B cell analyses after SARS-CoV-2 mRNA third vaccination reveals a hybrid immunity like antibody response

The continuous evolution of SARS-CoV-2 generated highly mutated variants able to escape natural and vaccine-induced primary immunity. The administration of a third mRNA vaccine dose induces a secondary response with increased protection. Here we investigate the longitudinal evolution of the neutralizing antibody response in four donors after three mRNA doses at single-cell level. We sorted 4100 spike protein specific memory B cells identifying 350 neutralizing antibodies. The third dose increases the antibody neutralization potency and breadth against all SARS-CoV-2 variants as observed with hybrid immunity. However, the B cell repertoire generating this response is different. The increases of neutralizing antibody responses is largely due to the expansion of B cell germlines poorly represented after two doses, and the reduction of germlines predominant after primary immunization. Our data show that different immunization regimens induce specific molecular signatures which should be considered while designing new vaccines and immunization strategies.

The continuous evolution of SARS-CoV-2 generated highly mutated variants able to escape natural and vaccine-induced primary immunity. The administration of a third mRNA vaccine dose induces a secondary response with increased protection. Here we investigate the longitudinal evolution of the neutralizing antibody response in four donors after three mRNA doses at single-cell level. We sorted 4100 spike protein specific memory B cells identifying 350 neutralizing antibodies. The third dose increases the antibody neutralization potency and breadth against all SARS-CoV-2 variants as observed with hybrid immunity. However, the B cell repertoire generating this response is different. The increases of neutralizing antibody responses is largely due to the expansion of B cell germlines poorly represented after two doses, and the reduction of germlines predominant after primary immunization. Our data show that different immunization regimens induce specific molecular signatures which should be considered while designing new vaccines and immunization strategies.
The emergence of SARS-CoV-2 variants of concern (VoCs) able to escape vaccine immunity elicited by the spike (S) protein of the original virus isolated in Wuhan, China, have decreased the impact of vaccination 1,2 . As a consequence, the COVID-19 pandemic continues to impact the health, the economy, and the freedom of people worldwide in spite of the several billion doses of vaccines already deployed. This scenario raises important questions about the use of the existing vaccines and the immunity they can provide to tackle current and future variants. Therefore, it became of utmost importance to understand the nature and the quality of the immune response elicited by the different vaccines used worldwide. In our previous study, we analyzed at a single-cell level the immune response induced by two doses of the BNT162b2 mRNA vaccine in naïve people and in people that had been previously infected by the SARS-CoV-2 virus 3 . In this study, we analyzed at a single-cell level the longitudinal B cell and neutralizing antibody response of the same naïve people after a third immunization. We found that, while the overall immune response after a third dose is similar to that observed in the hybrid immunity of vaccinated people previously infected by the virus, the B cell repertoire that stands behind the response is different.

B cells response after the third dose
To evaluate the longitudinal evolution of the neutralizing antibody response, four seronegative donors that participated in our previous study after two doses of BNT162b2 mRNA vaccine 3 , were re-enrolled after receiving the third dose. None of the subjects were exposed to SARS-CoV-2 infection between the second and third vaccination dose. Data after the third dose (seronegative third dose; SN3) described in this study were compared to those obtained from the same subjects after the second dose (seronegative second dose; SN2) and to those of subjects with hybrid immunity (seropositive second dose; SP2) previously described 3 . Three subjects received the BNT162b2 (VAC-001, VAC-002, and VAC-008) vaccine, while one subject (VAC-010) received the mRNA-1273 vaccine. Blood collection occurred at an average of 58 days post third vaccination dose. Subject details are summarized in Supplementary Table 1 and the gating strategy used is shown in Supplementary Fig. 1a. The frequency of CD19 + CD27 + IgD − IgM − memory B cell (MBCs) specific for the S protein + was 6.18-fold higher in SN3 compared to SN2 (Fig. 1a, b). This was also 2.26-fold higher than that previously observed in subjects with SARS-CoV-2 infection and subsequent vaccination with two doses of the BNT162b2 mRNA vaccine (SP2) 3 . No major differences between the two groups were observed in the CD19 + CD27 + IgD − IgM − /IgM + memory B cell (MBCs) compartments ( Supplementary Fig. 1b-e). In addition, SN3 showed higher binding to the S protein, receptor binding domain (RBD), and N-terminal domain (NTD), and a 5.82-fold higher neutralization activity against the original Wuhan SARS-CoV-2 virus compared to SN2 (Fig. 1c, d and Supplementary Fig. 1f-i).

