Samples used in this study for the severe SARS-CoV-2 cohort were from individuals hospitalized in the ICU St. Joseph’s Hospital (SJH Denver) or at the University of Colorado Hospital (UCH). An approved waiver of consent (UCH) or a legally authorized representative (SJH) provided informed consent to donate whole blood. Individuals who donated blood had a nasal swab to confirm the presence of the virus by polymerase chain reaction (PCR), were at least 18 years old, and mechanically ventilated for acute respiratory distress syndrome. The exclusion criteria for this study included individuals with solid organ or bone marrow transplants, hemoptysis, chronic lung disease, pregnancy, increased risk for bleeding, or immunosuppressed. Whole blood was collected from central venous catheters in sodium citrate collection tubes and processed per the manufacturer’s instructions (BD Biosciences, San Jose, CA), with plasma and PBMCs stored as previously described (30, 31).

The mild SARS-CoV-2 infection cohort samples were included in this study if they were confirmed to have the virus by nasal swab PCR or the presence of anti-SARS-CoV2 antibodies, did not require hospitalization for infection, and were able to donate whole blood in the convalescent stage. These samples were collected before vaccination was available to the general public. The samples used in this study for the immunized cohort were from individuals vaccinated against SARS-CoV-2 by Pfizer BNT162b2-mRNA or Moderna mRNA-1273 who were able to provide dates of their primary inoculation and booster. Informed consent was provided by these individuals. Whole blood was drawn by the Colorado Clinical and Translation Sciences Institute (CCTSI). PBMCs and plasma were collected and stored as previously described (30, 31).

Samples from healthy controls were obtained from Vitalant Blood Center (Denver, CO). PBMCs were isolated from plateletpheresis leukoreduction filter (LRS chambers) as previously described (30, 31). Since LRS chambers do not allow for the acquisition of plasma, in this study when plasma was examined, the immunized individuals were used as a comparator to SARS-CoV-2 individuals.

The Colorado Multiple Institutional Review Board (COMIRB) approved the use of human samples and this study, which was performed under the Declaration of Helsinki.

To characterize the frequency and phenotype of memory B cells, PBMCs were thawed as previously described (30, 31). PBMCs were stained on ice for 20min in 1X PBS with antibodies (clone): CD3 (OKT3), CD27 (M-T271), CD19 (SJ25C1), IgD (IA6-2), CD69 (FN50), CD86 (IT2.2), CD21 (Bu32), CD22 (HIB22), CD72 (REA231), IgM (MHM-88), and IgG (G18-145) with Live/Dead Blue for UV viability dye (Invitrogen, Eugene, OR). After surface stain, PBMCs were washed and fixed with 4% PFA (Fisher, Fair Lawn, NJ). To run on the flow cytometer, cells were resuspended in 1% BSA with 0.05% Sodium Azide in 1X PBS.

The spectral Cytek Aurora was utilized to acquire all flow cytometry data and data were analyzed using FlowJo software (v 10.8.1). Single color reference controls were generated using PBMCs from healthy humans. When necessary, Ultra Comp eBeads (Invitrogen) were used. Gating through lymphocytes, singlets, live cells, and CD3- cells, unswitched memory B cells were identified as CD19+ CD27+ IgD+. The IgD+only subset within unswitched memory B cells was identified by then gating on IgM negative cells. Gating through lymphocytes, singlets, live cells, then CD3- cells, switched memory B cells were gated as CD19+ CD27+ IgD-. The frequency of unswitched or switched memory B cells were enumerated out of the combined memory population (B cells that are CD27+) or out of total B cells (CD19+).

