Construction of spike-pseudotyped viruses containing single or combined mutations in spike

The identification of mutations in SARS-CoV-2 genome of considerable prevalence or that emerged independently over time is important, not only to track viral evolution and dynamics of lineages circulating worldwide, but also to identify recurrent mutations and their effects, and ultimately tailor preventive measures to contain the virus. For this study, we selected only mutations in spike (that emerged between 29^th of December 2019 and 25^th of March 2021, S1 Fig) because this protein is responsible for viral recognition of ACE2 receptor [6,49–52] and for inducing neutralizing antibodies in the host [10] and, therefore, mutations in spike may hamper therapies, populational immunity and vaccination-mediated protection [53–55]. Whilst mutations on the RBD may sterically block binding of spike to ACE2, mutations in different spike domains may affect the conformation of the protein thereby impacting antibody recognition. Given this, for this study we selected mutations in spike to include variants of concern and of interest (up to all mutations in lineages B.1.1.7, B.1.351 and P.1; S1 Table), of high prevalence (S1 Fig), or convergent appearance in different lineages over time, as shown in Fig 1B and described in S1 Table. In addition, the mutations in the RBD region are highlighted also in the open (‘up’) conformation (Fig 1C), showing the conformational changes that each amino acid residue undergoes in this region.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Location of the mutations in spike tested in this work.

(A) Schematic representation of the SARS-CoV-2 spike primary structure. NTD, N-terminal domain; RBD, receptor-binding domain; S1/S2, S1/S2 protease cleavage site; S2’, S2’ protease cleavage site; FP, fusion peptide; HR1, heptad repeat 1; HR2, heptad repeat 2; TM, transmembrane domain; CT, cytoplasmic tail. The protease cleavage sites are indicated by arrows. (B) The mutations in spike engineered in this study are mapped onto the spike protein surface, using a structure obtained by homology-based modelling with the structure PDB ID 6XR8 as a template. Mutations are highlighted using magenta spheres mapped on one of the monomers, which is represented with different colors to highlight important regions: NTD (green), RBD (yellow), FP (pink), HR1 (cyan). The remaining monomers are displayed in grey. (C) The structure of the spike protein with one RBD up and bound to ACE2 (PDB ID 7DF4) is shown to highlight the mutations used in this study located in the RBD in the context of its interaction with the receptor. The spike protein is represented in the same color scheme used in (B) and ACE2 is displayed in teal.

https://doi.org/10.1371/journal.ppat.1009772.g001

The different spike versions served to engineer lentiviral spike-pseudotyped particles through a three-plasmid approach: a plasmid harboring the lentiviral genome, lacking Gag-Pol and envelope proteins, and encoding a GFP reporter; a Gag-Pol expression plasmid; and a plasmid expressing SARS-CoV-2 spike protein (or VSV-G protein, as control). Cells infected by these lentiviruses express GFP and can be easily quantified with high-throughput analytical methods.

Neutralization assays on spike-pseudotyped viruses to evaluate synergetic effects of mutations

For the neutralization assay, we developed a 293T cell line expressing the human ACE2 receptor (S2 Fig). The specificity of the assay was assessed using an anti-spike polyclonal antibody and an RBD peptide that competes with the spike protein for viral entry in SARS-CoV-2 spike-pseudotyped but not in VSV G-pseudotyped lentivirus that was used in parallel (S3 Fig).

In this study, we evaluated evasion of host neutralizing antibodies. We characterized the serum antibodies from 12–16 health care workers, infected with SARS-CoV-2 during spring/summer 2020, for their ability to neutralize our spike-pseudotyped mutant lentiviruses. These sera were tested by ELISA for their anti-spike content and classified as low (endpoint titer up to 1:150), medium (endpoint titer 1:450) and high (endpoint titer from 1:1350) (S2 Table). We also used 4 negative sera as control (one pre-pandemic and 3 contemporary). Representative neutralization curves of the wild-type (WT, 614G) and original (614D) strains, as well as variants of concern B.1.1.7, B.1.351 and P.1 (containing all defining mutations, Fig 2A–2D and S1 Table) are shown in Fig 2E and the complete set of neutralization curves in S4 Fig. For each neutralization curve, we calculated the half-maximal neutralization titer (NT₅₀), defined as the reciprocal of the dilution at which infection was decreased by 50% (Fig 2F). Consistent with the literature, the ELISA titer was associated with the neutralizing titer of sera (S2 Table) [45,56].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Neutralization of SARS-CoV-2 spike mutants by convalescent human sera.

