Single-cell Sequencing of Circulating Human Plasmablasts during Staphylococcus aureus Bacteremia

Staphylococcus aureus is a human opportunistic pathogen that frequently causes infections, including the life-threatening S. aureus bacteremia (SAB) (1–3). Due to the emergence of multidrug-resistant strains, effective antibiotic treatment of SAB is increasingly complicated. This leads to increasing health care costs (€1.5 billion per year in Europe), prolonged hospital stays, and sometimes death. Because it has been difficult to find new antibiotics that act on multidrug-resistant strains, there is a clear need for the development of alternative therapies.

Ab-based biologicals could provide an alternative approach to improve treatment of staphylococcal infections (4, 5). An Ab molecule is derived from one B cell clone and has a variable domain that can be changed to bind an almost infinite variety of molecules. Abs fulfill various protective roles such as the neutralization of bacterial toxins and virulence factors (6, 7). Alternatively, Abs can contribute to bacterial killing via the human immune system. To do so, an Ab first needs to recognize bacterial surface structures (e.g., proteins or conserved glycans) via its Fab domains (8). Subsequently, the Ab Fc domain binds the Fcγ receptors on phagocytes or interacts with proteins of the human complement system to trigger a sequence of reactions that results in more effective labeling of bacterial cells (8, 9). This labeling facilitates engulfment of bacteria by immune cells (phagocytosis) for intracellular killing (10, 11).

In recent years, several advanced methods have been developed to identify human B cells with a certain specificity that are used to generate Ag-specific mAbs (12–15). B cells recognize a specific Ag via the BCR, the membrane-bound form of an Ab. Ig-Fab regions are assembled from germline variable (V), diversity (D; only in Ig H chain), and joining (J) elements. Recombination of these regions creates an almost infinite Ab repertoire (16). When the BCR binds foreign Ag, B cells proliferate while somatic hypermutation (SHM) introduces mutations in the variable domain (16, 17). High-affinity B cell clones are selected for survival (affinity maturation). Additionally, B cells may switch their isotype (from IgM to IgG or IgA, class switching) (18). Activated B cells can develop into highly specific memory B cells that provide long-term memory or differentiate into plasmablasts/plasma cells, which no longer display the BCR on their surface but secrete soluble Abs. Although plasmablasts are short-lived, plasma cells migrate to the bone marrow to become long-lived cells responsible for specific Ab titers that persist for years postinfection or vaccination (19, 20).

Ab interrogation can be approached via analysis of memory B cells (13, 15) or plasmablasts (21). Because plasmablasts circulate in the blood during an active infection, this cell type provides a valuable readout for Abs formed during infection (22, 23). For S. aureus, studies have shown that plasmablast responses are less specific because the bacterium produces a superantigen (staphylococcal protein A [SpA]) that interferes with the B cell response (24). Also, because S. aureus exposes hundreds of Ags on its surface, the hunt for an optimal target that induces Ab-mediated immune activation is a complex task. To expand on our current knowledge of the plasmablast response during S. aureus infections (24, 25), we performed high-throughput single-cell RNA sequencing to analyze Ig receptor sequences of plasmablast during SAB. A selection of Abs was produced to assess functionality in vitro, focusing on Abs that could bind to the bacterial surface and elicit Fc effector functions.

Peripheral blood from patients with S. aureus bacteremia (n = 17) was obtained between days 8 and 86 after the first S. aureus–positive blood culture. Before blood draw, written informed consent in accordance with the Declaration of Helsinki from patients and approval from the Institutional Review Board of the University Medical Center Utrecht (METC Protocol 19/495) was obtained. Samples were received within 1 h of blood draw and directly processed. Serum tubes were spun down for 10 min at 3,000 rpm and 4°C, and serum samples were aliquoted and stored at −80°C. PBMCs were isolated using Ficoll and directly used for experiments and/or frozen and stored at −80°C. Healthy donor PBMCs (METC Protocol 07-125/C) were isolated and processed in the same way, except with monitoring of the 1-h time limit between blood draw and start of processing.

S. aureus clinical isolates from the six SAB patients were obtained from the clinical department of the University Medical Center Utrecht at the moment of initial bacteremia diagnosis. Additional isolates from clinical follow-up between diagnosis and blood draw for PBMC isolation were also included when available. For Patient 1, we used a total of three isolates; for Patients 2 and 3, we used a total of two isolates, and for Patients 4–6, we used one isolate. Isolates were received on blood agar plates, grown in lysogeny broth (LB), and stored in −80°C glycerol stocks for later use. Using multilocus sequence typing, the three isolates from Patient 1 were characterized as the same isolate. The same is the case for the two isolates from Patient 3. One of the isolates from Patient 2 and the isolate from Patient 4 were classified as Staphylococcus haemolyticus. For further experiments, clinical isolates of S. aureus (M5134, M5116, M5132, M5572, M5571, N0303, N0302, N0475, N0280, and N1050) and Staphylococcus epidermidis (M6138, N0620, N1385, N1451, N4590, N4843, N5065, and ATCC14990) were grown overnight in 3 ml of LB medium at 37°C with agitation. S. aureus Newman Δspa/sbi (mAmetrine-labeled) (11) and S. aureus Newman were grown overnight in 3 ml of Todd–Hewitt broth. The next day, the overnight cultures were subcultured in fresh LB medium or Todd–Hewitt broth and grown to a midexponential growth phase, corresponding to an OD of 0.5 at 600 nm (OD₆₀₀). For phagocytosis experiments, S. epidermidis clinical isolates N4843 and N0620 were incubated in 500 µg/ml FITC (Sigma-Aldrich) in PBS for 1 h at room temperature. After labeling, the bacteria were washed and resuspended in RPMI 1640 containing 0.05% human serum albumin (RPMI-H), and the concentration was determined on a MACSQuant (Miltenyi). The bacteria were aliquoted and stored at −20°C.

