Extrafollicular B cell responses correlate with neutralizing antibodies and morbidity in COVID-19

High-dimensional B cell phenotyping of acute COVID-19.

The identification of new human B cell subsets^18,21,22 and their integration in a coherent classification²³ enable the application of high-dimensional flow cytometry (FCM) to accurately identify B cell profiles while providing functional context to understand infection response course. To this end, we studied 17 patients with confirmed COVID-19 infection using 24 marker FCM panels (plus non-B cell dump and viability staining) that accurately identify B cell populations, assess activation status and indicate homing potential through integrin and chemokine receptor modulation (Supplementary Table 1). Of the COVID-19 cohort, 10 patients were critically ill and required intensive care unit (ICU) admission (severe COVID-19; ICU-C), and 4 of the 10 patients died (Supplementary Table 2). This group was compared to 7 outpatients with milder illness (outpatient COVID; OUT-C) and 17 healthy donors (HD).

Despite reports of lymphocytopenia²⁴, patients with COVID-19 displayed elevated numbers of peripheral blood mononuclear cells (PBMCs), with CD19⁺ B cells significantly increased relative to HDs (Extended Data Fig. 1). B cell profiling with high-dimensional FCM identified primary cell populations (transitional (Tr), naive (N), DN, memory (M) and ASCs), which could be then fractionated into 14 nonredundant, secondary B cell populations of established significance (Fig. 1a–f and Table 1)²³. The three clinical groups displayed distinct B cell profiles characterized by either expansions of ASCs and DN2 cells (ICU-C) or transitional cells (OUT-C; Fig. 1a–f).

To aid in exploration of the dataset, a composite sample was created by representative downsampling (1,000 cells per patient) of FCM results obtained from HD (n = 12), OUT-C (n = 7) and ICU-C (n = 10) patients using an identical staining panel (Methods; FCM panel V2), and results were recombined for combined analysis. A uniform manifold approximation and projection (UMAP) algorithm for dimensionality reduction²⁵ was applied to the composite sample for all assessed fluorescence parameters, and cells from the composite sample were mapped in Cartesian space (Fig. 2a). Overlaying a 90% equal probability contour from each cohort revealed clear phenotypic B cell separation in the three clinical groups (Fig. 2b). Three major cell clusters are visually apparent in the projection (Fig. 2a–c). Cluster 1 is highly diverse, predominantly composed of transitional, naive and IgM⁺ memory B cell subsets (Fig. 2c,d). Cluster 2 is more homogeneous, consisting of DN populations 1–3 and the switched memory (sM) compartment (Fig. 2c,d). Importantly, in keeping with our previous identification of close transcriptional similarity between DN1 and sM cells¹⁸, there is a clear overlay between these populations, while DN2 B cells formed a distinct cluster within the larger DN density (Fig. 2c,d). Of note, these projections indicated a split in the DN1 and DN3 compartments seemingly driven by expression of human leukocyte antigen (HLA)-DR and CD19—an indication of heterogeneity within these populations that requires further interrogation (Fig. 2e). ASC populations split by the CD138 expression pattern defined cluster 3 (Fig. 2c,d). To aid in visual exploration of the dataset, selected marker expression mapped onto these UMAP projections are provided in Fig. 2e. Together, these data provide unique insight into the stark transitions in the B cell compartment following severe SARS-CoV-2 infection.

Increased DN2 B cells and ASCs correlate with severe COVID-19.