(A-D) SARS-CoV-2 spike secondary structure of the original strain (A), the B.1.1.7 variant first identified in the UK (B), the B.1.351 variant (SA) (C) and the P.1 variant (Brazil) (D). Insets show spike structure surrounding each mutated amino acid, and the corresponding structure in the WT. (E-H) Neutralization assays were performed by infecting 293T-ACE2 cells with pseudoviruses displaying WT or mutant SARS-CoV-2 spike, in the presence of serial dilutions of a convalescent human sera cohort (n = 12–16). (E) Representative neutralization curves of sera with low, medium, and high titers of anti-spike IgG antibodies against WT virus, original virus, and variants of concern B.1.1.7, B.1.351 and P.1. Relative infection is defined as the percentage of infection relative to cells infected in the absence of serum. (F) Paired analysis of neutralizing activity of each serum against WT vs original virus or variants of concern. NT₅₀ is defined as the inverse of the dilution that achieved a 50% reduction in infection. Dashed lines indicate the limit of detection of the assay (NT₅₀ = 30). ns, non-significant, **p<0.01, ***p<0.001 by two-tailed Wilcoxon matched-pairs signed-rank test. (G) Ratios of neutralization between variant and WT viruses. (H) Ratios of neutralization between mutant viruses with single, double, or triple mutations and WT virus. Red bars indicate the geometric mean. See S4 and S5 Figs and S2 Table for the complete set of neutralizing curves, neutralizing titers, anti-spike IgG titers determined by ELISA and the numerical values plotted in this figure.

https://doi.org/10.1371/journal.ppat.1009772.g002

In this paper, we used the Wilcoxon statistical test when comparing NT₅₀s to understand if variants are significantly different from the WT. However, to define whether variants are capable of evading neutralization, we use the 4-fold criteria, similar to what has been done with influenza virus, where differences in hemagglutination assays greater than 4-fold highlight the need for vaccine update [33,57,58]. Data show that, although the spike including all defining mutations of B.1.1.7 exhibited a modest reduction in neutralization titers relative to WT spike, it was still efficiently neutralized by convalescent sera, as observed before (Fig 2F and 2G) [32,59]. Conversely, the complete set of defining mutations of B.1.351 and P.1 consistently showed a reduction in neutralization capacity higher than 4-fold and are considered significantly different from the WT virus (Fig 2G). We then analysed a series of single, double and triple mutations (Figs 2H, S4 and S5). We found that the mutations N501Y, Δ69-70/N501 and Δ69-70/N501+P681H, associated with B.1.1.7 variant, behave as WT spike. The mutation E484K alone reduced sera neutralization capacity by 3.6-fold [ratio NT₅₀ mutant/NT₅₀ WT of 0.028±0.37 (geometric mean ± SD)], and inclusion of additional mutations (K417N and N501Y) further decreased the neutralization (0.08±0.05), emphasizing the role of synergic mutations on immune evasion and underscoring the need for evaluating single and combined mutations on variants [48]. Other mutations outside the RBD did not significantly alter sera neutralization efficiency. However, we observed a reduction in neutralization in some sera against N439K mutant (0.53±0.45) as reported in [38], which was not observed when Δ69–70 was added (Figs 2H and S5), and also against L452R mutant (0.48±0.23) that is associated with the most recent variants of concern B.1.427/9, as reported in [60].

Importantly, we tested BNT162b2-elicited plasmas (collected after the first and second dosage of the Pfizer–BioNTech vaccine) for their capacity to neutralize the spike-pseudotyped particles expressing WT spike protein or mutants bearing all defining mutations of the variants of concern B.1.1.7, B.1.351 and P.1. Amongst 10 individuals, plasma collected 12 days after the administration of the first vaccine dose displayed anti-spike IgG titers from 1:450 to 1:12150 (S3 Table), and only one of them exhibited neutralizing capacity (S3 Table and S6 Fig). BNT162b2-elicited plasma collected 12 days after the administration of the 2^nd dose had very high levels of anti-spike IgG (titers 1:36450–1:109350), higher than any sera collected upon natural infection, and all exhibited neutralizing activity against the variants of concern B. 1.351 and P.1, although with significantly lower efficiency than against the wild type or the B.1.1.7 variant (Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Neutralization of SARS-CoV-2 variants of concern by post-vaccination plasma.