To assess spontaneous Ab production, PBMCs were cultured at a density of 50,000 cells/well in a round-bottomed 96-well plate in Iscove’s modified Dulbecco’s medium supplemented with penicillin/streptavidin/ultraglutamine and 10% heat-inactivated FCS with no additional stimuli. Supernatants were collected after 7 d.

ELISA was performed to determine total IgG concentration in supernatants. To this end, sheep anti-IgG (ICN Biomedicals; 2 µg/ml in PBS) was coated overnight at 4°C on 96-well ELISA plates (Thermo Fisher Scientific). The next day, the plates were washed with PBS with 0.05% Tween 20 (PBST), blocked for 1 h at 37°C with PBST containing 4% BSA (Serva). After washing, the plates were incubated for 1 h at 37°C with diluted supernatants (in PBST 1% BSA). As a reference, IgG (Jackson) was titrated in a 1:1 serial dilution starting at 500 ng/ml. Goat anti-human IgG HRP (Southern Biotech) was incubated for 1 h at 37°C to detect bound mAbs. Development was done using tetramethylbenzidine (TMB), and the reaction was stopped by adding 1 N H₂SO₄ after a few minutes. Absorbance was measured at 450 nm in an iMark microplate absorbance reader (Bio-Rad).

Freshly grown bacteria were washed (7 min at 3500 rpm, 4°C) and resuspended in RPMI-H (to OD₆₀₀ 0.5). Next, the bacteria were diluted 30-fold and incubated with PBMC culture supernatants, HEK cell supernatants, or a concentration range of the purified monoclonal Abs for 30 min at 4°C. Subsequently, binding of Abs was assessed by incubating with anti-human IgG Alexa Fluor 647 (Southern Biotech) for 30 min at 4°C. The bacteria were measured on a FACSVerse and analyzed in FlowJo (both BD Biosciences). The bacterial population was determined using forward and side scatter. The Alexa Fluor 647 geomean for the bacterial population was analyzed. To correct for differences in background binding of HEK cell supernatants between different clinical isolates, all of the values were corrected by dividing the geomean binding of a supernatant by the geomean of an empty control of the same isolate.

PBMCs were stained with B cell phenotyping Ab panels including CD3, CD14, CD19, CD20, CD27, CD38, IgD, IgM, IgA, and IgG. For 10× sequencing experiments (10X Genomics), PBMCs from each donor were labeled with a unique Ab-oligonucleotide conjugates (Totalseq-C, BioLegend). Immediately prior to measurement, the dead cells were stained using SYTOX AADvanced dye (diluted 1/40) (Thermo Fisher). The clones, fluorochromes, and suppliers are listed in Supplemental Table I. The samples were measured on FACSVerse or measured and/or sorted on ARIA II or ARIA III (all BD Biosciences).

The cells were sorted using two different strategies. In the first strategy, plasmablasts/cells and other B cells were sorted in two separate vials per donor. First, all the plasmablasts/cells were loaded on the 10× chip (10X Genomics) followed by the other B cells to fill up the chip to obtain a higher plasmablast/cells frequency. In the second strategy, plasmablasts/cells of cells from individual donors were sorted in separate tubes, which were pooled just before loading the plasmablasts/cells only on the 10× chip (10X Genomics).

The sorted cells were kept on ice and transported to Single Cell Discoveries (Utrecht, The Netherlands) for same-day processing and sequencing with the 10X Genomics 5′ kit. Libraries were generated and sent for sequencing (Novogene). The sequencing data were mapped against the human genome from 10× GRCh38 2020A with 10× CellRanger count version 6.0. Initial clustering and differential gene expression analysis were performed in Seurat (26).

The data were further analyzed for single-cell gene expression and immune profile using Loupe browser 4.2.0/6.1.0 and Vloupe browser 3.0.0/4.0.0 (both from 10X Genomics). The resulting data were used to generate lists with clonotypes that express a specific set of markers and oligotags (Totalseq-C, BioLegend) derived from the single-cell gene expression data. All clonotypes had to meet the following expression criteria: CD27⁺CD38⁺CD20⁻. Additional criteria were Blimp1⁺, CD27⁺, CD71⁺, and t-SNE cluster with majority of plasmablasts for the enriched plasmablast sort. For the plasmablast sort using oligotags, the requirement for the clonotypes was an Ab oligotag level of >7 log2.). Data regarding somatic hypermutation and CDR3 length were generated using the IMGT/HighV-Quest server (27). Clonotypes in the plasmablast (CD27⁺CD38⁺CD20⁻ mRNA) list that contained incomplete or excess sequences to form Ab (n = 46) were kept in the analysis.

Selection of sequences for mAb production from the enriched plasmablasts/cells sort was based on the following criteria: frequency > 1, mRNA profile being Blimp1⁺ (transcription factor plasma cells), and CD71⁺ (recently activated) or in the t-SNE of each donor located in the cluster containing most plasmablasts/cells. Selection of sequences from the Ab oligotagged plasmablasts/cells sort was based on the following criteria: frequency > 1, H chain (HC) and L chain (LC) gene usage combination occurring > 1 (selecting one of the sequences with the same HC LC usage), or CDR3 sequence of heavy or L chain occurring multiple times.