By overlaying disease states onto the composite sample and subtracting overlapping densities, differential cellular signatures became visually apparent (Fig. 3a), with three main regions of interest (ROIs) distinguishing ICU-C and OUT-C patients, both from each other and from HD. Region 1 highlights an area of high CD11c expression by the ICU-C cohort demarcating aN and DN2 populations (Fig. 3b), which were significantly enriched in ICU-C patients (Fig. 3c,d). In accordance with previous studies¹⁸, both aN and DN2 cells expressed the highest levels of CD11c and the IFN-γ-inducible transcription factor T-bet (Fig. 3e,f). Region 2 comprises ASCs, a population whose expansion was readily observed following vaccination and other instances of acute infection^26,27. In COVID-19 infection, robust ASC formation correlated with negative disease outcomes, with ICU-C patients presenting with significantly higher frequencies of ASCs than other groups (Fig. 3g). Notably, ASC responses in ICU-C patients were highly enriched in more mature CD138⁺ cells relative to both OUT-C and HDs—a response feature previously observed in SLE¹⁷ (Fig. 3h,i). Finally, region 3 mapped neatly to the CD21^lo Tr subset (Fig. 3a), which was significantly increased in OUT-C patients. Overall, transitional cells constituted more than 10% of the total B cell compartment in OUT-C patients, with CD21^lo cells making up more than half of the overall frequency gain (Fig. 3j,k). CD21^lo Tr cells expressed higher levels of markers of the early transitional compartment (IgM, CD24, CD10), but less CD21 and CXCR5 (Fig. 3l), than CD21hi cells. Of note, in comparison with similarly identified cells from the HD cohort, CD21^lo Tr cells from OUT-C patients exhibited higher expression of CD138, a marker whose expression is typically considered to be restricted to ASCs (Fig. 3l). Altogether, these data suggest a robust activation of effector B cells in severe COVID-19, unmatched by the responses observed in more mild disease courses.

Severe COVID-19 drives SLE-like extrafollicular responses.

The use of secondary population frequencies as features for the hierarchical clustering of patients with COVID-19 demonstrated a near-perfect separation between the ICU-C and OUT-C cohorts, driven primarily by a coordinated increase in the more severe group of both aN and DN2 cell populations (Fig. 4a). A novel DN population (DN3), previously unreported in other conditions and defined by the absence of both CD21 and CD11c, consistently fell alongside aN and DN2 populations (Fig. 4a). Their association with these EF constituents and similarly reduced expression of CD21 suggests relevance to this effector pathway (Fig. 4a and Extended Data Fig. 2a,b). Increased frequency of all three activated CD21⁻ populations was highly correlated with the ASC expansion described above (Fig. 4a), forming a population cluster that defined the responses within the critically ill patients. EF pathway activation was not a reflection of infection duration as both EF-high (CoV-A) and EF-low (CoV-B) clusters were sampled at similar times after symptom onset (Fig. 4b).

We have previously observed that EF pathway activation, particularly the increased frequency of DN2 B cells, was accompanied by concordant reductions in the DN1 compartment^17,18,20. Direct comparison of the CoV-A cluster to patients with active SLE revealed highly similar DN features (Fig. 4c) with strong skewing toward the ASC-associated DN2 subset in both groups (Fig. 4d,e). Thus, DN2:DN1 ratios, an important reflection of EF to follicular response dynamics, were significantly higher in both CoV-A and active SLE groups (Fig. 4f). This phenotype was associated with a contraction of unswitched memory cells, a feature consistently observed in SLE and other autoimmune diseases^17,28,29 (Fig. 4g). In keeping with an EF nature, chemokine receptor analysis in patients with COVID-19 showed a decrease in follicular homing predisposition through CXCR5 and a corresponding increase in CXCR3—a mediator of homing to tissue undergoing IFN-γ-driven inflammation (Fig. 4h).

An important consideration in this study cohort was the heavy skewing of the ICU-C cohort toward African American (AA) patients, while HD and OUT-C groups included more patients of European ancestry. As AA patients have been previously shown to have higher baseline B cell responsiveness, and EF responses specifically¹⁸, it was important to understand if this activation signature was attributable to baseline demographic differences. Hence, we evaluated a retrospective cohort of AA HDs (n = 24), by FCM, to assess ASC response, DN2:DN1 ratio and usM reduction, relative to HD, OUT-C and ICU-C groups (Extended Data Fig. 3). In all three metrics, AA HDs were slightly skewed from our current HD cohort, but significantly different from the ICU-C cohort, giving confidence that the responses identified in the ICU-C cohort were not due to changes in baseline B cell composition across demographics. Accordingly, these data confirm that patients with severe COVID-19 display many of the hallmarks of dominant EF responses previously reported in SLE (Extended Data Fig. 4).