Neutralization assays were performed with pseudoviruses displaying WT SARS-CoV-2 spike or variants of concern B.1.1.7, B.1.351 and P.1, in the presence of serial dilutions of post-vaccination plasmas from 10 individuals, 12 days after the first and the second rounds of vaccination. (A) Representative neutralization curves of post-vaccination plasma against WT virus and variants of concern. (B) Paired analysis of neutralizing activity each of plasma against WT vs variants of concern. Dashed lines indicate the limit of detection of the assay (NT₅₀ = 30). *p<0.05, ***p<0.001 by two-tailed Wilcoxon matched-pairs signed-rank test. (C) Ratios of neutralization between variant and WT viruses. Red bars indicate the geometric mean. See S6 Fig and S3 Table for the complete set of neutralizing curves, the anti-spike IgG titers of each plasma and the numerical values plotted in this figure.

https://doi.org/10.1371/journal.ppat.1009772.g003

Collectively, these data show that the spike-pseudotyped lentiviral particles used in this study, in which GFP is the output and is measured by high throughput and cheap methodologies, is suitable to quantitatively assess how single and synergetic mutations in full-length spike affect neutralizing-antibody responses.

Frequency of contacts spike-antibodies and spike-ACE2

We then asked if we could find signatures in the spike protein leading to escape in antibody neutralization. To be able to escape antibodies while maintaining the ability to efficiently infect cells, the mutations introduced must destabilize the interaction with antibodies while keeping a high binding affinity for ACE2. Thus, we reasoned that mutations occurring in residues that are relevant for antibody binding but not very relevant for the interaction with ACE2 can be advantageous for the virus and, therefore, we needed to establish which amino acid residues are involved in antibody binding and in RBD binding.

To determine which spike protein residues are more frequently targeted by neutralizing antibodies, we analyzed 57 structures retrieved from the protein data bank (PBD) containing the spike protein (or only the RBD region) bound to antibodies (S4 Table). Our results revealed that the receptor binding motif (RBM), in RBD, is the region with the highest frequency of contacts (Fig 4A and 4B), consistent with what has been documented [11,61–64]. This region is important for the SARS-CoV-2 infection since it mediates contacts with ACE2 [50,65,66] and, for this reason, the binding of antibodies to this region would hinder ACE2 binding and, consequently, the entry of the virus in the cell.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Spike protein-antibody and RBD-ACE2 contact analysis.

(A) Histogram of the contact frequency of each spike residue with antibodies using a cut-off value 4.5 Å to define a contact. (B) Zoom in of the histogram shown in (A) in the region comprising the residues with the highest frequency of antibody contacts (located in the RBD). (C) Antibody contact map represented on the spike protein surface, seen from front (left side) and from top (right side); the map was built with VMD [108] and the color gradient is used to represent the log of the frequency of contact with Ab as shown in the scale bar. (D) Histogram of the contact frequency of RBD residues with ACE2, obtained from MD simulations, using a cut-off value 4.5 Å to define a contact. (E) Scatter plot correlating the RBD residues’ frequency of contact with ACE2 and antibodies; each point represents one residue and those that were studied experimentally are highlighted in bold; the quadrants are filled with different colors to highlight the relevance of different residues as potential mutation hotspots, with the most relevant residues located in quadrant IV and shown as red points. (F-I) Neutralization assays were performed against pseudoviruses displaying WT or S494P spike mutants, in the presence of serial dilutions of convalescent sera (F, G) or post-vaccination sera 1 month after the first and the second rounds of vaccination (H, I). (F, H) Paired analysis of neutralizing activity of convalescent sera (F) or post-vaccination sera (H) against WT vs S494P, S494P/N501Y, E484K/S494P or E484K/S494P/N501Y mutants. Dashed lines indicate the limit of detection of the assay (NT₅₀ = 30). ns, non-significant, *p<0.05, **p<0.01, ***p<0.001 by two-tailed Wilcoxon matched-pairs signed-rank test. (G, I) Ratios of sera neutralization between S494P mutants and WT viruses. Red bars indicate the geometric mean. See S7 and S8 Figs and S4 Table for protein-antibody contact frequency histogram, overall representation of amino acid residues relevant for antibody and ACE2 binding and list of references of the structures of the antibodies used, respectively; S9 and S10 Figs and S5 and S6 Tables for the complete set of neutralizing curves, the anti-spike IgG titers of each serum and the numerical values plotted in this figure.