Human mAbs were produced as previously described (28, 29), with minor modifications. Briefly, eBlocks (Integrated DNA Technologies) were ordered containing the VH and VL sequences including flanking vector sequences for cloning (online dataset, Supplemental Table II, https://doi.org/10.5061/dryad.ncjsxkt54). The HC hG3 vector was analog to the pcDNA3.4 hG1 vector containing murine Ig signal peptide, and the AgeI and XhoI restriction sites were used for insertion of the VH sequences in form of the constant sequence of hIgG3 (GenBank accession number PQ310558). The κ vector was a pcDNA3.4 vector containing a HAVT20 signal peptide, and restriction sites XhoI and BsiWI were used for insertion of variable κ sequence in front of the κ constant sequence (GenBank accession number PQ310559). The λ vector was a pcDNA3.4 vector containing the murine Ig signal peptide, and the restriction sites NheI and XhoI were used for insertion of the variable λ sequence in front of the constant λ sequence (GenBank accession number PQ310560). Sequences were cloned into vectors using Gibson assembly (New England Biolabs) that were transformed to Escherichia coli Top10 cells. Plasmids were isolated to verify sequences, followed by transfection of Expi293F cells (Thermo Fisher Scientific) using PEI (polyethylenimine HCl MAX; Polysciences). For Ab production, two different protocols were used: Protocol S1 included 100 µl in 96-well flat-bottomed culture plates on a rotation platform (300 rotations/min) at 37°C, 8% CO₂, and Protocol S2 included 4 ml in 6-well plates (125 rotations/min) at 37°C, 8% CO₂. For both production strategies, 1 μg DNA/ml cells (ratio of HC and LC plasmids was 2:3) was added to OPTIMEM (1:10 of total volume; Life Technologies) and gently mixed. After adding PEI (1 μg/ml; ratio PEI to DNA is 5:1), the mixture was incubated at room temperature for 20 min. For Protocol S1, the cells were added to this mixture, whereas with Protocol S2, the mixture was added dropwise to the cells. At 4–6 d after transfection, IgG3 Abs were isolated by collecting supernatant. For protocol S2, IgG3 was isolated using a protein G HP multitrap (Cytiva). Subsequently, Abs were dialyzed in PBS overnight at 4°C and filter-sterilized through 0.22-μm Spin-X filters.

As positive controls, we recombinantly generated human IgG3 variants of previously described mAbs known to react with α-GlcNAc–wall teichoic acid (WTA) (4461), β-GlcNAc-WTA (4497) (WO/2014/193722 A1) (30, 31), peptidoglycan (PG) (M130, US20030228322A), lipoteichoic acid (LTA) (A120, WO-03059260-A3), poly-N-acetyl glucosamine (PNAG) (F598, US/2006/0115486 A1), and GlcNAc-containing proteins of the SDR family (anti-SDR or rF1, WO/2016/090040_Seq14) (32, 33). As a negative control, we used an IgG3 variant of an Ab recognizing DNP (anti-DNP) (34).

To test whether SpA can bind the VH3 of our mAbs, we used in-house produced SpA-B^KK (28). Briefly, 96-well ELISA plates (Thermo Fisher Scientific) were coated with SpA-B^KK (3 µg/ml in 0.1 M NaCO₃, pH 9.6) overnight at 4°C. The next day, the plates were washed with PBST and blocked for 1 h at 37°C with PBST containing 4% BSA (Serva). The plates were washed again and then incubated for 1 h at 37°C with 1 µg/ml mAb (in PBST 1% BSA). Goat anti-human IgG HRP (SouthernBiotech) was incubated for 1 h at 37°C to detect bound mAbs. Development was done using TMB, and the reaction was stopped by adding 1 N H₂SO₄ after a few minutes. Absorbance was measured at 450 nm in an iMark microplate absorbance reader (Bio-Rad).

IgG binding to PG, PNAG, and LTA was determined by ELISA. Maxisorp plates (VWR 735-0083) were coated with 2 μg/ml LTA (purified from S. aureus DSM 20233, a kind gift from Siegfried Morath and Thomas Hartung (University of Konstanz) (35); PG (purified from S. aureus Cowan EMS (36); 0.6 μg/ml PNAG (a kind gift from Gerald Pier, Harvard Medical School, Boston, MA) (37) in sodium carbonate buffer, pH 9.6, overnight at 4°C. The wells were blocked with 4% BSA in PBST for 1 h at room temperature with shaking. The wells were washed with PBST, followed by incubation with 1 µg/ml of different Abs for 1 h at 4°C with shaking. After washing with PBST, Ab binding was detected using HRP-labeled goat anti-human IgG (diluted 1:5,000, SouthernBiotech). The plates were washed with PBST and incubated with TMB substrate to detect binding. The reaction was stopped with H₂SO₄, and the OD₄₅₀ was measured.

IgG binding to WTA was determined using flow cytometry and synthetic WTA beads that were produced as previously described (38). In short, biotinylated RboP hexamers (a kind gift of Jeroen Codee, Leiden University) were incubated with recombinant glycosyltransferases TarS or TarM and UDP-GlcNAc. Glycosylated or empty RboP hexamers were coupled to Dynabeads M280 streptavidin (Thermo Fisher Scientific) and subsequently washed and stored at 4°C until use. To assess Ab binding to synthetic WTA beads, 5 × 10⁶ beads/ml were incubated with 1 µg/ml Ab in PBS, 0.1% human serum albumin, 0.05% Tween (PBS-HT) for 30 min at 4°C. Subsequently, the beads were washed in PBS-HT twice using a magnetic plate holder and resuspended in 3 µg/ml goat anti-human κ-AF488 (SouthernBiotech, 2060-30) for 30 min at 4°C. Afterwards the beads were washed twice again in PBS-HT and resuspended and diluted in PBS-HT to be analyzed with a MACSQuant VYB flow cytometer (Miltenyi Biotec).

