns. Among the latter, we considered reacting with plasma from patients and/or vaccinees (see Methods for details). The curation of all signals resulted in the selection of 28 peptides for further analyses ( Figure 1 ). Note that peptide S-813 covering part of the “fusion peptide” around the S2’ sub-cleavage site did not pass our background selection filter. However, since we found very high reactivity among all vaccinees and patients cohorts and because this region is important for virus entry into the host cell (21), we decided to include this peptide in our further study. Of note, the 28 selected peptides are distributed throughout the Spike protein sequence, including its receptor binding domain (RBD). Nevertheless, Figure 1 shows a more dense coverage of the Spike S1 SD1 and SD2 domains as well as of CD and HR2 domains within the S2 part.

The overall signal patterns of the 28 selected as well as all reactive peptides ( Figure 1 ; Supplementary Figure S2B ) showed many overlaps between the sample cohorts. For example, six of the selected peptides (S-553, S-657, S-813, S-937, S-1145, and S-1177) displayed very high antibody signals for both vaccinees and patients. Others showed a more focused distribution with prevalent reactivity only for one cohort, e.g.S-213 for AZD1222, S-409 for BNT162b2, or S-789 for the patient samples. A formal analysis of the cohorts with respect to all reactive peptides corroborated overlapping as well as cohort-specific reactivity profiles (see Supplementary Figure S3 ; Venn diagram). Since peptide-specific signals were derived from pools of 10 plasma samples each, which in turn may contain polyclonal and polyspecific antibodies, the recorded signals reflect the summation of the reactivity of many antibodies.

We next set out to resolve the sample pools. To that end, we used ELISA and determined the reactivity of each individual plasma against the 28 selected peptides. For this series of experiments, we included a total of 71 samples: 20 BNT162b2 vaccinees, 20 AZD1222 vaccinees, 10 COVID-19 patients with active disease, 10 pre-COVID samples, and an additional 11 samples representing NVX-CoV2373 vaccinees ( Supplementary Table S2 ). Against expectation, one of the pre-COVID plasma samples, pre-COV5, displayed very high reactivities towards 11 peptides and was therefore defined as an outlier and excluded from further analyses ( Supplementary Table S2 ).

Clustering of the ELISA results showed high signals beyond OD 2.0 (orange and red tiles in the heatmap) in five of the 10 samples from patients with active disease, whereas in the vaccine and pre-COV cohorts, only one sample each from groups AZD1222 and BNT162b2 reached this level ( Figure 2 ). In particular, sample COV5C stood out with the highest reactivities for seven out of the 28 peptides (S-537, S-553, S-573, S-657, S-685, S-1145, and S-1161) In total, we found eight peptides with significant differences between groups (Kruskal-Wallis p < 0.05 and Dunn’s post hoc values p < 0.05) with prominent reactivities against S-553, S-813, and S-1145 (median signals > 0.5 OD; maximum value >1 OD) among patients ( Figure 3 ). These peptides derive from the subdomain SD1 region of Spike S1 (S-553), the fusion peptide (S-813), and the connector domain CD of Spike S2 (S-1145), respectively. The statistical post-hoc analyses confirmed robust differences between patients and NVX-CoV2373 vaccinees (S-553), between patients and all vaccination groups (S-813), and between patients and nucleic acid-based vaccine cohorts (S-1145) ( Figure 3 ).

The heatmap in Figure 2 illustrates high variability and sample-specific reactivity patterns rather than cohort-specific ones. When focusing on the vaccine cohorts, there was only one small vaccine-specific cluster consisting of three NVX-CoV2373 samples (N8, N4, and N10). Other than that, most peptides reacted selectively with a small subset of plasma samples. With respect to signal intensities, nine peptides displayed signals above 1 OD, while eight never reached values over 0.8 OD. Interestingly, peptide S-445, featuring the highest number of recognitions with many values above 1 OD (7/51 of all vaccine cohort samples; three of them from the NVX-CoV2373 cohort), is part of the Spike RBD. Altogether, the humoral response to vaccination turned out to be individually diverse yet relatively limited in its breadth with respect to the recognition of the 28 selected peptides. However, six peptides stood out because of their differences between cohorts (Kruskal-Wallis p-value < 0.05; see Supplementary Table S3 ). At closer inspection though, five of these, S-293, S-309, S-805, S-849, and S-937 yielded very low signal intensities and upon post-hoc tests, three of them (S-805, S849, and S-937) failed to reveal group-specific differences (Dunn’s test p > 0.05).