https://doi.org/10.1371/journal.ppat.1009772.g004

In addition to contacts with the RBD, we also identified other regions of the spike protein with antibody binding sites. The antibodies FC05 and DH1050.1 bind to segments in the N-terminal region (between residues 143–152 and 246–257), region identified as an antigenic site [54], and 2G12 binds to residues 936–941 in the stalk (Fig 4B and 4C).

To choose the cut-off above which to select the most important residues for antibody binding, we built a histogram (S7 Fig). Three distinct groups appear: the first includes residues that have an interaction probability below 20%, the second includes residues whose interaction is between 20% and 45% and the last corresponds to frequent binders (interaction probability over 45%). Within this last group, we find Y449, L455, F456, E484, F486, N487, Y489, Q493, S494, and Y505 residues (Fig 4A and 4B). More than half of these residues have hydrophobic side chains, showing that, within the batch of antibodies we studied, the most common binding mode is through hydrophobic contacts. These residues are important for antibody binding, which means that mutations within this group may enable the virus to escape antibodies from patients previously exposed to the WT or another variant, or people vaccinated using WT sequences of spike. In fact, prior studies have shown that mutations on most of these residues influence antibody escape [2,67]. For residues L455, F456 and F486, mutations for less hydrophobic residues have been shown to reduce binding by polyclonal antibodies [67], and mutants at site 487 escape human monoclonal antibodies COV2-2165 and COV2-2832 [2,63]. Interestingly, our results also show that there is a high prevalence of antibodies binding to E484 (~50%, Fig 4B).

To determine which amino acid residues in spike engage in ACE2 binding, we performed molecular dynamics (MD) simulations of the RBD bound to ACE2. These allowed us to determine which RBD residues are relevant for binding to ACE2, and the persistence of these interactions (Fig 4D). Combining the information on antibody and ACE2 binding allows us to predict which are the most relevant amino acid residues (high affinity for antibodies and low affinity for ACE2). Using these criteria, two residues stand out: E484 and S494 (Fig 4E, IV quadrant circled in red). These results are consistent with the evidence that E484 is an important mutation site and bring to light a new relevant site: S494. Mutations in S494 have been found in circulating strains, and the S494P mutation was shown to reduce the binding by polyclonal plasma antibodies [67] whilst having no [4] or modest effect [68] in RBD-ACE2 binding.

To elucidate the role of this mutation in antibody neutralization, we used our neutralization assay and found that mutation S494P alone leads to a reduction in neutralization by convalescence sera that is significant (Figs 4F and S9, and S5 Table) despite not reaching our 4-fold criteria (ratio NT₅₀ of mutant in relation to WT of 0.39±0.28, Fig 4G). This value was further reduced when this mutation was combined with N501Y (0.28±0.21, Fig 4F and 4G), or when combined with E484K (0.25±0.32, Fig 4F and 4G). We constructed these spike mutants because we aimed at understanding if mutations that increased the ability to bind ACE2, such as N501Y, could synergize with mutations escaping antibody neutralization. Interestingly, the E484K/S494P/N501Y triple mutant reduced neutralization over the 4-fold criteria applied (0.17±0.34, Fig 4F and 4G). We also tested the ability of these mutants to evade neutralization by post-immunization sera, but the effect was less pronounced (Figs 4H, 4I and S10, and S6 Table). In fact, S494P alone did not significantly reduce neutralization, becoming significant when N501Y (0.70±0.4), or E484K (0.5±0.3) mutations were introduced. The triple mutant displayed the more significant reduction in neutralization, despite not reaching the 4-fold difference (0.34±0.18, Fig 4H and 4I). Importantly, S494P mutation appeared independently on several occasions and is under positive selection [68], and is now a variant in monitoring, especially when combined with E484K or N501Y [20,69]. Our results indicate that the S494P mutation can facilitate the escape of the virus from antibodies, maintaining the ability to bind to the receptor and enter host cells, and should be surveilled worldwide.