Germline clonotypes dominate the COVID-19 ASC repertoire.

To understand the nature of the repertoire in severe COVID-19, ASCs from a critically ill patient from the CoV-A cluster (Fig. 4a) was analyzed by single-cell V(D)J repertoire sequencing. Isotype analysis revealed that ASCs utilized a balanced IgM, IgG1 and IgA1 signature (Fig. 5a). Moreover, a large fraction of all the clonotypes identified (>3% of all clonotypes and 60% of the top 15 clonotypes) displayed contemporaneous connections between IgM and IgG1 or IgA1, demonstrating ongoing isotype switching (Fig. 5b). Also consistent with ongoing maturation and antigen selection, the ASC V(D)J repertoire was characterized by the presence of oligoclonal expansions, with the top ten clonotypes making up more than 12% of the overall repertoire (n = 2,017 clonotypes; Fig. 5c). Bulk V(D)J sequencing of CD138-enriched ASCs from two additional ICU-C patients demonstrated similar oligoclonality, but with more pronounced clonotype expansions, individually contributing 1–8% of the total repertoire (Fig. 5c). This observation was similar to previous studies characterizing ASCs in both flaring SLE and in response to influenza or tetanus vaccination¹⁷. In addition, single-cell sequencing identified individual multimember clonal lineages, including the two largest clonotypes, with complex branching patterns, class switching and broad ranges of somatic hypermutation (SHM) indicative of robust antigen selection (Fig. 5c,d). Nevertheless, a majority of the clonotypes identified in this patient had remarkably low mutation frequencies, especially clonotypes making up the IgA1 and IgG1 compartments (Fig. 5e). Indeed, more than half of the clonotypes obtained expressed germline V_H genes (Fig. 5f). This pattern is consistent with the presence of newly recruited EF clones, as we previously reported in SLE¹⁷.

Another hallmark of the SLE repertoire is defective tolerance resulting in increased frequencies of disease-specific IgHV4–34 B cells and VH4–34-encoded 9G4-idiotype autoantibodies. These antibodies display intrinsic autoreactivity against a variety of self-antigens (nucleic acids and blood-group antigens among others), which is mediated by a hydrophobic patch encoded in the germline framework 1 (FR1)³⁰. In healthy individuals, elimination of the FR1 patch through somatic mutation allows the expression of non-autoreactive, protective IgHV4–34 antibodies (clonal redemption)³¹. In contrast, the FR1 patch was present in 85% of all VH4–34-expressing clonotypes from the single-cell sequencing, including expanding lineages with evidence of SHM (Fig. 5g). In keeping with the sequencing data, ICU-C patients expressed significantly higher concentrations of serum 9G4 antibodies whose detection relies on the preservation of the VH4–34 FR1 patch (Fig. 5h). These data suggest that, in addition to effector pathway activation, perturbations in the ASC repertoire are highly consistent with those previously characterized in EF-driven responses.

Robust extrafollicular responses correlate with neutralizing antibody titers.

EF response characterization has identified the pathway as peripherally focused, inflammatory, and highly associated with IL-6 and IP-10 (refs.^19,20,32). Both of these factors have now been identified as biomarkers of poor prognosis in the COVID-19 literature¹ and could be identified at increased expression in the CoV-A cluster (Fig. 6a,b). In keeping with those studies, four of the five patients with the highest concentrations of IL-6 did not survive (Fig. 6b). C-reactive protein (CRP), an acute phase reactant in the IL-6 pathway commonly used as a biomarker of COVID-19 severity, was also increased in the ICU-C cohort (Fig. 6c) and correlated with IL-6 and IP-10 concentrations in patients with COVID-19 (Fig. 6d,e). Importantly, the DN2 magnitude within the DN compartment was also significantly associated with CRP concentrations, indicating that the extent of EF activation may serve as a correlate of disease severity (Fig. 6f).