In contrast, peptide S-657 exhibited robust signals that distinguished the AZD1222 and BNT162b2 vaccinees not only from the pre-pandemic control group but also from the NVX-CoV2373 vaccinees ( Figure 4A ; Dunn’s post hoc test p-values < 0.05). Though the BNT162b2 plasma samples exhibited a trend towards higher median values (0.461 vs. 0.337), the difference to the AZD1222 group was not statistically significant (Dunn’s post hoc p-value = 0.278). The differences between AZD1222 and BNT162b2 on the one hand and NVX-CoV2373 on the other were also apparent for Spike RBD recognition as measured by the diagnostic immunoassay ( Figure 4B ). This latter finding suggested that the differences in the extent of the humoral response between vaccine groups were not limited to S-657. As opposed to our linear peptide ELISA, the diagnostic anti-SARS-CoV-2 immunoassay uses a recombinant protein representing the whole RBD as antigen and can therefore detect antibodies towards discontinuous epitopes. In summary, our approach yielded the linear peptide S-657 – located close to the S1/S2 furin cleavage site - that elicited antibody responses following vaccination with AZD1222 and BNT162b2, but significantly less so after NVX-CoV2373 vaccination.

Access to plasma samples from vaccine groups AZD1222 and BNT162b2, who received a third, i.e. booster immunization, allowed us to look at the dynamics of the response to selected linear B-cell epitopes. We compared the time point 6 months (T1) to 12 months (T2, i.e. 3 months after the booster) and 18 months (T3, 9 months after the booster) after primary immunization. Of note, AZD1222 participants were vaccinated following a heterologous regimen, since boosters consisted of mRNA vaccines (n=7 BNT162b2 and n=1 mRNA1273 from Moderna). In addition, two vaccinees from the BNT162b2 cohort received a second booster at 15 months.

For longitudinal analyses, we focused on seven peptides that showed either group-specific differences or high individual responses. The resulting profiles over the three-time points showed individual time courses ( Figure 5 ; data in Supplementary Table S4 ). Most frequently, ELISA signals increased from T1 to T2 and waned again towards T3 with overall robust reactivity. Examples for this scenario are AZD1222 cohort members A18 and A51 or BNT162b2 vaccinee B18 (all for the three peptides S-657, S-1145, and S-1177). The statistical analysis stated that within the AZD1222 cohort, six of seven peptides had the highest response at time point T2 (Friedman test p < 0.05; exact values given in Supplementary Table S5 ). The same was true for two peptides (S-657 and S-1177) among the BNT162b2 vaccinated cohort. In another scenario, high ELISA signals remained elevated over the three-time points, as seen in donors A28 (S-1177), A46 (S-813), or B1 (S-409). However, there was also a scenario where initial reactivity was not boosted, e.g. cases A13, B1, and B43 (all for peptide S-593). Looking for differences between the AZD1222 and BNT162b2 vaccine cohorts, we found significantly higher reactivities for the AZD1222 group and peptides S-813 (at T2 as well as T3) and S-1177 (T3; Mann-Whitney test with p < 0.05; Supplementary Table S5 ). In summary, vaccinees benefited from booster vaccines in terms of increased antibody reactivities against various epitopes.

In order to evaluate whether changes in linear peptide reactivity paralleled the more general anti-Spike(RBD) response, we focused again on peptide S-657 and compared its reactivity to anti-RBD titers and neutralizing capacities. Indeed, we found significant correlations for the individual donors between linear peptide and S(RBD) reactivity and between linear peptide reactivity and neutralizing capacity for the time point T2 at 12 months ( Figure 6 ; Supplementary Table S5 ). Spearman correlations were at ρ > 0.8 and significance level p < 0.012 for the AZD1222 cohort for both correlations. For the BNT162b2 plasma samples, we determined ρ = 0.6 at p = 0.048 for the peptide to S(RBD) antibody value correlation and ρ = 0.9 with p < 0.001 for the comparison between peptide reactivity and neutralization capacity. In summary, the linear S-657 peptide response appeared as a surrogate for both anti-S(RBD) binding antibody units (BAU) and international units (IU) of neutralizing capacity.