Some residues that are important for antibody binding are also involved in frequent interactions with ACE2. Seven of these residues remained bound to ACE2 throughout the whole simulation in all the replicates: L455, F456, F486, N487, Y489, Q493 and Y505 (Fig 4D, top residues in quadrant II). This group includes residue F456, which was found to be a hotspot for escape mutations [67]. Although this may seem contradictory, the same study shows that mutations in this residue were very rarely found in nature and, because of lack of prevalence, were not analyzed [67]. We hypothesize that the low mutation rate for this site may be due to the high relevance of this residue for RBD-ACE2 binding. In fact, deep mutational analysis shows that when mutating F546 to any amino acid residue besides its SARS-CoV-1 counterpart (a leucine), the ACE2 binding affinity is reduced [4]. Even though a viral variant with a single mutation at the site 456 could escape antibodies, it would be unfit to subsist in nature, due to a reduced ability to infect host cells. In addition, some residues such as N501, S417, L452 were shown to be strong ACE2, but weak antibody binders, raising the possibility that these amino acids may undergo mutations that cooperatively synergize with antibody binders to increase the viral fitness of immune escape competent SARS-CoV-2 variants. Interestingly, N501Y, L452R and S417N have been associated with increased viral transmission [18,19,26,36] and, when combined with E484K, augmented resistance to neutralizing antibodies.

Together, the analysis presented here is supported by experimental evidence on mutation-driven binding to ACE2 and escape to antibodies [2–4], as well was with the emergence of natural variants [28,32,38,55,70–72], and may be used to predict mutation hotspots.

Ethics statement

This study was approved by the Ethics committees of Centro Hospitalar Lisboa Ocidental and Centro Hospitalar Lisboa Central, in compliance with the Declaration of Helsinki, and follows international and national guidelines for health data protection. All participants provided informed written consent to take part in the study.

Global variants frequencies

A variant surveillance report was retrieved from GISAID on March 26^th, 2021. A total of 864131 submissions were used to establish the world frequency of each variant using specific amino acid mutations (e.g., S_D614G, etc.) by using command line search text tools. Counts were then divided by week and year, using available metadata for collection date and just variants over indicated prevalence are reported and plotted per graph.

Cells and plasmids

Human hepatocellular carcinoma HuH-7 cells (a kind gift from Dr Colin Adrain, Instituto Gulbenkian de Ciência, Portugal), Human Embryonic Kidney 293T (provided by Prof Paul Digard, Roslin Institute, UK) and 293ET cells (from Dr Colin Adrain) were maintained in Dulbecco’s Modified Eagle Medium (DMEM, Gibco, 21969035) supplemented with 10% fetal bovine serum (FBS, Gibco, 10500064), 1% penicillin/streptomycin solution (Biowest, L0022) and 2mM L-glutamine (ThermoFisher, 25030024), at 37°C and 5% CO₂ atmosphere. Plasmids used in this study were obtained as follows: pLEX.MSC was purchased from Thermo Scientific; psPAX2 and pVSV.G were a kind gift from Dr Luís Moita (Instituto Gulbenkian de Ciência, Portugal); pEGFP-N1 was provided by Dr Colin Crump (University of Cambridge, UK); and pCAGGS containing the SARS-Related Coronavirus 2, Wuhan-Hu-1 Spike Glycoprotein Gene (pCAGGS-SARS-CoV-2-S, NR-5231) was obtained through BEI Resources.