Boosting antibody potency and breadth
To evaluate the longitudinal B cell response, antigen-specific classswitched MBCs (CD19 + CD27 + IgD − IgM − ) were single-cell sorted using as bait the Wuhan prefusion SARS-CoV-2 S protein trimeric antigen, which was encoded by the mRNA vaccine. Sorted cells were incubated for 2 weeks to naturally release human monoclonal antibodies (mAbs) into the supernatant. A total of 4100 S protein + MBCs were sorted and 2436 (59.4%) produced mAbs able to recognize the S protein prefusion trimer in ELISA. Of these, 350 neutralized the original Wuhan live SARS-CoV-2 virus when tested at a single point dilution (1:10) by cytopathic effect-based microneutralization assay (CPE-MN). Overall, the fraction of S protein-specific B cells producing neutralizing antibodies (nAbs) was 14.4% which is 2.2-fold higher than what was observed in SN2 and comparable to what was observed for SP2 dose vaccinees (14.8%) in our previous study ( Supplementary Fig. 2a Fig. 1e-l). To increase the power of the analyses, in Fig. 1, we included all 52 nAbs isolated from the five seronegative donors enrolled in our previous study 3 . Overall, nAbs isolated in SN3 were higher in frequency and potency against all tested viruses compared to SN2 and showed a significantly higher 100% inhibitory concentration (IC 100 ), except for BA.1 and BA.2 VoCs, where SN2 did not have enough nAbs to make a meaningful comparison (Fig. 1e-l). The IC 100 geometric mean (GM-IC 100 ) in SN3 was 3.25, 7.57-, 3.31-, 3.21-, 5.70-, and 2.83-fold lower compared to SN2 for the Wuhan virus, the alpha, beta, gamma, delta, and BA.1 VoCs, respectively ( Fig. 1 and Supplementary Fig. 3). Only for BA.2 a higher GM-IC 100 was observed in SN2 compared to SN3. Interestingly, SN3 also showed a higher neutralization potency compared to SP2 against all tested viruses with a 1.13-, 3.49-, 2.81, 2.23-, 3.15-, 2.22-, and 1.20-fold lower GM-IC 100 ( Fig. 1l and Supplementary Fig. 3). In addition, we observed that nAbs induced following a third booster dose retained a high ability to crossneutralize all tested SARS-CoV-2 VoCs. Indeed, while 67.3, 21.1, 25.0, 51.9, 1.9, and 7.7% of nAbs maintained the ability to neutralize alpha, beta, gamma, delta, omicron BA.1, and omicron BA.2, respectively, in the SN2 group 3 , the same viruses were neutralized by 77.2, 41.7, 39.3, 61.2, 21.8, and 25.2% of nAbs in the SN3 group ( Fig. 1 and Supplementary Table 2). Finally, we assessed if a third booster dose would enhance the ability of nAbs to recognize and cross-neutralize the distantly related SARS-CoV-1 virus. Two of the four SN3 donors presented nAbs able to recognize SARS-CoV-1 ( Supplementary Fig. 4a). However, only 2.4% (5/206) and 0.9% (2/206) of all nAbs were able to bind the SARS-CoV-1 S protein and neutralize this sarbecovirus respectively, showing a lower frequency than the one observed in SN2 and SP2 nAbs ( Supplementary Fig. 4a, b) 3 .