The bacteria (7.5E5) were first incubated with the mAb (0–10 μg/ml) with or without 1% IgG/IgM-depleted pooled healthy donor serum in RPMI-H for 15 min at 37°C with shaking at 600 rpm. Next, freshly isolated neutrophils were added to the mAb-labeled bacteria (neutrophils:bacteria 1:10) and incubated for an additional 15 min at 37°C with shaking at 600 rpm. The reaction was stopped by fixing with 2% paraformaldehyde (in RPMI-H), and the neutrophils were measured on a FACSVerse. The data were analyzed using FlowJo.

To study whether the circulating plasmablasts derived from patients with SAB are a potential source to identify Abs against S. aureus, we set up an observational, cross-sectional study to recruit SAB patients. Specifically, we included 12 patients that were diagnosed with SAB by means of a positive blood culture. Because it is known that (Ag-specific) plasmablasts transiently appear in blood upon infection (23) and vaccination (21), venous blood was collected between 8 and 21 d after the first positive blood culture. Five additional SAB patients were included from whom blood was drawn >1 mo after the first positive blood culture. We expected that as the acute infection in these patients is diminished, the B cell frequencies returned to frequencies similar to healthy donors. Lastly, blood from six anonymous healthy donors was obtained as controls.

To determine whether SAB patients have an active and specific B cell response, we first characterized the percentage of total B cells (CD19^+/dimCD3⁻CD14⁻) in PBMCs and the distribution between naive B cells (CD20⁺CD27⁻), memory B cells (CD20⁺CD27⁺), and plasmablasts/cells (CD27⁺⁺CD38⁺⁺CD20⁻) in circulation. To this end, PBMCs were isolated and the frequency of plasmablasts (CD27⁺CD38⁺CD20⁻) within the total B cells (CD19^+/dim) population was determined using flow cytometry (gating strategy; Supplemental Fig. 1). As expected, the percentage of total B cells in PBMCs was not different between patients and healthy donors (Fig. 1A, left panel). However, in 4 of 12 patients, we observed a plasmablast/cell frequency greater than 1% of the total B cell population, which is higher than the frequency in healthy donors (Fig. 1A, right panel). In the >1-mo samples, one of the five donors had an increased plasmablast/cell frequency. No statistical differences between the three groups were found. Although the frequency of plasmablasts only increased in 4 of 12 patients in the day 8–21 samples, the lower plasmablast frequency in the >1-mo and healthy donor samples shows that we have the highest chance to identify S. aureus–specific plasmablasts between days 8 and 21. Furthermore, the plasmablast frequency can be used to determine which donors have the highest plasmablast count to retrieve the most plasmablast sequences.

To determine whether circulating plasmablasts from the four SAB patients resulted from an ongoing immune response against S. aureus, we assessed their capacity to spontaneously produce Abs and then analyzed the specificity of those Abs. Freshly isolated total PBMCs from the four SAB patients and three healthy donors were cultured in vitro without additional stimuli (patients indicated in color in Fig. 1A). Although memory and naive B cells require additional stimuli to proliferate and differentiate into Ab-secreting plasmablasts/cells (28, 29), plasmablasts can secrete Abs in culture in the absence of stimulation (39). Indeed, we observed that unstimulated PBMCs from the SAB patients secreted higher quantities of IgG than healthy donors (Fig. 1B). To test whether the plasmablasts produced surface-specific Abs, supernatants from cultures with PBMC from patients were incubated directly with S. aureus Newman Δspa/sbi (40). The S. aureus Newman Δspa/sbi strain lacks SpA and Sbi proteins, preventing nonspecific binding of Ig to the bacterial surface (41). Ab binding was detected using fluorescent anti-IgG Abs and flow cytometry. Forward and side scatter gating was used to identify the bacterial population (example plot in Fig. 1C) and to determine the level of IgG binding to bacteria. We found that patient plasmablast cultures contained IgGs against S. aureus Newman Δspa/sbi (Fig. 1D), indicating that the recently formed plasmablasts in the PBMC population are the result of an ongoing infection.