Cloning and spike protein mutations

To facilitate incorporation of spike protein into lentiviral pseudovirions, the C-terminal 19 amino acids encompassing the ER-retention domain of the cytoplasmic tail of spike were deleted. This construct, named SARS-CoV-2 S_trunc, was generated by PCR amplification of the C-terminal region of S without amino acids 1255–1273 (a premature STOP codon was added in the reverse primer), using pCAGGS-SARS-CoV-2-S as template (primers in S7 Table). The amplified sequence was cloned back into pCAGGS-SARS-CoV-2-S, between AgeI and SacI restriction sites, generating the expression vector pCAGGS-SARS-CoV-2-S_trunc. Mutations were introduced in the spike protein by site-directed mutagenesis, with the Quickchange Multi Site-Directed Mutagenesis kit (Agilent) following the manufacturer’s instructions (primers in S7 Table). Lentiviral reporter plasmid pLEX-GFP was produced by PCR amplifying GFP from pEGFP-N1 (primers in S7 Table) and cloning the insert into BamHI–XhoI restriction sites in the multi-cloning site of pLEX.MCS vector. For lentiviral plasmid pLEX-ACE2 production, human ACE2 coding sequence was amplified from Huh7 cDNA and cloned into pLEX.MCS, using XhoI and MluI restriction sites (primers in S7 Table).

Production of 293T cells stably expressing human ACE2 receptor

To produce VSV-G pseudotyped lentiviruses encoding the human ACE2, 293ET cells were transfected with pVSV-G, psPAX2 and pLEX-ACE2 using jetPRIME (Polyplus), according to manufacturer’s instructions. Lentiviral particles in the supernatant were collected after 3 days and were used to transduce 293T cells. Three days after transduction, puromycin (Merck, 540411) was added to the medium, to a final concentration of 2.5μg/ml, to select for infected cells. Puromycin selection was maintained until all cells in the control plate died and then reduced to half. The 293T-ACE2 cell line was passaged six times before use and kept in culture medium supplemented with 1.25 μg/ml puromycin. ACE2 expression was evaluated by flow cytometry (S2 Fig).

Production and titration of spike pseudotyped lentiviral particles

To generate spike pseudotyped lentiviral particles, 3x10⁶ 293ET cells were co-transfected with 8.89μg pLex-GFP reporter, 6.67μg psPAX2, and 4.44μg pCAGGS-SARS-CoV-2-S WT or mutants (or pVSV.G, as a control), using jetPRIME according to manufacturer’s instructions. The virus-containing supernatant was collected after 3 days, concentrated 10 to 20-fold using Lenti-XTM Concentrator (Takara, 631231), aliquoted and stored at -80°C. Pseudovirus stocks were titrated by serial dilution and transduction of 293T-ACE2 cells. At 24h post transduction, the percentage of GFP positive cells was determined by flow cytometry, and the number of transduction units per mL was calculated.

Flow cytometry

Flow cytometry was performed as in [89]. In brief, HEK 293T and 293T-ACE2 cells were prepared for flow cytometry analysis by detaching from the wells with trypsin, followed by fixation with 4% formaldehyde). For analysis of ACE2 expression, cells were stained with a primary antibody against ACE2 (4μg/ml, R&D Systems, catalog no. AF933) followed by a secondary antibody labelled with Alexa Fluor 568 (1:1000; Life Technologies). Analysis of cell populations was performed in a Becton Dickinson (BD) LSR Fortessa X-20 equipped with FACS Diva and FlowJo (BD, Franklin Lakes, NJ) software’s.

Human convalescent sera and post-vaccination plasma/serum

Venous blood was collected by standard phlebotomy from health care providers who contracted COVID-19 during spring/summer 2020, as tested by RT-PCR on nasopharyngeal swabs. Serum was prepared using standard methodology and stored at -20°C. During this period, the variant B.1.1, encoding the WT spike protein, was prevalent in Portugal.

Peripheral blood from health care workers was collected by venipuncture into EDTA tubes at day 12 and day 33 post-immunization with Pfizer BTN162b2 mRNA vaccine and immediately processed. Alternatively, post-vaccination serum was collected at day 22 and day 51 post-immunization with Pfizer BTN162b2 mRNA vaccine. Peripheral blood was diluted in PBS 1x (VWR), layered on top of biocoll (Biowest) and centrifuged at 1200g for 30 min without break. Plasma was collected to cryotubes and stored at -80°C ultra-low freezer until subsequent analysis.