To evaluate signaling downstream of the BCR in memory B cells, PBMCs were thawed then incubated in warm serum-free RPMI Medium 1640 for 45 min at 37˚C with 5% CO2 to reduce basal phosphorylation levels while in the presence of the following antibodies (clone): CD3 (OKT3), CD27 (M-T271), and CD19 (SJ25C1) with Live/Dead Blue for UV viability dye. PBMCs were resuspended in warm RPMI with 5% FBS and stimulated with either 10ug/mL Rabbit anti-human IgG (H+L) F(ab’)2 (Southern Biotech, Birmingham, AL) or positive control 75μM pervanadate for 5 minutes in a 37˚C water bath. After stimulation, PBMCs were BD Cytofix/Cytoperm for 20 minutes on ice and washed in 1X Perm/Wash (BD). The following antibodies (clone): pSYK Y348 (I120-722), IgD (IA6-2), and pPLCγ2 Y759 (K86-689.37) were used for intracellular staining for 30 min on ice in 1X Perm/Wash before data acquisition.

PBMCs were thawed as described above and total B cells were enriched using the Mojo Human anti-CD3 selection kit (Biolegend, San Diego, CA) with the addition of anti-CD16-biotin (clone 3G8) and the Easy Sep magnet (Stem Cell Technologies, Cambridge, MA) to deplete non-B cells. Enriched B cells were then incubated on ice for 20min with the following antibodies CD3 (OKT3) and CD19 (SJ25C1) in the presence of Zombie green viability dye (Biolegend), washed in 1X PBS and resuspended in sorting buffer comprised of 1X PBS with 1mM EDTA, 25mM HEPES, and 1% BSA then sorted on the BD Aria Fusion Cell Sorter at the University of Colorado Allergy and Clinical Immunology Flow Cytometry Facility. Purified B cells were plated in 96 well tissue culture plates in RPMI 1640 media with 10% FBS, 1X Glutamax, and 1X Penicillin/streptomycin (Gibco) for 18 hours at 37˚C with 5% CO2 in the presence of 50ng/mL PMA (Sigma) with 1ug/mL Ionomycin (Sigma) or 10ug/mL anti-Ig (H+L) F(ab)’2 (Southern Biotech) and 100ng/mL CD40L (Biolegend) with either 1ug/mL TL3RL (Poly I:C), 1ug/mL TLR7/8L (R848), or 2.5ug/mL TLR9L (CpG) (all from InvivoGen, San Diego, CA) as indicated in the figure legends. The supernatant was collected after incubation and cytokines were enumerated using the U-PLEX Assay according to the manufacturer’s instructions (Meso Scale Discovery, Rockville, MD). The QuickPlex SQ 120 Instrument was used the read the Mesoscale plate (Meso Scale Discovery).

To determine whether memory B cells were capable of producing IL-6 and TNFα cytokines, PBMCs were thawed, incubated for 15 min on ice with antibodies IgD (IA6-2) and IgM (MHM-88), washed by centrifugation 1,500 rpm for 5min in 1X PBS then incubated for 4 hours at 37˚C with 5% CO2 in RPMI 1640 media with 10% FBS, 1X Glutamax and 1X Penicillin/streptomycin (Gibco) with the stimuli described above in the presence of Golgi Plug (BD). After incubation, cells were washed by centrifugation at 1,500 rpm for 5 min in 1X PBS, stained for 20 min on ice with antibodies CD19 (SJ25C1) and CD27, washed again, resuspended in BD Cytofix/Cytoperm for 20 minutes on ice. Cells were then washed in 1X Perm/Wash (BD) and incubated in 1X Perm/Wash with antibodies (clone): IL-6 (MQ2-13A5), TNFα (Mab11), IgG (G18-145) or isotype controls for intracellular staining for 30 min on ice. Cells were washed with 1X Perm/Wash by centrifugation at 1,500 rpm for 5 min at 4˚C, fixed by incubation in 4% PFA before data acquisition. Isotype controls were used to set gates.

Plasma was thawed on ice and levels of TNFα in plasma and quantified using the U-PLEX Assay (Meso Scale Discovery, Rockville, MD). The QuickPlex SQ 120 Instrument was used to evaluate cytokine levels (Meso Scale Discovery).

Autoreactive IgG antibodies and total IgG levels were enumerated in plasma samples as previously described (30, 31).