To further characterize circulating plasmablasts from SAB patients, we performed single-cell RNA sequencing (10X Genomics) to determine gene expression in combination with single-cell V(D)J sequencing. Using this approach, each individual B cell is barcoded to link all related mRNA levels and the Ab sequences. This makes it possible to identify an Ab sequence from a single cell with a specific gene expression profile. In the commercially provided 10X Genomics setup offered by Single Cell Discoveries (Utrecht, The Netherlands), a large excess of cells is needed to guarantee a certain number of cells in the data. Because the amount of plasmablasts we could obtain from the patients is limited, we used two modified experimental approaches to obtain information from as many plasmablasts as possible (schematic overview of experimental and analysis approaches in shown in Supplemental Fig. 2). In the first approach, Patients 1 and 6 were chosen because we expected the highest plasmablast number due to the available amount of PBMCs. B cells from both patients were isolated using FACS and subsequently loaded together on the 10× microfluidics chip. To enrich for plasmablasts on the 10× chip, two B cell populations were sorted. First plasmablasts (CD3⁻CD14⁻CD19^+/dimCD27⁺CD38⁺CD20⁻) and afterward the other B cells (CD3⁻CD14⁻CD19⁺) from the same patient were loaded. To identify Ag-experienced CD27⁺ B cells and plasmablasts in the datasets of both patients, we used mRNA levels of the plasmablasts CD markers used in sorting: CD27, CD38, and CD20. Total mRNA levels of all individual cells were used to generate a t-SNE plot that clustered those cells with similar mRNA expression profiles (Fig. 2A, left panel). Additionally, a plot of all cells with a gene expression of CD27⁺CD38⁺CD20⁻ plasmablasts (Fig. 2A, upper right panel, colored in purple) shows the plasmablasts are mostly located in one of the six clusters identified in the t-SNE plot (brown cluster). This cluster contains cells with the highest CD27 levels (Fig. 2A, lower right panel), which is expected for plasmablasts in comparison with other B cells. Furthermore, markers for B cell activation (CD71) (33), proliferation (Ki67) (42), and plasma cell differentiation (Blimp1) (43) were all present, indicating that the cells in this cluster are recently formed plasmablasts. Finally, clonotypes of all Ag-experienced B cells (CD27⁺) and the plasmablasts/cells (CD27⁺CD38⁺CD20⁻) were analyzed for their isotype. This analysis shows that IgM was the most dominant isotype of memory B cells, whereas most plasmablasts had a switched isotype (not shown). Together, this approach resulted in characterization of 88 CD27⁺CD38⁺CD20⁻ plasmablasts.

In the second approach, we further increased the number of plasmablasts in our analysis by specifically FACS sorting the plasmablast population. For this, all four patients with increased plasmablast frequency (Patients 1, 2, 3, and 6) and two additional patients (Patients 4 and 5) were selected. To preserve donor information, PBMCs were labeled with Abs that contained a different oligonucleotide tag for each patient (hashed sort). All CD3⁻CD14⁻CD19^+/dimCD27⁺CD38⁺CD20⁻ sorted plasmablasts were combined to perform 10× sequencing. Analysis of the data resulted in the sequences of an additional 341 CD27⁺CD38⁺CD20⁻ plasmablasts.

Combining the data of all CD27⁺CD38⁺CD20⁻ plasmablasts from both approaches, we made one dataset including the immune profile information of 429 plasmablasts (online dataset, Supplemental Table III, https://doi.org/10.5061/dryad.ncjsxkt54). V(D)J analyses revealed a high plasmablast diversity with 309 unique clonotypes (cells sharing the same V, D, and J genes and having the same CDR3 sequence in both heavy and L chain). In 46 of the 309 clonotypes, we could not identify one HC/LC pair. These clonotypes either consisted only of the HC or LC sequence, or contained additional sequences such as an extra HC, LC, or both. Even though we do not know whether one or multiple Abs were produced in these plasmablasts, we kept all clonotypes in our analysis. Subsequently, we analyzed the isotypes and found that SAB plasmablasts were mainly comprised of IgA (the majority being IgA1) and IgG isotypes, which is consistent with previous data (24) (Fig. 2B). Analysis of the VH family usage showed that VH3 was most abundant, although all VH families were found (not shown). A more detailed analysis of VH/VL gene usage showed that 172 of the clonotypes have a unique VH VL gene combination (online dataset, Supplemental Table IV, https://doi.org/10.5061/dryad.ncjsxkt54). The degree of SHM and CDR3 length in the H chain across the individual donors was similar, ranging between 0 and 51 (nucleotide) mutations (SHM) and 6–28 CDR3 (amino acid) length (Fig. 2C). Similar mutation frequencies were obtained for the κ and λ LC (not shown). Thus, plasmablasts circulating in S. aureus–infected patients are genetically diverse resulting in a large range in the number of mutations and CDR3 length.

To test whether the patient-derived Ab sequences are specific against S. aureus, we selected sequences for mAb production. To increase the chance of selecting specific Abs, we used several selection criteria (Supplemental Fig. 2, analysis 2). We created a new clonotype list using the sequencing data of all CD27⁺CD38⁺ cells, including CD20-negative and -positive cells, because the latter may be indicative for (early) plasmablasts that are in the process of losing CD20 but still contain CD20 mRNA. From all clonotypes in the new list, we selected the following groups: 1) all clonotypes that occurred more than once. For the clonotypes from approach one, we additionally used the gene expression data to identify: 2) plasmablasts that were recently activated (characterized by expression of CD71 [33] and Blimp-1 [35]) and 3) all (early) plasmablasts sequences in the t-SNE plot located in the plasmablast cluster of the datasets (for example, the brown cluster in Fig. 2A). For the selection of sequences from approach 2, we looked at CDR3 sequences and HC/LC usage to identify clonotypes that were similar. If two clonotypes with the same CDR3 amino acid sequence were found, at least one of the two sequences was selected. In the process of somatic hypermutations, CDR3 is the hotspot for mutations to occur, leading to a different clonotype while maintaining the same VH/VL combination. These combinations can therefore give a first hint if there are clonal lineages in our dataset. Interestingly, three VH/VL combinations: IGVH3-23/IGKV1-5, IGHV3-23/IGKV3-20, and IGHV1-2/IGKV3-20, occurred more than five times. Because these combinations could contain clonally related sequences, they are more interesting than sequences with VH/VL combinations that occur less frequently. Thus, we selected one representative sequence with this combination. Finally, we confirmed that we conducted a representative analysis that included sequences from all six patients, and if possible, at least 20 sequences per donor. In total, we selected 224 VH/VL sequences that are a representation of all the plasmablasts, patients, isotypes, etc., found in the sequencing data (online datasets, Supplemental Tables III and V, https://doi.org/10.5061/dryad.ncjsxkt54).