ELISA assay

Direct ELISA was used to quantify IgG anti-full-length Spike in convalescent sera. The antigen was produced as described in [90]. The assay was adapted from [91] and semi-automized to measure IgG in a 384-well format, according to a protocol to be detailed elsewhere. For titer estimation, sera were serially diluted 3-fold starting in a 1:50 dilution and cut off was defined by pre-pandemic sera (mean + 2 standard deviation).

ELISA assay on post-vaccination plasma and serum was performed based on the protocol [92] and modified as described in Gonçalves et al. [93]. Briefly, 96 well plates (Nunc) were coated with 50 μl of trimeric spike protein at 0.5 μg/mL and incubated overnight at 4°C. On the following day, plates were washed three times with 0.1% PBS/Tween20 (PBST) using an automatic plate washer (ThermoScientific). Plates were blocked with 3% bovine serum albumin (BSA) diluted in 0.05% PBS/T and incubated 1h at room temperature. Samples were diluted using 3-fold dilution series starting at 1:50 and ending at 1:10,9350 in 1% BSA-PBST/T and incubated 1h at room temperature. Plates were washed three times as previously and goat anti-human IgG-HRP secondary antibodies (Abcam, ab97215) were added at 1:25,000 and incubated 30 min at room temperature. Plates were washed three times and incubated ~7min with 50 μl of TMB substrate (BioLegend). The reaction was stopped with 25μl of 1M phosphoric acid (Sigma) and read at 450nm on a plate reader (BioTek). Each plate contained 6 calibrator samples from two high-, two medium-, and two low- antibody producer from adult individuals collected at Hospital Fernando Fonseca that were confirmed positive for SARS-CoV-2 by RT-PCR from nasopharyngeal and/or oropharyngeal swabs in a laboratory certified by the Portuguese National Health Authorities [93]. As negative control, we used pre-pandemic plasma samples obtained from healthy donors collected prior July 2019. The endpoint titer was defined as the last dilution before the absorbance dropped below OD₄₅₀ of 0.15.

Neutralization assay

Heat-inactivated serum and plasma samples were four-fold serially diluted over 7 dilutions, beginning with a 1:30 initial dilution. Dilutions were then incubated with spike pseudotyped lentiviral particles for 1h at 37°C. The mix was added to a pre-seeded 96 well plate of 293T-ACE2 cells, with a final MOI of 0.2. At 48h post-transduction, the fluorescent signal was measured using the GloMax Explorer System (Promega). The relative fluorescence units were normalized to those derived from the virus control wells (cells infected in the absence of plasma or serum), after subtraction of the background in the control groups with cells only. The half-maximal neutralization titer (NT₅₀), defined as the reciprocal of the dilution at which infection was decreased by 50%, was determined using four-parameter nonlinear regression (least squares regression without weighting; constraints: bottom = 0) (GraphPad Prism 9). To assess the specificity of our assay, an anti-spike antibody neutralization assay and an RBD competition assay were performed in SARS-CoV-2 spike pseudotyped and vesicular stomatitis virus (VSV) G pseudotyped lentiviral particles, in parallel (S3 Fig). The anti-spike antibody neutralization assay was performed as above, starting with a concentration of 50μg/ml (Thermo Fisher Scientific, PA5-81795) and three-fold serially diluting over 7 dilutions. For the RBD competition assay, 293T-ACE2 cells were pre-incubated with three-fold serial dilutions of SARS-CoV-2 spike’s receptor binding domain, starting with a concentration of 400μg/ml. After 1h incubation at 37°C, supernatant was aspirated and pseudoviruses were added at an MOI of 0.2. The half maximal inhibitory concentration (IC₅₀) was determined using four-parameter nonlinear regression (least squares regression without weighting; constraints: bottom = 0) (GraphPad Prism 9).