Prism Graph-Pad Software (v 9.2.0) was used for statistical analysis. Paired t test was used to determine the significance of differences between unswitched or switched memory B cells. The significance of differences between cohorts (mild or severe infection vs. healthy controls vs immunized samples) was evaluated using one-way ANOVA or between stimulations tested in vitro. The significance of differences between TNFα levels with the frequency of memory populations or between the frequency of memory populations with autoreactive antibody titers was examined using a Pearson correlation. Significance was defined as p< 0.05 and the type of statistical test used is indicated in each figure legend. Each dot on the scatter plot is a single human donor. Each figure legend states the number of human donors examined in each assay.

To better understand the phenotypic and functional differences between unswitched and switched memory B cells, particularly in the context of viral infection, we acquired PBMCs from healthy controls, people immunized via the mRNA platform against SARS-CoV-2, individuals with mild SARS-CoV-2 in the convalescent stage, and patients with severe SARS-CoV-2 who were hospitalized in the ICU at local hospitals ( Supplemental Table 1 ). For this study, we evaluated total memory B cells irrespective of antigen specificity. Blood was collected between 7-507 days post booster for vaccinated individuals, between 28-305 days post positive PCR test for mildly infected individuals or between 2-35 days post admittance to hospital for severely infected subjects ( Supplemental Table 1 ). The memory B cell compartment comprises unswitched memory B cells identified in human PBMCs via CD27 and IgD (CD27+ IgD+) expression, whereas switched memory B cells identified by expression of CD27 but not IgD (CD27+IgD-) ( Figure 1A ). In healthy controls, the unswitched memory B cell population was less frequent (~33% of CD27+ B cells) than the switched memory B cell population (~66% of CD27+ B cells; Figure 1B ). Similarly, unswitched memory B cells were less frequent than switched memory B cells within the memory compartment of the other cohorts ( Figure 1B ).

Comparing the frequency of unswitched memory B cells in peripheral blood between cohorts, we observed a significant reduction in the proportion of unswitched memory B cells within the memory compartment during severe SARS-CoV-2 infection when compared to healthy controls, immunized or mildly infected individuals and a corresponding increase in switched memory B cells with severe infection ( Figure 1B, C ). Examination of the unswitched memory B cells within the total B cell population revealed a reduction in the percentage of unswitched memory B cells with severe SARS-CoV-2 infection when compared to mild infection, vaccination, or healthy controls ( Figure 1D ), in keeping with previous reports (22–25, 27).

Individuals with RA have been reported to harbor a reduced frequency of unswitched memory B cells, however, RA individuals who respond positively to anti-TNFα treatment display normal frequencies of unswitched memory B cells in peripheral blood (17), supporting the possibility that elevated levels of systemic TNFα may promote loss of unswitched memory B cells. We have previously reported that individuals with severe SARS-CoV-2 infection have significantly increased levels of TNFα in plasma (30). Accordingly, we compared the frequency of unswitched memory B cells with plasma levels of TNFα in the same cohorts and found that the frequency of unswitched memory B cells demonstrates a significant inverse correlation with levels of TNFα, a relationship that was not observed with switched memory B cells ( Figure 1E ).

Multiple reports provide evidence that unswitched memory B cells are protective during viral infection; the frequency of unswitched memory B cells negatively correlated with disease activity during SARS-CoV-2 infection (22), unswitched memory B cells in principal component analysis of clinical biomarkers clustered with patient discharge from the hospital (25) and, although identified by different phenotypic criteria (CD38^-CD24⁺), unswitched memory B cells negatively correlate with symptom duration (29). Here, we compared the frequency of unswitched memory B cells from individuals with severe SARS-CoV-2 whose clinical outcome was survival versus death. These analyses did not find a significant difference in the frequency of unswitched memory between patients who survived versus those that succumbed to viral infection ( Supplemental Figure 1 ), demonstrating frequency does not predict clinical outcome in our cohort.

