Study population and samples

COVID-19 recovered patients (n = 397) were prospectively recruited at the pulmonology service of Hospital Clínic de Barcelona (Spain) between April and December 2020. Inclusion criteria were as follows: (1) adults aged ≥ 18 and ≤ 90 years; (2) confirmed or suspected SARS-CoV-2 infection; (3) approximately 35 ± 5 days since symptom onset, positive RT-qPCR, or hospital discharge; (4) ability to perform all the clinical tests and understand the study’s process and purposes; and (5) signed written informed consent. The exclusion criteria applied were: (1) immunocompromised state or chronic immunosuppressive medication; (2) untreated or uncontrolled chronic viral infection (e.g. HIV, HBV, HCV); (3) malignancy requiring active treatment within the past three months; and (4) any physical or psychological disorder that might interfere with protocol compliance. Additionally, 39 serum samples were obtained from the Blood and Tissue Bank (BTB) in Barcelona from anonymous donors, and 73 plasmas were collected by ISGlobal from SARS-CoV-2 IgG-positive healthcare workers, resulting in a total sample size of 509 participants (Fig. 1). The demographics of the anonymous donors were not available.

Study samples flow-chart.

The study was approved by the Clinical Research Ethics Committee of the Hospital Clínic de Barcelona (Ethics Approval Reference: HCB/2020/0332) and was conducted in accordance with applicable legislation (Spanish Biomedical Research Law 14/2007). Written informed consent was obtained from all study participants. Samples were anonymized with a successive numerical code that can only be related to the clinical history by the study investigators. Demographics of the anonymous patients (n = 112) were not available.

For patients attending Hospital Clínic, a single visit was conducted to: (1) record socio-demographic and clinical data, (2) collect finger-prick blood samples, and (3) confirm IgG and/or IgM seropositivity against SARS-CoV-2. Socio-demographic and clinical data included: age, sex, date of symptom onset and diagnosis, diagnostic method and result, symptoms (malaise, fever, cough, myalgia, anosmia, ageusia, headache, gastrointestinal symptoms, dyspnoea, others), presence of co-infection, hospitalization and length of stay, oxygen therapy, intensive care unit (ICU) admission and length of ICU stay, days under mechanical ventilation, treatment, and complications (acute respiratory distress syndrome [ARDS], bilateral pneumonia, organizing pneumonia, coagulopathy, others). Patients were classified into four groups depending on COVID-19 severity: (1) asymptomatic (absence of symptoms during infection), (2) mild (presence of mild clinical signs or symptoms only requiring symptomatic treatment), (3) moderate (concomitant medication [other than symptomatic] was added, and/or oxygen therapy requirement, and/or hospital admission), and (4) severe (intermediate care unit or ICU admission, and/or high flow oxygen, and/or mechanical ventilation requirements).

Lateral flow immunoassay classification

Finger-prick blood samples were used to confirm seropositivity against SARS-CoV-2 using lateral flow immunoassays (Abingdon Health, York, UK). The results were classified into four groups: (1) negative, (2) IgM positive, (3) IgM and IgG positive, and (4) IgG positive. IgG results were further categorized into (A) weak positive (IgG+), (B) positive (IgG++), and (C) strong positive (IgG+++). This classification was performed visually by a single, trained researcher to ensure consistency across the samples. The results were cross-validated by multiple trained individuals to minimize subjectivity and ensure reliability of the visual inspection.

Additionally, blood samples were subsequently processed and stored at -80ºC at IDIBAPS Biobank until Luminex immunoassays.