Based on the selected DNA sequences, full-length Abs were synthetized, cloned, and expressed as IgG3 in Expi293F cells. IgG3 was chosen because the Fc part of this IgG subclass cannot interact with surface-bound Ig-binding protein SpA (28) on S. aureus. In the first round of screening, IgG3 expression was performed in 96-well plates. To determine binding, supernatants containing the synthesized mAbs were incubated with the clinical S. aureus isolates from the six SAB patients, irrespective of sequence origin. Example histograms of a buffer control, negative, and positive supernatant can be found in Fig. 3A (left panel). For our initial screen, we considered an arbitrary cutoff of corrected binding of >2 as a positive binding event. Our analysis showed that 47 of 224 (21%) supernatants contained Abs that bind to at least one of the clinical isolates (Fig. 3A, right panel).

Although generation of Abs as IgG3 circumvents nonspecific binding by SpA to the Ig-Fc domain, SpA can also interact with VH regions encoded by germlines of the VH3 family of certain Ig-Fabs (36). This means that the 47 identified candidates could still be bound by SpA. As described previously (24), comparing Ab binding between SpA-negative and -positive S. aureus strains can be used to determine IgG binding by SpA. Therefore, we tested binding of IgG3 supernatants to a S. aureus strain deficient in surface SpA and Sbi (Newman Δspa/sbi) and the wild-type Newman strain. This revealed three different categories: 1) six Abs that bind both the wild-type and Δspa/sbi strain (SpA/Sbi independent: Fig. 3B, green dots); 2) 26 Abs that bind to wild-type Newman but not the Δspa/sbi strain (SpA/Sbi dependent: lower right quadrant, Fig. 3B); and 3) 15 Abs that neither bind Newman wild-type nor Δspa/sbi strains (specific for one or multiple clinical isolates; lower left quadrant, Fig. 3B).

Because the Newman Δspa/sbi strain provides a tool to test the potency of our Abs without interference of SpA or Sbi, the Abs binding both Newman wild-type and Δspa/sbi (category a) were expressed in larger volume to allow further characterization. Of these, four mAbs were successfully expressed and bound back to the bacterial surface as purified IgG3. We wondered whether there were commonalities between these four Abs (online supplemental Table VI, https://doi.org/10.5061/dryad.ncjsxkt54), so we compared their sequences to the entire dataset of plasmablasts from our six SAB patients. Notably, all four Abs (BC019, BC020, BC021, and BC153) were derived from IgG2 plasmablasts. All four mAbs had undergone SHM, and the number of mutations and the CDR3 length are near the median of the entire dataset. Interestingly, BC019, BC020, and BC021 are clonally related and likely the result of ongoing affinity maturation (Fig. 3C).

Next, we determined whether the four purified mAbs could bind to the different clinical isolates from the SAB patients, and we confirmed their binding to S. aureus Newman Δspa/sbi (Fig. 3D). When comparing the binding capacities of the different Abs, we observed that BC019 and BC153 strongly bound to bacteria isolated from three and four patients, respectively (greater than 10 times above background). Interestingly, although BC019, BC020, and BC021 are clonally related, BC020 binds 2- to 6-fold less compared with BC019. Because SpA can interact with VH regions of certain VH3 families (36) and all new S. aureus mAbs have a VH3 sequence that could potentially be recognized (Supplemental Table VI, https://doi.org/10.5061/dryad.ncjsxkt54), we investigated whether our Abs can be bound by SpA. Previously it has been shown that the single SpA-B domain with mutations to abrogate IgG-Fc binding (SpA-B^KK; Q9K and Q10K) can be used to study the interaction between SpA and the Ab VH3 region (28). Therefore, we used recombinantly produced SpA-B^KK to determine the binding to our Abs in an ELISA. Fig. 3E shows that SpA-B^KK can bind BC019 and BC021 but not BC020. This demonstrates that the binding values in Fig. 3D for BC019 and BC021 are not only caused by the Ab binding to the clinical isolate but also partly by surface-bound SpA recognition of VH3 part of the Ab. Thus, the binding values in Fig. 3D are likely an overestimation of these Abs’ potency. In contrast, BC020 and BC153 are interesting S. aureus recognizing Abs because of the lack of SpA binding to the VH3. In conclusion, from 224 sequences, we identified four novel mAbs that can bind the surface of S. aureus.

Next, we compared the binding of our newly identified Abs with two previously published S. aureus Abs: 4497 targeting β-GlcNAc-WTA (anti–β-WTA) (WO/2014/193722 A1) (30, 31) and rF1 targeting GlcNAc-containing proteins of the SDR family (anti-SDR) (32). As a negative control, we used an Ab recognizing DNP (anti-DNP) (34). All three Abs were produced locally as IgG3. After titrating our Abs, we observed a dose-dependent binding to Newman Δspa/sbi (Fig. 4A), consistent with our previous experiments. Although BC153 showed a comparable binding to anti–β-WTA and anti-SDR Ab, the other mAbs showed a lower binding.