The unswitched memory B cell population (CD27+ IgD+) includes two subpopulations; the IgD+IgM+ memory subset and the IgD+IgM- subset which has lost expression of IgM and is termed ‘IgD+only’ memory cells ( Figure 2A ). It has been reported that IgD+only memory cells have somatically mutated V genes indicative of antigen experience (32), however, their role in viral infection has not been evaluated to date. In healthy controls, we found that amongst the unswitched memory B cell population, the IgD+only memory subset is approximately 5% and much less frequent than the IgD+IgM+ subset ( Figure 2B ). Similarly, immunized individuals and mild or severely infected subjects also harbor an IgD+only memory subset that is less frequent than the IgD+IgM+ subset within the unswitched memory population ( Figure 2B ). Notably, however, we observed a significant increase in the IgD+only memory B cell population with severe SARS-CoV-2 infection representing approximately 12.5% and a corresponding decrease in the IgD+IgM+ subset out of the unswitched memory population ( Figure 2C ). These results indicate that viral infection promotes the expansion of the IgD+only memory subset or, conversely, a loss of IgD+IgM+ memory B cells.

We and others have reported increased titers of autoreactive antibodies with severe SARS-CoV-2 infection demonstrating viral infection promotes a breach in B cell immunological tolerance (30, 33–36). IgD+only memory cells from healthy human controls are enriched in high-affinity autoreactive BCRs with specificity towards anti-nuclear antigens, ss/dsDNA, LPS, and insulin (7). Given that we observe an increased frequency of IgD+only cells, we asked whether there was an association between this subset and SARS-CoV-2 infection-induced autoantibody production. Importantly, we find a significant correlation between the frequency of IgD+only cells within the total B cell population and plasma titers of not only VH4-34 IgG but also with levels of IgG anti-chromatin, anti-Smith, and anti-cardiolipin ( Figure 2D ). Furthermore, neither the frequency of CD27+IgD+ unswitched memory B cells nor CD27+IgD- switched memory B cells correlated with levels of autoantibodies ( Supplemental Figure 2 ), highlighting the specificity of the relationship between the presence of (CD27+IgD+IgM-) IgD+only memory cells and autoantibody titers. These data demonstrate that as the frequency of IgD+only cells increases with SARS-CoV-2 infection, autoreactive antibody titers also increase, such that IgD+-only B cells appear to be a novel contributor to autoreactive antibody production during viral infection.

Since severe SARS-CoV-2 infection resulted in an alteration of the unswitched memory compartment with an increase in the percentage of IgD+only B cell population and a concomitant decrease in IgM+IgD+ B cells ( Figure 2 ), we questioned whether the switched memory population subsets were also altered with severe viral infection, thus we examined the total switched memory population, rather than antigen-specific B cells. The switched memory population is comprised of memory B cells expressing immunoglobulin isotype class IgM (pre-switched) or IgG ( Supplemental Figure 3A ) and a substantial subset of IgM-IgG- which are IgA-expressing memory B cells ( Supplemental Figure 3B ). Healthy controls harbor IgG-expressing and IgM-IgG- (IgA+) memory B cells that are more frequent relative to IgM-expressing memory B cells ( Supplemental Figure 3C ). A similar composition was observed in switched memory subsets from vaccination or mild infection ( Supplemental Figure 3C ). However, comparing switched memory subsets between the cohorts revealed that there is a significant reduction in IgG+ memory B cells during severe SARS-CoV-2 infection compared to healthy controls and mildly infected individuals, and a significant increase in IgM-IgG- (IgA+) memory B cells when comparing mild to severe SARS-CoV-2 infection ( Supplemental Figure 3D ). Together these data demonstrate that severe SARS-CoV-2 infection modulates the memory compartment immunoglobulin isotype.

To evaluate how viral infection modulates the activation status of unswitched memory B cells, we evaluated the expression of the CD86 and CD69 activation antigens as well as the CD21 co-receptor on total memory B cells (rather than antigen-specific B cells). These analyses demonstrated a significant increase in expression of CD86, but not CD69, and a significant reduction of CD21 on unswitched memory B cells with severe SARS-CoV-2 infection ( Figures 3A, B ), phenotypic changes indicative of an activated population. Similarly, we observed that switched memory B cells had significantly increased expression of CD86 and a significant loss of CD21 with severe SARS-CoV-2 infection compared to mild infection, vaccination, or healthy controls ( Figures 3C, D ). Switched memory B cells also displayed a significant increase in the expression of CD69 with severe SARS-CoV-2 infection ( Figures 3C, D ). Switched memory B cells were observed to have higher expression of CD86 compared to unswitched memory B cells in severe infection, whereas unswitched memory B cells displayed higher expression levels of CD21 than switched memory ( Figure 3E ). Interestingly, despite the increase in CD69 expression on switched memory B cells with SARS-CoV-2 infection, expression of this protein never reached the levels of CD69 expression on unswitched memory ( Figure 3E ), such that unswitched memory B cells reside at a higher activation state in general. These data indicate that both the unswitched and switched memory B cell compartments are highly activated with severe viral infection.