Computational pipeline to refine bioinformatic B-cell epitope predictions

We implemented Brewpitopes⁹, a computational pipeline to refine B-cell epitope predictions from linear predictors (BepiPred v2.0¹² and ABCpred¹⁰) and for conformational tools (Discotope v2.0¹¹) on the S protein and predict the epitopes most likely to be recognised by neutralising antibodies. Our pipeline integrated predictions of linear and structural epitopes into consensus epitope regions using the in-house Epixtractor and Epiconsensus modules⁹. Then, these candidates were prioritized based on multiple factors influencing the surface accessibility of the epitopes and exposition to neutralizing antibodies. These consisted of the localization on the viral membrane or extra virality, which was predicted using CCTOP¹³ and the Epitopology module; the absence of glycosylated residues in the epitope sequence, predicted using Net-N-Glyc¹⁴, Net-O-Glyc¹⁵ and Epiglycan⁹; and the surface accessibility in the 3D conformation of their parental protein, which was predicted using the Episurf module to calculate the residue solvent accessibility (RSA) of the epitope residues⁹. Residues with higher RSA than 0.2 were considered exposed. Additionally, predicted peptides shorter than 6AA were removed due to lack of sequence specificity. All this pipeline has been streamlined to enhance its throughput to facilitate proteome-wide predictions – as described for the SARS-CoV-2-proteome in our previous study⁹.

Conservation of epitope candidates across SARS-CoV-2 variants was evaluated using a complementary in silico analysis. Amino acid mutations corresponding to the main SARS-CoV-2 Variants of Concern (VOCs) — Alpha, Beta, Gamma, Delta, and Omicron — were retrieved from the CoVariants server (https://covariants.org), which is powered by GISAID. Using these data, we reconstructed variant-specific S protein sequences by introducing the corresponding mutations into the ancestral sequence via an in-house R script (fasta_mutator.R), as previously described⁹. Our candidate peptides were then mapped onto these sequences to assess their conservation across variants. This step was incorporated to prioritize epitope regions that are resilient to viral antigenic drift and potentially suitable for broad-spectrum antibody targeting.

Peptide synthesis

Peptide epitopes (peptides) were synthetized by BCN Peptides (BCN Peptides, Spain). During the process, amino acids were obtained as protected derivatives to prevent unwanted reactions. Solvents such as dimethyl sulfoxide (DMSO) and methanol were utilized to dissolve reagents and establish the necessary reaction environment. Coupling reagents like DIC and HOBt facilitated bond formation between amino acids. Temporary blocking of specific functional groups on amino acids was achieved using protecting groups like Fmoc and Boc to prevent undesired reactions. Peptides were typically synthesized via solid-phase synthesis, wherein the initial amino acid was attached to a solid support, and subsequent amino acids were sequentially added to construct the peptide chain. Activation and coupling occurred by activating amino acid residues with a coupling reagent and coupling them to the growing peptide chain on the solid support. Once the desired peptide sequence was assembled, protecting groups were removed to expose functional groups. Subsequently, the peptide was cleaved from the solid support using a cleavage reagent such as trifluoroacetic (TFA). The crude peptide mixture underwent purification techniques such as chromatography to yield a highly pure peptide. Finally, the synthesized peptide was characterized using methods like high-performance liquid chromatography (HPLC) and mass spectrometry to confirm its identity and purity.

Antigen epitope mapping by luminex

Plate format and sample processing

Each Luminex assay plate used 384-well plates to allow for high-throughput processing. Each plate included 126 pre-pandemic plasma samples to determine seropositivity cut-offs. The multiplexing was done in line with the standard Luminex protocol for epitope mapping, ensuring consistency across wells and plates¹⁶.

Epitope mapping

Antigen epitope mapping was conducted through x-MAP technology (Luminex Corp., USA) at ISGlobal to identify those epitopes exhibiting the highest IgG reactivity (measured by median fluorescence intensity, MFI) in convalescent serum samples. Initially, peptides were conjugated to X-MAP avidin magnetic beads via biotin conjugation with polyethylene glycol 12 (PEG-12), following the manufacturer’s instructions (Luminex Corp.). Briefly, 1 mL of beads (1.25 × 10⁶/mL) was sonicated for 20 s, washed and resuspended at 10,000 beads/ul in PBS-BSA 1%, followed by 20 s of vortexing and 20 s of sonication. Beads were washed thrice with PBS-BSA 1%, resuspended in freshly prepared 5 µg/1 × 10⁶ beads of each peptide and incubated in a rotator for 30 min at room temperature, shielded from light. Subsequently, beads were washed thrice and resuspended at 10,000 beads/ul in PBS-BSA 1%. Finally, beads were stored protected from light at 4 °C until the commencement of the assay.