While mouse EF responses can effectively undergo antigen-driven affinity maturation¹³, and human SLE EF responses are enriched for autoreactivity even in clones with low frequency of SHM¹⁷, we wondered whether such responses would be associated with low titers of SARS-CoV-2-specific serum antibodies in COVID-19 infection. Instead, serum antibody responses against the receptor-binding domain (RBD) of the SARS-CoV-2 spike protein were higher in the more severe group (Fig. 6g). Consistent with the B cell antigen receptor-sequencing data, SARS-CoV-2-specific responsiveness was broad, with participation of IgM, IgA and IgG antibodies (Fig. 6g). Importantly, and in accordance with previous findings⁶, ICU-C responses were higher than outpatient responses and occurred very early in the infection course with significantly elevated titers by day 5 after symptom onset (Fig. 6h). To understand if these elevated titers were capable of neutralization, selected serums were tested using an in vitro neutralization assay. Concordant with anti-RBD activity, CoV-A cohort serums were consistently higher in neutralization capacity than CoV-B or HD serums, thereby confirming antibody functionality (Fig. 6i). In total, these data suggest that high concentrations of neutralizing antibodies are correlated with robust EF response activation but are insufficient to resolve the disease course in severe COVID-19.

Human participants.

All research was approved by the Emory University Institutional Review Board (Emory IRB nos. IRB00058507, IRB00057983 and IRB00058271) and was performed in accordance with all relevant guidelines and regulations. Written informed consent was obtained from all participants or, if they were unable to provide informed consent, from designated healthcare surrogates. Healthy individuals (n = 36) were recruited using promotional materials approved by the Emory University Institutional Review Board. Individuals with COVID-19 (n = 19) were recruited from Emory University Hospital, Emory University Hospital Midtown and Emory St. Joseph’s Hospital (all Atlanta, USA). All non-healthy individuals were diagnosed with COVID-19 by PCR amplification of SARS-CoV-2 viral RNA obtained from nasopharyngeal or oropharyngeal swabs. Individuals with COVID-19 were included in the study if they were between 18 to 80 years of age, were not immunocompromised and had not been given oral or intravenous corticosteroids within the preceding 14 d. Peripheral blood was collected in either heparin sodium tubes (PBMCs) or serum tubes (serum; both BD Diagnostic Systems). Baseline individual demographics are included in Supplementary Table 2. Study data were collected and managed using REDCap electronic data capture tools hosted at Emory University⁶⁰.

Peripheral blood mononuclear cell isolation and plasma collection.

Peripheral blood samples were collected in heparin sodium tubes and processed within 6 h of collection. PBMCs were isolated by density gradient centrifugation at 1,000g for 10 min. Aliquots from the plasma layer were collected and stored at −80 °C until use. PBMCs were washed twice with RPMI at 500g for 5 min. Viability was assessed using trypan blue exclusion, and live cells were counted using an automated hemocytometer.

Flow cytometry.

Isolated PBMCs (2 × 10⁶) were centrifuged and resuspended in 75 μl FACS buffer (PBS + 2% FBS) and 5 μl Fc receptor block (BioLegend, no. 422302) for 5 min at room temperature. For samples stained with anti-IgG, it was observed that Fc block inappropriately interfered with staining, so a preincubation step of the anti-IgG alone for 5 min at 22 °C was added before the addition of the block. Next, 25 μl of antibody cocktail (Supplementary Table 3) was added (100 μl staining reaction), and samples were incubated for 20 min at 4 °C. Cells were washed in PBS, and resuspended in a PBS dilution of Zombie NIR fixable viability dye (BioLegend, no. 423106). Cells were washed and fixed at 0.8% paraformaldehyde (PFA) for 10 min at 22 °C in the dark before a final wash and resuspension for analysis.

For intracellular staining, the 0.8% PFA fixation step was omitted, and an additional wash was added following viability staining. Fixation and permeabilization of the cells were carried out per the manufacturer’s instructions using the True-Nuclear Transcription Factor Buffer Set (BioLegend, no. 424401). Following permeabilization, cells were stained with antibodies for intracellular cytokine staining (ICS) and diluted in permeabilization buffer (Supplementary Table 3).