Because severe viral infection results in phenotypically activated unswitched and switched memory B cells ( Figure 3 ), we sought to determine if activation of the memory B cell populations also coincides with a reduction in inhibitory receptors by memory B cells through evaluation of CD72 and CD22. Unswitched memory B cells had a significant reduction in CD72 expression during severe SARS-CoV-2 infection when compared to mild infection, vaccination, or healthy controls ( Figures 4A, B ). A similar reduction in the expression of CD72 was observed on switched memory B cells from severely infected SARS-CoV-2 patients ( Figures 4A, C ). Interestingly, immunization to SARS-CoV-2 with mRNA vaccination increased the expression level of CD22 on both unswitched and switched memory B cells as compared to healthy controls such that there was a significant difference in CD22 expression between immunized and severely infected individuals ( Figures 4A, C ). Comparison of unswitched to switched memory B cells revealed that unswitched memory B cells have higher expression of CD72 and CD22 than switched memory B cells during severe SARS-CoV-2 infection ( Figure 4D ), indicating differential regulation of these populations.

We have previously reported that Double Negative B cells signal via the BCR during viral infection (31); however, the BCR signaling capacity of unswitched memory B cells during viral infection has not been reported. To address this, we stimulated PBMCs with and without anti-human Ig to induce BCR signaling and assessed expression levels of phospho-SYK (pSYK) and pPLCγ2, effector molecules in the BCR signal transduction cascade. The results from these analyses revealed that BCR signal transduction is intact in unswitched memory B cells from all cohorts as evidenced by significant increases in pSYK and pPLCγ2 levels upon BCR stimulation ( Figure 5A ). Similar results were also observed for switched memory B cells ( Figure 5B ). Thus, despite the activated and dysregulated state of memory B cells with severe SARS-CoV-2 infection, they maintain the ability to signal through the BCR. Importantly, upon comparison of unswitched to switched memory B cells from severe SARS-CoV-2 infection, unswitched memory B cells express higher levels of pSYK and pPLCγ2 after BCR stimulation than switched memory B cells ( Figure 5C ), (which was surprising given higher expression of inhibitory receptors CD72 and CD22 on unswitched memory B cells) demonstrating that unswitched memory B cells are better poised to respond to antigen recognition than switched memory B cells during severe viral infection.

B cells from healthy individuals are capable of producing cytokines in response to TLR ligand stimulation (21, 37) and we and others have previously reported that individuals with severe SARS-CoV-2 infection have increased levels of pro-inflammatory cytokines (30, 38, 39). Accordingly, we assessed whether B cells could contribute to the inflammatory milieu during viral infection. To address this, we purified total B cells from a small subset of healthy controls and mild SARS-CoV-2 infection, stimulated the cells in vitro, and measured cytokine production in the supernatant. Stimulation of total B cells with PMA and ionomycin for 18 hours significantly increased the secretion of cytokines IL-6 and TNFα, but not IL-1β ( Figure 6A ). To determine if B cells participating in immune response (after BCR stimulation and T cell help) could secrete these cytokines in response to viral-specific TLR ligands, we stimulated purified B cells from healthy controls and mild SARS-CoV-2 infection with ligands specific for TLR3, TLR7/8, or TLR9 in the presence of polyclonal BCR crosslinking and T cell help. These results demonstrated that B cells can contribute to the inflammatory cytokine milieu by secreting pro-inflammatory IL-6 and TNFα, in response to all three viral TLR ligands ( Figure 6B ). We did not see differences in the amount of IL-6 or TNFα when comparing healthy controls with mild SARS-CoV-2 infection ( Supplemental Figure 4 ). Interestingly, the highest levels of secreted cytokines were detected in stimulations with TLR7/8L or TLR9L as compared to TLR3L ( Figure 6B ) with the highest levels of IL-6 or TNFα secreted in response to TLR9L, even above levels from PMA and ionomycin stimulation ( Figure 6 ). Together these data indicate that B cells are able to and likely contribute to the pro-inflammatory cytokine milieu during viral infection.