For the assay, serum samples were incubated with the multiplex at 4ºC overnight in agitation at 800 rpm. The following day, plates were washed and subsequently incubated with PE-Labeled anti-human IgG (MossBio, GTIG-001) for 30 min at room temperature with agitation at 800 rpm. Beads were washed and acquired using a FlexMap3D instrument. Statistical comparisons of MFI values across different groups were performed using the Mann–Whitney U test for independent sample comparisons, and one-way ANOVA followed by Bonferroni correction for multiple comparisons. These tests were selected due to their suitability for non-normally distributed and multiple-group data, respectively, and were applied using GraphPad Prism 9.3.0.

Each plate included a positive control curve, blanks controls, and 126 pre-pandemic samples as negative controls to determine the seropositivity cut-offs for each peptide¹⁷. Typically, home-made cut-off values are estimated using known independent negative sera, sometimes alongside positive ones, which are included in the titer plates amongst the unknown samples. A general formula for determining a cut-off value involves calculating the mean and standard deviation (SD) of independent negative control readings, (i.e., cut-off = mean + 3SD)^18,19. In our study, cut-off values were defined following this approach, using the mean plus three standard deviations (mean + 3SD) of the MFI values obtained from 126 pre-pandemic negative control samples included in each plate.

The frequencies of positive samples (above the seropositivity cut-off) for each epitope, were utilized to evaluate the most immunogenic peptides. Multiplexing tests were carried out, and the MFIs between singleplex and multiplex assays were compared at various dilutions (1:50, 1:1000, 1:10000, 1:20000, 1:40000), demonstrating consistent MFI values across formats, confirming the reliability of multiplexing.

Production of monoclonal antibodies

mAbs were produced using Phage Display technology by ProteoGenix S.A.S (https://www.proteogenix.science/, Schiltigheim, FRANCE). The three selected epitopes, 3, 11 and 25 (synthesized as peptides < 20 AA with >95% purity, as required by Proteogenix S.A.S to scale up to antibodies production), were conjugated to BSA, OVA, KLH, and biotin carriers. A human immune scFv (single-chain variable fragment) library, LiAb-SFCOVID-19TM, derived from patients who recovered from COVID-19 (with a high diversity of 1.19 × 10^10 variants), was used for panning. Preliminary depletion against BSA, OVA, and KLH was performed before panning.

The elution of phage binders was performed using Glycine-HCl. The concentration of eluted phages (output pfu) was determined using E. coli TG1, which was also used for the amplification of the eluted phages.

ELISA immunoassays were conducted to screen for polyclonal (phage pools) and monoclonal phages (single phage binders) using an anti-phage-HRP antibody. ELISA plates were coated with either peptide conjugated to carriers or with the carriers alone. For monoclonal phages, single TG1 clones were randomly picked from plates and cultured with a helper phage followed by supernatant collection and screening. Positive single clones were then sequenced, followed by recombinant expression of the antibodies using ProteoGenix’s proprietary XtenCHOTM cell line and purification on a ProteinA/G resin. The QC of recombinant antibodies was performed by Bradford assay using BGG standard (concentration determination), SDS-PAGE (purity and integrity assessment) and ELISA (functionality assessment).

Engineering framework for ScFv

The epitope binder scFv fragments were engineered onto human IgG heavy and light chain constant regions. The human framework used was based on standard therapeutic antibody formats, ensuring compatibility for downstream applications, including therapeutic and diagnostic purposes.