Antibody classes of cross-protection
To understand the epitope region recognized by our RBD targeting antibodies (n = 154) after a third vaccination dose, we performed a competition assay with three known antibodies. As previously described, we used the Class 1/2 antibody J08 4 , the Class 3 antibody S309 5 , and the Class 4 antibody CR3022 6 to map the epitope regions targeted by our RBD-binding nAbs 3 . The most abundant class of nAbs targeted the Class 1/2 epitope region (98/154; 63.6%), followed by Class 3 targeting nAbs (30/154; 19.5%) and nAbs that recognize the Class 4 region (3/154; 1.9%). The remaining 23 (14.9%) nAbs did not compete with any of the three antibodies used in our competition assay. As shown in Fig. 2a, in SN3, we observed a marked increase in Class 3 and 4 nAbs compared to SN2 and a similar distribution to SP2. When we compared the GM-IC 100 of classes of nAbs in the SN3 with the SN2 and SP2 groups, we observed similar neutralization potency when tested against the SARS-CoV-2 virus originally isolated in Wuhan, China ( Supplementary Fig. 5). Following, we aimed to understand which classes of RBD (n = 154) and NTD (n = 43) targeting nAbs in SN3 were mainly responsible for the cross-protection against the SARS-CoV-2 VoCs. Overall, we observed that Class 1/2 nAbs are the most abundant family of antibodies against all tested variants, followed by Class 3, Not-competing, NTD, and Class 4 nAbs (Fig. 2b, c). Finally, we compared the functional antibody response between SN3 and SP2 against the highly mutated omicron BA.1 and BA.2 viruses. Interestingly, we observed that a third mRNA booster dose increases neutralization potency and evasion resistance of Class 1/2 nAbs against both omicron BA.1 and BA.2 compared to SP2. The opposite trend was observed for Class 3 nAbs, which had a lower frequency of cross-protection in SN3 compared to SP2 (Fig. 2d, e).
Conversely to what found in SP2 3 , we did not observe the expansion of Class 3 targeting antibody germline IGHV2-5;IGHJ4-1, which so far has been observed only in previously infected vaccinees or subjects immunized with adenoviral vectors 3,13,14 . Following this, we aimed to identify expanded clonal families within the same four donors after a second and third vaccination dose. Sequences (SN2 n = 43; SN3 n = 288) were clustered by binning the clones to their inferred germlines (centroids) and according to 80% nucleotide sequence in the VoCs. a The graph shows the frequency of CD19 + CD27 + IgD − IgM − able to bind the SARS-CoV-2 S protein trimer (S protein + ) in SN2 and SN3 donors (n = 4/group). The black line and bars denote the geometric mean. b The table summarizes the frequencies of the B cell populations in SN2 and SN3. c The graph shows the neutralizing activity of plasma samples against the original Wuhan SARS-CoV-2 virus for SN2 (n = 4) and SN3 (n = 4). Technical duplicates were performed for each experiment. d The table summarizes the 100% inhibitory dilution (ID 100 ) of each COVID-19 vaccinee and the geometric mean for SN2 and SN3 (n = 4/group). e-k Scatter dot charts show the neutralization potency, reported as IC 100 (ng ml −1 ), of nAbs tested against the original Wuhan SARS-CoV-2 virus (e) and the alpha (f), beta (g), gamma (h), delta (i), omicron BA.1 (j), and omicron BA.2 (k) VoCs. The number of samples tested (n) and percentage of nAbs from SN2 vs SN3, fold change, and statistical significance are denoted on each graph. l The table shows the IC 100 geometric mean (GM) of all nAbs pulled together from SN2, SN3, and SP2 against all SARS-CoV-2 viruses tested. " †" indicates previously published data 3 . Technical duplicates were performed for each experiment. A nonparametric Mann-Whitney t-test was used to evaluate the statistical significance between groups. Two-tailed p value significances are shown as *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. Source data are provided as a Source Data file.
Article https://doi.org/10.1038/s41467-022-35781-6 heavy complementary determining region 3 (CDRH3). Clusters were defined as antibody families including at least five or more members as previously described 15 . Of the 331 sequences, 226 (68.3%) were orphans (i.e., did not cluster with other sequences), and only in six cases sequences from SN2 and SN3 were binned to the same centroid (Fig. 3b). Only five clusters were identified, three of which were composed by antibodies belonging exclusively from the SN3 group. Of these clusters, two were formed by antibody sequences shared between the SN2 and SN3 groups. The smallest cluster was composed of 11 antibody members encoded from the IGHV1-58;IGHJ3-1 germline, while the biggest cluster, composed of 18 antibody members, were encoded by the IGHV3-48/IGHV3-53/IGHV3-66 germlines (Fig. 3b). Finally, we evaluated the V gene mutation levels, neutralization potency and breadth in nAbs encoded by reduced and expanded germlines following a third vaccination dose. Our data showed that IGHV3-53;IGHJ6-1 and IGHV3-66;IGHJ4-1 germlines, expanded in SN2 and reduced in SN3, have a similar level of V gene somatic mutations, are poorly cross-reactive, especially against omicron BA.1 and BA.2, and show medium neutralization potency (IC 100 mainly between 100 and 1000 ng ml −1 ). Differently, the antibody germlines IGHV1-58;IGHJ3-1 and IGHV1-69;IGHJ4-1, mainly expanded in SN3 and poorly represented in SN2, show between 5 and 15-fold higher V gene somatic mutation levels, higher neutralization potency and breadth against all SARS-CoV-2 VoCs (Fig. 3c-f).