Previously, we have shown that several mAbs recognizing the surface of S. aureus, including anti-SDR, could also cross-react with S. epidermidis (33). Thus, we wondered whether our S. aureus–specific Abs could also recognize S. epidermidis. We tested binding of the four mAbs to eight different S. epidermidis clinical isolates. Three mAbs (BC019, BC020, and BC021) bound to all S. epidermidis clinical isolates in our panel (Fig. 4B). In contrast, BC0153, which performed best on S. aureus, did not bind any of the tested S. epidermidis strains. As a control, we also measured binding of anti-SDR (33) and observed binding to all eight isolates. However, BC019, BC020, and BC021 showed on average a higher and more consistent binding to all eight strains tested than anti-SDR (Fig. 4C). After titrating the Abs, we observed a dose-dependent binding of BC019, BC020, and BC021 to S. epidermidis clinical isolate N484 that was stronger than anti-SDR. Thus, we have shown that three of our S. aureus–specific mAbs can also cross-react with S. epidermidis.

To explain the binding data, we set out to identify the surface targets of our Abs. Because all Abs were derived from IgG2 plasmablasts, we suspected they would recognize surface carbohydrates. To test this, we analyzed Ab reactivity with key surface carbohydrates of S. aureus and S. epidermidis. First, we used ELISA to assess binding to purified polysaccharides that are shared between S. aureus and S. epidermidis. No binding was detected to purified PG and PNAG (Fig. 5A, 5B), whereas IgG3 variants of positive control Abs M130 (recognizing PG) and F598 (recognizing PNAG) showed strong binding to PG and PNAG, respectively. However, we observed strong binding of BC019, BC020, and BC021 to LTA that was purified from S. aureus (35). Binding of BC019, BC020, and BC021 was more potent than the control LTA Ab (A120) (Fig. 5C). No binding to LTA was observed for BC153 (Fig. 5C). Because LTA is present in the cell wall of all Gram-positive bacteria, these data corroborate with our finding that BC019, BC020, and BC021 can bind to both S. aureus and S. epidermidis. Because BC153 was specific for S. aureus, we suspected WTA, a major component of S. aureus cell wall that is not found in S. epidermidis, as the target. To test this, we assessed binding to synthetic WTA beads (38) hat contain either the ribitol phosphate backbone (RboP) alone, or the RboP backbone modified with TarS (to obtain β-GlcNAc-WTA) or TarM (to obtain α-GlcNAc-WTA). This revealed that BC153 specifically bound β-GlcNAc–modified WTA, but not the RboP backbone or α-GlcNAc-WTA (Fig. 5D). Binding of BC153 to β-GlcNAc–modified WTA was similar to the Ab targeting β-WTA (4497). In all, we show that BC019, BC020, and BC021 recognize LTA, whereas BC153 binds to β-GlcNAc–modified WTA.

Finally, we evaluated whether the Abs induce phagocytosis of bacteria by human neutrophils. We first studied the capacity to directly engage Fcγ receptors in the absence of human complement by incubating S. aureus with different concentrations of mAb and freshly isolated human neutrophils. In this set up, we found that only BC153 (the strongest S. aureus binder) stimulated FcγR-mediated phagocytosis of S. aureus Newman Δspa/sbi (Fig. 6A). However, the concentration needed for phagocytosis was higher than the concentration of previously identified Abs anti-WTA and anti-SDR. Next, we added 1% normal human serum depleted of IgG and IgM (ΔNHS) as a complement source to study the effect of mAbs in presence of complement. Also in this setting, only BC153 contributed to phagocytosis of S. aureus (Fig. 6B). Additionally, we performed both phagocytosis experiments on two clinical isolates from patient 2 (M5572) and patient 5 (N0280). These isolates were selected as they showed highest binding (Fig. 3D). Interestingly, all mAbs induce FcγR-mediated phagocytosis on these clinical isolates (Fig. 6C, Supplemental Fig. 3A), and the addition of complement did not greatly enhance phagocytosis (Fig. 6D).

Our next step was to determine whether the Abs trigger phagocytosis of the S. epidermidis strain using the same assays. In absence of complement, the anti-SDR control mAb was able to stimulate phagocytosis. Interestingly, the three newly identified Abs could all induce phagocytosis of S. epidermidis, in both the absence (Fig. 6E) and the presence (Fig. 6F) of complement. This suggest that the identified Abs activate phagocytosis mainly via the Fcγ receptor. Comparable binding and phagocytosis results were obtained using another S. epidermidis clinical isolate (Supplemental Fig. 3B, 3C). Overall, we identified four Abs directed against S. aureus that stimulate phagocytosis, of which three show similar results in the phagocytosis of S. epidermidis.

In this study, we employed high-throughput single-cell RNA sequencing to analyze Ab repertoires of plasmablasts during SAB. From the resulting >300 unique VH/VL sequences, we identified four novel Abs that could bind and engage Fc effector functions on S. aureus. Three of the four Abs show cross-reaction with S. epidermidis.

Since 2021, over 100 mAbs have been approved by the U.S. Food and Drug Administration (44), of which the majority is used for the treatment of cancer. Although the medical need is high, there are no Food and Drug Administration–approved mAbs against S. aureus. Abs targeting the surface of S. aureus and subsequently activating the immune system are considered interesting for the treatment of multidrug-resistant S. aureus. In addition, Ab therapies are also proposed as adjunctive therapy (next to antibiotics) to improve bacterial clearance of antibiotic-sensitive S. aureus in patients that have a compromised immune system hampering natural Ab production (intensive care patients, patients receiving chemotherapy and/or immunomodulatory agents).