Cells were analyzed on a Cytek Aurora flow cytometer using Cytek SpectroFlo software. Up to 3 × 10⁶ cells were analyzed using FlowJo v10 (Treestar) software with DownSample (v3.3) and UMAP (V3.1) plugins for UMAP generation and visualization (see below).

UMAP visualization of flow cytometric data.

Contour plots were generated using ‘contour’ visualization in FlowJo (equal probability contouring). For UMAP projections, all samples stained with panel 2 were downsampled to 1,000 cells using the DownSample plugin (v3.3) available on the FlowJo Exchange. All samples were concatenated to create a single, 29,000-cell composite, and a UMAP algorithm for dimensionality reduction was applied using the UMAP plugin (v3.1) available on the FlowJo Exchange. The composite sample then was re-gated as indicated for all primary and secondary populations (Table 1) to aid in visual overlays in exploration of the UMAP projections. Density plots with levels set to 10% of gated populations were then projected onto the UMAP coordinates, with the outermost density representing 90% of the total gated cells (Fig. 2c,d). Alternatively, 90% densities were identified from total B cells from HD, OUT-C or ICU-C cohorts (Fig. 2b). These densities and underlying UMAP projections were exported to be superimposed and processed for display in Adobe Illustrator. Overlapping densities were removed using Adobe Illustrator’s intersection tools to reveal only densities occupied by OUT-C or ICU-C groups (Fig. 2a).

Analysis software.

Computational analysis was carried out in R (v3.6.2; release 12 Dec 2019). Heat maps were generated using the pheatmap library (v1.0.12), with data prenormalized (log-transformed z-scores calculated per feature) before plotting. Clustering was carried out using Ward’s method. Custom plotting, such as antigen-specific response curves, was performed using the ggplot2 library for base analysis, and then post-processed in Adobe Illustrator. Circos plotting was carried out using Circos software (v0.69–9). Lineage trees were calculated using GLaMST (Grow Lineages along Minimum Spanning Tree) software^61,62, run on MATLAB and then visualized in R using the iGraph package (v1.2.4.2). Exported trees then were post-processed in Adobe Illustrator. Statistical analyses were performed directly in R, or in GraphPad Prism (v8.2.1).

Flow cytometry and sorting of B cell subsets for repertoire sequencing.

Frozen cell suspensions were thawed at 37 °C in RPMI + 10% FCS and then washed and resuspended in FACS buffer (PBS + 2% FCS). The cells were incubated with a mix of fluorophore-conjugated antibodies for 30 min on ice. The cells were washed in PBS and then incubated with the live/dead fixable aqua dead cell stain (Thermo Fisher) for 10 min at 22 °C. After a final wash in FACS buffer, the cells were resuspended in FACS buffer at 10⁷ cells per ml for cell sorting on a three-laser BD FACS (BD Biosciences).

For single-cell analysis, total ASCs were gated as CD3⁻CD14⁻CD16⁻CD19⁺CD38⁺CD27⁺ single live cells, whereas naive B cells were gated as CD3⁻CD14⁻CD16⁻CD19⁺CD27⁻IgD⁺CD38⁺ single live cells.

For bulk sequencing preparations, B cells were enriched using StemCell’s Human Pan-B Cell Enrichment Kit (no. 19554; negative selection of CD2, CD3, CD14, CD16, CD36, CD42b, CD56, CD66b and CD123). CD138⁺ ASCs were enriched further using CD138⁺ selection beads according to the manufacturer’s instructions (Miltenyi Biotec, no. 130-051-301).

V(D)J repertoire library preparation and sequencing.

Cytokine immunoassays.

Plasma concentrations of IL-6 were quantified with ELISA using a Human IL-6 Quantikine ELISA Kit according to the manufacturer’s instructions (R&D Biosystems, no. HS600C). CRP was measured in a singleplex immunoassay (MilliporeSigma) in a Luminex-200 platform following manufacturer’s protocol (25 μl of 1:40,000 dilution in duplicates).

Carbodiimide coupling of microspheres to SARS-CoV-2 antigens.