Although it has been reported that total (unswitched and switched) memory B cells (CD27+) from healthy controls produce more cytokines than naive B cells (21), we asked whether unswitched memory B cells per se were capable of producing cytokines. Stimulation of PBMCs with PMA and ionomycin from a subset of healthy controls and mild SARS-CoV-2 infection resulted in a significantly increased proportion of IL-6-expressing unswitched memory B cells ( Figure 7A ), that were predominantly IgD+IgM+ unswitched memory B cells rather than the IgD+only subset ( Supplemental Figure 5A ). Similarly, PMA and ionomycin stimulation significantly increased the frequency of IL-6-expressing switched memory B cells ( Figure 7A ) and the majority of the IL6-expressing switched memory cells were IgG+ followed by a fraction of IgM-IgG- (IgA+) cells in both healthy controls and mildly-infected individuals ( Supplemental Figure 5B ). Examination of TNFα expression revealed that both unswitched and switched memory B cells produce TNFα in response to PMA and ionomycin stimulation ( Figure 7B ), with the dominant fraction of TNFα-expressing unswitched memory IgD+IgM+ and switched memory IgM-IgG- (IgA+) in both healthy controls and mildly-infected individuals ( Supplemental Figures 5A, B ).

Comparing unswitched to switched memory B cells, the frequency of IL-6-expressing memory B cells were significantly higher for unswitched memory B cells compared to switched memory B cells after stimulation with PMA and ionomycin stimulation ( Figure 7C ), whereas there was a significantly higher frequency of TNFα- expressing switched memory compared to unswitched memory ( Figure 7C ). These findings demonstrate unswitched memory cells are more poised to produce the proinflammatory IL-6 cytokine, whereas, switched memory B cells are better poised to produce TNFα. We further queried whether memory B cells were polyfunctional in their ability to produce both IL-6 and TNFα. Although polyfunctional IL-6- and TNFα-expressing unswitched memory and switched memory B cells were observed, the majority of each population from healthy controls and mild SARS-CoV-2 infection produced a single cytokine rather than both with unswitched memory B cells mostly producing IL-6 and switched memory B cells mostly expressing TNFα ( Figure 7D ). Together these results indicate that unswitched memory B cells are capable of producing pro-inflammatory cytokines.

We next determined if unswitched and switched memory B cells participating in an immune response upon BCR engagement and provision of T cell help could produce IL-6 and TNFα in response to viral TLR ligands. Thus, we stimulated PBMCs from healthy controls and mild SARS-CoV-2 infection with ligands specific for TLR3, TLR7/8, or TLR9 in the presence of polyclonal BCR crosslinking and T cell help. These data revealed that stimulation with TLR7/8L and TLR9L, but not TLR3L, significantly increased the frequency of IL-6-expressing unswitched and switched memory B cells ( Figure 8A ). Finally, stimulation with TLR7/8L or TLR9L but again not TLR3L significantly increased the frequency of TNFα-expressing unswitched or switched memory B cells ( Figure 8B ). We did not see differences in the frequency of IL-6 or TNFα- expressing unswitched or switched memory B cells when comparing healthy controls to individuals with mild SARS-CoV-2 infection ( Supplemental Figures 5C, D ). Together these data indicate that memory B cells differentially respond to viral TLR ligands (more poised to respond to ssRNA or unmethylated DNA rather than dsRNA) and demonstrate that unswitched memory B cells likely contribute to the pro-inflammatory cytokine milieu during viral infection.