All four Abs identified in this study effectively activated Fc-mediated phagocytosis of S. aureus by human neutrophils, with the BC153 Ab (targeting WTA) being the most potent. This supports previous data that WTA-targeting Abs are potent drivers of immune activation (31, 38, 45). This is likely because WTA is a high-density Ag that is not hidden by other surface structures (46). Although we previously observed for other mAbs targeting Gram-positive bacteria that activation of the complement system greatly enhances phagocytosis (29), this was not observed for BC153-IgG3. We think this is because BC153-IgG3 alone strongly stimulates bacterial uptake via FcγR. Because S. aureus has evolved mechanisms to escape neutrophil killing (47, 48), the activity of BC153 could be further enhanced by conjugation of drugs that enhance intracellular killing of S. aureus inside neutrophils (31). In addition, engineering of the IgG-Fc tail can be used to improve both the half-life and/or opsonophagocytic killing of anti-staphylococcal IgGs (33, 45, 49, 50). Whether a single Ab would be sufficient to clear S. aureus in vivo can be debated. Especially because S. aureus has evolved many mechanisms to evade and attack the immune response via the secretion of several immune evasion proteins and toxins (47), it is anticipated that immune therapy against this versatile pathogen is most effective when immune-activating Abs are administered in combination with neutralizing Abs that target evasion molecules.

Although less virulent than S. aureus, S. epidermidis is also considered a significant problem in the clinic. Preterm neonates and elderly that have undergone surgery (e.g., joint replacement) are especially vulnerable to the development of S. epidermidis–induced sepsis (51, 52). Together, S. aureus and S. epidermidis account for most of the hospital-acquired sepsis and foreign body–related infections (53, 54). During infections, both species are often found (55), and the exact bacterial strain is not known when a physician starts treatment. Therefore, Abs that recognize both species such as the LTA-targeting BC019, BC020, and BC021 would be beneficial for therapeutic purposes.

An important complication in Ab therapy of staphylococci is that S. epidermidis and S. aureus are both notorious for their ability to form biofilms on implanted devices (e.g., heart valves, catheters, prosthetic joints) and host tissues (e.g., chronic wounds, endocarditis, osteomyelitis) (56). Although biofilm formation could prevent Ab binding, we have previously shown for Abs targeting WTA that Abs recognizing planktonic cells can also bind to biofilm (40). Potentially, the herein described Abs could also serve as vehicles to carry biofilm-degrading agents to the site of infection (57).

Next to BC019, BC020, BC021, and BC153, the other >300 unique VH/VL sequences identified in this study provide potential leads for S. aureus–specific Abs. Because we specifically searched for Abs that could bind to the bacterial surface, it remains to be determined whether our dataset includes Abs recognizing virulence factors secreted by S. aureus.

This paper is not the first to sequence Ab repertoires from circulating plasmablasts during S. aureus infections (24, 25) Within our cohort, several SAB patients had increased plasmablast levels, supporting an active B cell response. Still, the frequency was lower compared with several viral infections (22, 23, 58, 59), suggesting that either SAB patients have a limited capacity to produce S. aureus–specific B cells or that the timing of the blood draw was not optimal to capture the transient plasmablast peak (22). For six patients, we performed single-cell RNA sequencing using 10X Genomics to retrieve VH/VL sequences and corresponding gene expression from >400 plasmablasts. An important difference with previously described approaches (24, 25) is that the 10X Genomics platform does not require single-cell sorting, so the plasmablasts can be sorted in bulk. In this way, it is possible to sequence all plasmablasts from a single SAB patient and provide information on Ab isotype, sequences, and activation markers of each cell. A drawback of the bulk sequencing approach is the loss of material in the different preparation steps, especially when the number of plasmablasts per patient is limited. Therefore, the limited total number of plasmablasts and our selection criteria only provides a snapshot of Ab sequences present during infection. Most plasmablasts in our dataset were recently formed and had a switched isotype. As in a previous study (24), most plasmablast sequences were IgA, the isotype associated with mucosal immune response (60). Furthermore, 9% of the plasmablasts were IgG2, the isotype associated with Abs binding bacterial sugars (61). For all the IgA-derived sequences, it would be interesting to assess their functionality in the context of recombinant dimeric IgA instead of IgG3. Finally, it is interesting that most of our Abs were bound by SpA. Even though our mAbs were expressed with IgG3 Fc regions, we observed that 26 of the 47 mAbs were bound by SpA. This is in line with previous work by Pauli et al. (24), who showed that 48% of activated plasmablasts bind to SpA, and suggests that SpA-induced nonspecific expansion of plasmablasts also occurs during SAB. In conclusion, our study confirms that SAB patients have an ongoing immune response against S. aureus and that circulating plasmablasts are a potential source to identify novel mAbs against S. aureus and S. epidermidis.

R.J.M., J.S., F.J.B., and A.G. are (former) employees at Genmab BV and have ownership interests (including stocks, warrants, patents, etc.). The other authors have no financial conflicts of interest.

We thank the research nurses Henny Ophorst, Karin Brakke, and Marjoleine van Opdorp for the inclusion of all patients into the study. We acknowledge the Utrecht Sequencing Facility (USEQ) for providing sequencing service and data. The Utrecht Sequencing Facility is subsidized by the University Medical Center Utrecht and the Netherlands X-omics Initiative (Netherlands Organization for Scientific Research Project 184.034.019). We thank Jeroen Codee and Astrid Hendriks for providing synthetic WTA beads, and Kulsum Dawoodbhoy and Marije van’t Wout, for critically reading the manuscript.