Discussion
In agreement with other longitudinal studies 7,8,16 , we found that the third dose of mRNA vaccine induces an immune response similar to the hybrid immunity observed in people vaccinated after SARS-CoV-2 infection. This antibody response is characterized by a small increase in S protein binding antibodies, a strong increase in neutralizing potency, and a considerable increase in antibodies able to crossneutralize emerging variants, including omicron BA.1 and BA.2. The increased potency and breadth are due to a significant expansion of S protein-specific MBCs, which is even higher than that observed in subjects with hybrid immunity, and by a strong increase of V gene somatic mutations. What is interesting in our study is that the increased potency and breadth observed after a third booster dose was due mostly to Class 1/2 nAbs, while Class 3 antibodies had a lower frequency and breadth compared to subjects with hybrid immunity. In addition, we found that a third mRNA dose did not induce a strong response against the distantly related SARS-CoV-1, suggesting that additional doses of homologous vaccines against SARS-CoV-2 will focus the antibody response against this virus instead of broadening cross-protection to other coronaviruses. Another important observation of this study is that the increased neutralization potency and breadth is not due to a linear evolution of the B cells producing nAbs after two vaccine doses but is due mostly to the expansion of new B cells, which were not detected after primary immunization. Indeed, the secondary response induced by a third vaccine dose did not derive from expanded B cell clones, but it was dominated by singlets that constituted 68% of the entire repertoire. A possible explanation for this scenario is that newly expanded B cell clones needed further maturation, induced by a third booster dose, before acquiring a high level of functionality and being selected for further expansion. In addition, the germlines IGHV3-53;IGHJ6-1/IGHV3-66;IGHJ4-1 dominated the neutralizing response in donors infected with the original Wuhan virus and in subjects immunized with two doses of mRNA vaccines 3,4,17-19 , decreased in frequency and did not improve in potency or crossneutralization after a third dose. Conversely, the antibody germlines IGHV1-58;IGHJ3-1 and IGHV1-69;IGHJ4-1 became largely responsible for the improved potency and cross-neutralization observed after a third dose. Interestingly, the IGHV1-58 germline predominant in this group was previously shown to recognize a "supersite" on the S protein surface and to be expanded following beta or omicron breakthrough infection in vaccinated individuals 10,12,20 . An interesting finding of this work is that the highly cross-reactive germline IGHV2-5;IGHJ4-1, found in subjects with hybrid immunity and in people vaccinated with viral vectors 3,14 , is absent after a third mRNA vaccine, suggesting that the induction of this germline may require endogenous production of the S protein. The lack of expansion of the IGHV2-5;IGHJ4-1 germline after three mRNA vaccine doses could partially explain the above-mentioned loss of cross-reactivity by Class 3 nAbs. Overall, our study provides a high-resolution picture of the functional and genetic properties of a third mRNA vaccination and, despite we observed important similarities with hybrid immunity, unravels features of the antibody response specifically produced after a third mRNA vaccine dose. These observations are very important and should be considered while designing new vaccines and implementing vaccination regimens for booster doses, especially in low-middle-income countries where less than 20% of people received at least one vaccine dose 21 .