Monoclonal antibodies affinity and kinetic analysis with surface plasmon resonance biosensor

The determination of binding kinetics and affinity constants of monoclonal antibodies mAb11 (mAb11-G12 and mAb11-E1 clones) and mAb25 (mAb25-F9 and mAb25-H6 clones) for the peptide 11 (Pep11) and 25 (Pep25) was conducted using a Surface Plasmon Resonance (SPR) biosensor, enabling real-time monitoring of biomolecular interactions. A proprietary biosensor previously described was utilized, operating under wavelength interrogation²⁰. Pep11 and Pep25 conjugated to BSA carrier protein (Pep11-BSA and Pep25-BSA) were employed. The conjugates were covalently immobilized on gold sensor chip surfaces, previously modified with carboxylic groups. Briefly, a mixed self-assembled monolayer (SAM) comprising 16-mercaptohexadecanoic acid (MHDA) and 11-mercapto 1-undecanol (MuOH) (MHDA/MUOH, 1:5 ratio) was formed on the surface as previously described²¹.

Carboxyl groups were then activated via carbodiimide ester formation using a mixture of 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC) and N-hydroxysulfosuccinimide ester (NHS) (0.2 M/0.05 M in MES buffer) to facilitate the covalent binding to the Lys groups of the BSA in the peptide conjugates (100 µg/mL in acetate buffer pH 4.5). A final step of blocking with ethanolamine (EA, 1 M) was performed to inactivate any remaining activated groups. The reaction was monitored in real-time within the biosensor to appropriately select optimal biofunctionalization conditions. The SARS-CoV-2 S protein (S1 subunit, 2019 nCOV, original variant) was directly immobilized following an analogous protocol (20 µg/mL in MES buffer pH 5). A control sensor chip consisting of a BSA biofunctionalized surface was prepared for control experiments employing the same conditions as described previously for the peptide conjugates ([BSA] = 20 µg/mL). MilliQ water was used as a running medium throughout the immobilization protocols.

The assessment of binding and association characteristics of each antibody for the Pep11 and Pep25 and the S1 protein was conducted in real-time. The running buffer was switched to PBS 10 mM pH 7.5 and successive injections of the antibody at various concentrations in PBS 10 mM (ranging from 6.6 to 264 nanomolar [nM] depending on the antibody) were performed. A constant flow rate of 20 µL/min was maintained throughout the measurements. Association and dissociation curves were generated for each concentration. The surfaces modified with peptides and protein were fully regenerated employing NaOH 25 mM injected at a flow rate of 30 µL/min to completely dissociate any remaining interactions. The real-time association curves were fitted to a binding-kinetics interaction model (Association and then Dissociation equation) using GraphPad Prism 9.3.0. This equation determines a comprehensive kinetic profile by globally fitting the data obtained with multiple analyte concentrations (nM) and analysing the sequential association (ka) and dissociation (kd) rates from the plot of ΔλSPR (nM) vs. time (min) of a minimum of four dose-responsive curves with their respective replicates. All the analyses assumed that binding follows the law of mass action, and the fitting method used was the Least Squares Regression. To provide a thorough analysis of the goodness of the fit, the adjusted R-squared was reported as an indicator of the fitting confidence, the Akaike’s Information Criterion (AIC) value was reported for model validation, and the Sy.x value was reported for the dispersion measure of the residuals. A control surface with native BSA immobilized following a similar protocol was prepared to assess any potential nonspecific interactions of the antibodies with the protein, considering the highest concentration of Ab tested in each case (Figure S1 in Supplementary Information). The association and dissociation curves were fitted using the association-dissociation model equation via non-linear least squares regression (GraphPad Prism 9.3.0). Model performance was assessed using adjusted R², Akaike Information Criterion (AIC), and the standard deviation of residuals (Sy.x), ensuring robustness of parameter estimation.