Two SARS-CoV-2 proteins were coupled to MagPlex Microspheres of different regions (Luminex). Nucleocapsid (N) protein expressed from Escherichia coli (N-terminal His6) was obtained from Raybiotech (230-01104-100) and the RBD of spike (S) protein expressed from HEK293 cells was obtained from the laboratory of J. Wrammert⁶³ at Emory University. Coupling was carried out at 22 °C following standard carbodiimide coupling procedures. Concentrations of coupled microspheres were confirmed by Bio-Rad T20 Cell Counter.

Luminex proteomic assays for measurement of anti-antigen antibody.

Approximately 50 μl of coupled microsphere mix was added to each well of 96-well clear-bottom black polystyrene microplates (Greiner Bio-One) at a concentration of 1,000 microspheres per region per well. All wash steps and dilutions were accomplished using 1% BSA, 1× PBS assay buffer. Sera were assayed at 1:500 dilutions and surveyed for antibodies against N or RBD. After a 1-h incubation in the dark on a plate shaker at 800 r.p.m., wells were washed five times in 100 μl of assay buffer, using a BioTek 405 TS plate washer, then applied with 3 μg ml⁻¹ PE-conjugated goat anti-human IgA, IgG and/or IgM (Southern Biotech). After 30 min of incubation at 800 r.p.m. in the dark, wells were washed three times in 100 μl of assay buffer, resuspended in 100 μl of assay buffer and analyzed using a Luminex FLEXMAP 3D instrument (Luminex) running xPONENT 4.3 software. MFI using combined or individual detection antibodies (anti-IgA, anti-IgG or anti-IgM) was measured using the Luminex xPONENT software. The background value of assay buffer was subtracted from each serum sample result to obtain MFI minus background (MFI-B; net MFI).

Quantification of 9G4-idiotype IgG in serum.

Polysorp 96-well plates (Thermo Fisher, no. 475094) were coated overnight with 0.5 μg ml⁻¹ rat anti-human 9G4 IgG diluted in PBS. SuperBlock (Thermo Fisher, no. 37515) was used as blocking reagent. Serum samples were then incubated for 60 min at 22 °C at a dilution of 1:8,000 (9G4⁺ IgG) followed with three twofold dilutions. Positive 9G4 IgG sera were detected by AP-conjugated goat anti-human IgG (Fc specific) at a dilution of 1:15,000 (Sigma, no. A9544). The reaction was visualized by the addition of 100 μl KPL BluePhosÒ Microwell Phosphatase Substrate System (SeraCare, no. 5120–0059). The ELISA OD 9G4 IgG results were quantified with a human 9G4⁺ monoclonal IgG standard curve. Plates were washed three times with washing buffer (PBS (pH 7.4) containing 0.1% (vol/vol) Tween 20) after each step.

Focus-reduction neutralization test.

Indicated dilutions of plasma were incubated with 102 focus-forming units (FFU) of SARS-CoV-2 for 1 h at 37 °C. Antibody–virus complexes were added to indicated cell monolayers in 96-well plates and incubated at 37 °C for 1 h. Subsequently, cells were overlaid with 1% (wt/vol) methylcellulose in MEM supplemented with 2% FBS. Plates were harvested 30 h later by removing overlays and fixed with 4% PFA in PBS for 20 min at 22 °C. Plates were washed and sequentially incubated with 1 mg ml⁻¹ of CR3022 anti-S antibody^64,65 and HRP-conjugated goat anti-human IgG in PBS supplemented with 0.1% saponin and 0.1% BSA. SARS-CoV-2-infected cell foci were visualized using TrueBlue peroxidase substrate (KPL) and quantified on an ImmunoSpot microanalyzer (Cellular Technologies). Data were processed using Prism software v8.0 (GraphPad).

Statistical analysis.

Statistical analysis was carried out using Prism (GraphPad). For each experiment, the type of statistical testing, summary statistics and levels of significance can be found in the figures and corresponding legends. All measurements displayed were taken from distinct samples.

Reporting Summary.

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