Enrollment of COVID-19 vaccinees and human sample collection
This work results from a collaboration with the Azienda Ospedaliera Universitaria Senese, Siena (IT) that provided samples from COVID-19 vaccinated donors, of both sexes, who gave their written consent. The study was approved by the Comitato Etico di Area Vasta Sud Est (CEAVSE) ethics committees (Parere 17065 in Siena) and conducted according to good clinical practice in accordance with the declaration of Helsinki (European Council 2001, US Code of Federal Regulations, ICH 1997). This study was unblinded and not randomized. No statistical methods were used to predetermine the sample size.
ELISA assay with SARS-CoV-2 and SARS-CoV-1 S protein prefusion trimer mAbs and plasma binding specificity against the S protein trimer was detected by ELISA as previously described 3 . Briefly, 384-well plates (microplate clear, Greiner Bio-one) were coated with 3 µg/mL of streptavidin (Thermo Fisher) diluted in carbonate-bicarbonate buffer (E107, Bethyl laboratories) and incubated at RT overnight. The next day, plates were incubated for 1 h at RT with 3 µg/mL of SARS-CoV-2 or SARS-CoV-1 S protein, and saturated with 50 µL/well of blocking buffer (phosphate-buffered saline, 1% BSA) for 1 h at 37°C. Following, 25 µL/ well of mAbs or plasma samples, diluted 1:5 or 1:10, respectively, in sample buffer (phosphate-buffered saline, 1% BSA, 0.05% Tween-20), were added serially diluted step dilution 1:2 and then incubated at 1 h at 37°C. Finally, 25 µL/well of alkaline phosphatase-conjugated goat antihuman IgG and IgA (Southern Biotech) diluted 1:2000 in sample buffer were added. S protein binding was detected using 25 µL/well of PNPP (p-nitrophenyl phosphate; Thermo Fisher) and the reaction was measured at a wavelength of 405 nm by the Varioskan Lux Reader using the SkanIt Software Microplate Readers 6.0.1 (Thermo Fisher Scientific). After each incubation step, plates were washed three times with 100 µL/well of washing buffer (phosphate-buffered saline, 0.05% Tween-20). A sample buffer was used as a blank and the threshold for sample positivity was set at twofold the optical density (OD) of the blank. Technical duplicates were performed for mAbs and technical triplicates were performed for sera samples.
ELISA assay with RBD, NTD, and S2 subunits mAbs identification and plasma screening of vaccinees against RBD, NTD, or S2 SARS-CoV-2 protein were performed by ELISA as previously described 3 . Briefly, 3 µg/mL of RBD, NTD, or S2 SARS-CoV-2 protein diluted in carbonate-bicarbonate buffer (E107, Bethyl laboratories) were coated in 384-well plates (microplate clear, Greiner Bio-one) and blocked with 50 µL/well of blocking buffer (phosphate-buffered saline, 1% BSA) for 1 h at 37°C. After washing, plates were incubated for 1 h at 37°C with mAbs diluted 1:5 in samples buffer (phosphate-buffered saline, 1% BSA, 0.05% Tween- 20) or with plasma at a starting dilution 1:10 and step diluted 1:2 in sample buffer. Anti-Human IgG −Peroxidase antibody (Fab specific) produced in goat (Sigma) diluted 1:45,000 in sample buffer was then added and samples incubated for 1 h at 37°C. Plates were then washed and incubated with TMB substrate (Sigma) for 15 min before adding the stop solution (H 2 SO 4 0.2 M). The OD values were identified using the Varioskan Lux Reader (Thermo Fisher Scientific) at 450 nm. Each condition was tested in triplicate and samples tested were considered positive if the OD value was twofold the blank.