Spike protein-antibody contact analysis

To determine which residues of the SARS-CoV-2 spike protein contribute the most for the binding of antibodies, we studied 57 PDB structures containing the spike protein (or only the RBD region) bound to antibodies (S4 Table). These complexes were chosen based on their availability in the PDB repositorium [94] and are all neutralizing antibodies. We first used the MDAnalysis library [95,96] to identify the residues of the antibody and the spike protein located in the interface between the two proteins. This was done for all PDB structures using in-house Python scripts. To determine the relevance of each spike protein/RBD residue in the binding, a distance cut-off value of 4.5 Å was applied as a criterion (i.e., a contact is observed when two residues have a minimum distance lower than 4.5 Å). Finally, the numerical python library (NumPy [97]) was used to calculate the frequency of contact for each spike protein residue with an antibody residue, from the sample of PDB structures analyzed.

MD simulation of the RBD-ACE2 complex

Molecular dynamics (MD) simulations of the RBD bound to the ACE2 protein were performed with the GROMACS 2020.3 package [98], using the Amber14sb [99] force field, starting from the 6m0j structure [49], in a truncated dodecahedron box filled with water molecules (minimum of 1.2 nm between protein and box walls). The TIP3P water model [100] was used and the total charge of the system (-23, including the constitutive Zn^- and Cl^- ions bound to ACE2) was neutralized with 23 Na⁺ ions. Additional Na⁺ and Cl⁺ ions were added to the solution to reach an ionic strength of 0.1M. The system was energy-minimized using the steepest descent method for a maximum of 50000 steps using position restraints on the heteroatom positions by restraining them to the crystallographic coordinates using a force constant of 1000 kJ/mol in the X, Y and Z positions. Before performing the production runs, an initialization process was carried out in 5 stages of 100 ps each. Initially, all heavy-atoms were restrained using a force constant of 1000 kJ/mol/nm, and at the final stage only the only C-α atoms were position-restrained using the same force constant. In the first stage, the Berendsen temperature algorithm [101] was used to initialize and maintain the simulation at 300 K, using a temperature coupling constant of 0.01 ps, without pressure control. The second stage continues to use the Berendsen temperature algorithm but with a force constant of 0.1 ps. The third stage kept the same temperature control settings, but introduced pressure coupling with the Berendsen pressure algorithm [101] with a pressure coupling constant of 5.0 ps and applied isotropically. The fourth stage changed the temperature algorithm to V-rescale [102], with a temperature coupling constant of 0.1 ps, and the pressure algorithm to Parrinello-Rahman [103] with a pressure coupling constant of 5.0 ps. The fifth stage is equal to the fourth stage, but position restraints are only applied to C-α atoms. During the simulation, the equations of motion were integrated using a timestep of 2 fs. The temperature was maintained at 300 K, using the V-rescale [102] algorithm with a time constant of 0.1 ps, and the pressure was maintained at 1 bar using the Parrinello−Rahman [103] pressure coupling algorithm, with a time constant of 5 ps; pressure coupling was applied isotropically. Long-range electrostatic interactions were treated with the PME [104,105] method, using a grid spacing of 0.12 nm, with a cubic interpolation. The neighbor list was updated every twenty steps and the cutoff scheme used was Verlet with 0.8nm as the real space cut-off radius. All bonds were constrained using the LINCS algorithm [106]. The system was simulated for 8 μs in 5 replicates (to a total of 40 μs).

RBD-ACE2 contacts frequency

To determine the residues that contribute the most for the interaction between RBD and ACE2, and the persistence of these interactions, we performed a contact analysis throughout the simulation. We started by eliminating the first equilibration μs of all replicates. The MDAnalysis library [95,96] was then used to pinpoint the residues of the RBD that are in contact with the ACE2 protein. A distance cut-off value of 4.5 Å was applied as a criterion (i.e., a contact is observed when two residues have a minimum distance lower than 4.5 Å). Finally, we determined the percentage of time for which a given RBD residue is at less than 4.5 Å of ACE2.

Homology-based models of spike protein variants

Homology-based models of all the variants analyzed in this study were generated using the software Modeller [107], version 9.23, using the structure of the WT enzyme (PDB code: 6XR8) [12] as a template. The protocol used only optimizes the atoms belonging to the mutated residues and the residues that are located within a 5 Å radius from these residues, maintaining the remaining atoms fixed with the coordinates found in the template structure. The optimization parameters were set to the software default values. The final model corresponds to the one with the lowest value of the DOPE function out of 20 generated structures.