Flow cytometry-based competition assay
To classify mAbs candidates on the basis of their interaction with Spike epitopes, we performed a flow cytometry-based competition assay as previously described 3,9 . Briefly, magnetic beads (Dynabeads His-Tag, Invitrogen) were coated with histidine-tagged S protein according to the manufacturer's instructions. Then, 20 µg/mL of coated S protein beads were pre-incubated with unlabeled nAbs diluted 1:2 in PBS for 40 min at RT. After incubation, the mix of Beadsantibody was washed with 100 µL of PBS-BSA 1%. Then, to analyze epitope competition, mAbs able to bind RBD Class 1/2, (J08), Class 3 (S309), and Class 4 (CR3022) were labeled with three different fluorophores (Alexa Fluor 647, 488, and 594) using Alexa Fluor NHS Ester kit (Thermo Scientific), were mixed and incubated with S protein-beads. Following 40 min of incubation at RT, the mix of Beads-antibodies was washed with PBS, resuspended in 150 µL of PBS-BSA 1%, and analyzed using BD LSR II flow cytometer (Becton Dickinson). Beads with or without S protein incubated with labeled antibodies mix were used as a positive and negative control, respectively. FACSDiva Software (version 9) was used for data acquisition and analysis was performed using FlowJo (version 10).

SARS-CoV-2 authentic viruses neutralization assay
All SARS-CoV-2 authentic virus neutralization assays were performed in the biosafety level 3 (BSL3) laboratories at Toscana Life Sciences in Siena (Italy) and Vismederi Srl, Siena (Italy). BSL3 laboratories are approved by a Certified Biosafety Professional and are inspected every year by local authorities. To evaluate the neutralization activity of identified nAbs against SARS-CoV-2 and all VoCs and evaluate the breadth of neutralization of this antibody is a cytopathic effect-based microneutralization assay (CPE-MN) was performed 3,4 . Briefly, the CPEbased neutralization assay sees the co-incubation of the antibody with a SARS-CoV-2 viral solution containing 100 median Tissue Culture Infectious Dose (100 TCID 50 ) of virus for 1 hour at 37°C, 5% CO 2 . The mixture was then added to the wells of a 96-well plate containing a subconfluent Vero E6 (ATCC, Cat#CRL-1586) cell monolayer. Plates were incubated for 3-4 days at 37°C in a humidified environment with 5% CO 2 , then examined for CPE by means of an inverted optical microscope by two independent operators. All nAbs were tested at a starting dilution of 1:5 and the IC 100 was evaluated based on their initial concentration, while plasma samples were tested starting from a 1:10 dilution. Both nAbs and plasma samples were then diluted in step 1:2. Technical duplicates were performed for both nAbs and plasma samples. In each plate, positive and negative control were used as previously described 5 .

SARS-CoV-1 neutralization assay
For neutralization assay, HEK293TN-hACE2 cells were plated in white 96-well plates in a complete DMEM medium. 24 h later, cells were infected with 0.1 MOI of SARS-CoV-1 pseudoparticles that were previously incubated with serial dilution of purified or not purified (cell supernatant) mAb. In particular, a seven-point dose-response curve (plus PBS as untreated control), was obtained by diluting mAb or supernatant, respectively fivefold and threefold. Thereafter, nAbs of each dose-response curve point was added to the medium containing SARS-CoV-1 pseudoparticles adjusted to contain 0.1 MOI. After incubation for 1 h at 37°C, 50 µl of mAb/SARS-CoV-1 pseudoparticles mixture was added to each well, and plates were incubated for 24 h at 37°C. Each point was assayed in technical triplicates. After 24 h of incubation, cell infection was measured by luciferase assay using Bright-Glo™ Luciferase System (Promega) and Infinite F200 plate reader (Tecan) was used to read luminescence. Obtained relative light units (RLUs) were normalized to controls and dose-response curve were generated by nonlinear regression curve fitting with GraphPad Prism to calculate Neutralization Dose 50 (ND 50 ).

Functional repertoire analyses
nAbs VH and VL sequence reads were manually curated and retrieved using a CLC sequence viewer (Qiagen). Aberrant sequences were removed from the data set. Analyzed reads were saved in FASTA format and the repertoire analyses was performed using Cloanalyst (http:// www.bu.edu/computationalimmunology/research/software/) 30,31 .

Radar plot distribution of IGHV;IGHJ germlines
A radar plot was generated to display the distribution of IGHV;IGHJ germlines among the three nAbs groups: seronegative second dose, third dose, and seropositive second dose. Each star in the radar represents a particular IGHV;IGHJ germline combination, alphabetically sorted. Germline abundance is represented as a percentage of the total of each group. In addition, the log2 fold change for each combination of the three groups was calculated and represented as a concentric heatmap; the higher the fold change it is, the bigger is the first group with respect to the other, and vice versa. The figure was generated with R v4.1.1. and assembled with ggplot2 v3.3.5.

Network plot of clonally-expanded antibody families
To investigate the genetic similarity within and between lineages, a network map was built by representing each clonal family with a centroid and connecting centroids sharing a similar sequence. The centroid sequence was computed with Cloanalyst to represent the average CDRH3 sequence for each clonal family, and Hamming distance was calculated for each antibody CDRH3 sequence to represent the relationship within the clonal family. Levenshtein distance was calculated between each centroid representative of each clonal family to investigate the relationship between clonal families. Levenshtein distance was calculated with the R package stringdistm v0.9.8 (https://cran.rproject.org/web/packages/stringdist/index.html) and normalized between 0 and 1. A network graph was generated with the R package ggraph v2.0.5 (https://ggraph.data-imaginist.com/index.html) with Fruchterman-Reingold layout algorithm and the figure was assembled with ggplot2 v3.3.5. The size of the centroid is proportional to the number of antibodies belonging to the same clonal family, while the color of each node represents the antibody origin: light blue and dark blue for the seronegative second dose and seronegative third dose, respectively.

Statistics and reproducibility
Statistical analysis was assessed with GraphPad Prism Version 8.0.2 (GraphPad Software, Inc., San Diego, CA). Nonparametric Mann-Whitney t-test was used to evaluate the statistical significance between the groups analyzed in this study. Statistical significance was shown as * for values ≤0.05, ** for values ≤0.01, *** for values ≤0.001, and **** for values ≤0.0001. No statistical methods were used to predetermine the sample size of subjects and nAbs screened. All four subjects previously analyzed after two vaccine doses were re-enrolled after receiving the third dose. All isolated nAbs were analyzed in this study. No data were excluded from the analyses. The experiments were not randomized and investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
Source data are provided with this paper. All data supporting the findings in this study are available within the article or can be obtained from the corresponding author upon request. SARS-CoV-2 variant sequences were deposited and accessible from https://github.com/ dasch-lab/SARS-CoV-2_nAb_third_dose Source data are provided with this paper.