The primary goal for this study was to test the hypothesis that virus-reactive antibody levels are reflecting on the frequency of Bmem cells. Based on long held “textbook knowledge”, such a close correlation could be expected, as (a) plasma cells are thought to be long-lived and continuously produce antibody, (b) Bmem cells are also assumed to be long-lived, and (c) both cell types were thought to arise during the course of a B cell immune response at a constant ratio from a common precursor cell. The recent literature, discussed below, including the data presented in the following communication, call into question that such a direct correlation would be the rule after immune responses to viral infections.
Our first concern addressing this hypothesis was to optimize the accuracy of measuring both antigen-reactive antibody levels and the frequency of antigen-reactive Bmem cells. For the former, we performed standard ELISAs involving serial dilutions of the donor plasma samples in conjunction with a “µg/mL equivalent scale” for analyzing the data [52–55]. This approach is superior to the “area under the curve” approach as it leverages an internal (plate-specific) reference standard and thus is independent of assay-associated variability related to development time and temperature.
While the methodology for the quantification of antigen-reactive antibodies in plasma by ELISA is well-established, we first needed to develop such for the objective and accurate counting of antigen-reactive Bmem cells in the blood. Empirically, antigen-reactive B cell ImmunoSpot® assays result in diverse spot morphologies  (see also Figs. 1B and 1C). This outcome is expected, as the secretory foot-prints of individual ASC is defined by a multitude of parameters , including (a) the net amount of antibody produced by the individual B cells during the assay’s duration, (b) the kinetics of antibody production by the ASC, and (c) the functional affinity of each ASC’s antibody for the antigen (that can encompass a broad spectrum among an antigen-reactive B cell repertoire defining the capture and dissociation rates of antibody binding to antigen). Furthermore, pan-well or regional “ELISA” effects modulate the background membrane staining when antibody that “escaped” into the supernatant is captured distally from the source ASC. Crowding of spots also interferes with unambiguous counting. For all these reasons, the first part of this manuscript is dedicated to establishing unambiguous frequencies of antigen-reactive B cells in PBMC.
3.1. Establishing unambiguous frequencies of antigen-reactive B cells in PBMC
The accurate detection and counting of antigen-reactive B cells in PBMC predicts a linear relationship between the cell numbers plated into the ImmunoSpot® assay, and the antigen-reactive SFU counted. This is because B cells are the only cell type that secrete antibodies, and because each B cell produces antibody with pre-defined specificity. For each antigen reported in this study, we have established the ideal test conditions for accurate frequency measurements. PBMC have been polyclonally activated with R848 and IL-2 for five days to promote the differentiation of resting Bmem cells (which do not secrete antibody) into ASC that can be detected in ImmunoSpot® assays via their secretory foot-prints [35–37]. Such pre-stimulated PBMC were plated in serial dilution into antigen-coated assay plates while increasing the number of replicate wells at lower cell numbers to account for the expected Poisson variation. Suppl. Figure 2 shows raw data for such experiments; in the example shown two representative donors that have recovered from COVID-19 were evaluated for ASC reactivity against the SARS-CoV-2 S1 protein.
As expected, the number of detectable SFU decreased with the number of stimulated PBMC plated (Fig. 1A and Suppl. Figure 2A). However, secretory foot-prints of individual ASC were not readily resolved at the highest cell inputs due to the confluence of spots and elevated background membrane staining arising as a consequence of the elevated antigen-reactive ASC frequency. As expected, at lower cell inputs the foot-prints of individual ASC became clearly discernable and facilitated accurate enumeration of individual antigen-reactive ASC in such wells (Fig. 1B). Also as expected, the SFUs displayed a wide spectrum of sizes and fluorescent intensities (Figs. 1B and 1C) reflecting on the individual ASC’s range of productivity and functional affinity for the SARS-CoV-2 S1 protein. Therefore, when we used machine reading for the enumeration of SFU, we counted all SFU, including all sizes and densities. Counting of such “ungated events” showed a close to perfect linear relationship between SFU counts and cell numbers plated per well for (but only for) wells that contained low numbers of SFU: for COVID-19 Donor 1, for example, these were wells containing ≤ 2 x 104 PBMC/well (Fig. 1E). For this donor, SFU counts for cell numbers exceeding 2 x 104 PBMC/well were reduced below the expected count (Fig. 1D). Once the frequency-dependent linear range has been established for a subject, linear regression permitted to reliably calculate the frequency of antigen-reactive B cells within the PBMC: for COVID-19 Donor 1 this number was computed to be 195 SARS-CoV-2 S1-reactive ASC/105 PBMC, for COVID-19 Donor 2, providing another example, it was 163 S1-reactive ASC/105 PBMC (Suppl. Figure 2B).
The above data establish that accurate counting of antigen-reactive B cells in ImmunoSpot® assays requires to establish the range in which SFU counts and cell input per well exhibit a linear relationship from which frequencies can be reliably extrapolated. This is critical as the numbers of antigen-reactive B cells can span several orders of magnitudes even between antigen-primed individuals (see below). We optimized test conditions, and verified these basic assumptions for the accurate counting of SFUs for all antigen-reactive B cell ImmunoSpot® assays reported in this publication: representative results from such experiments are shown in Suppl. Figures 2–4.
As a positive control, we also performed total IgG+ ImmunoSpot® assays in parallel to verify the functionality of B cells following their in vitro stimulation. In this assay variant, the membrane is coated with anti-human immunoglobulin light chain (IgL) (anti-Igκ/Igλ) capture antibody instead of the antigen itself, capturing the secretory foot-print of all ASC irrespective of their binding reactivity. Applying the rules described above we have established the frequency of total IgG+ ASC in each test subject through serial dilution (the raw data and results for six representative samples are depicted in Suppl. Figure 5). Because the frequency of total IgG+ ASC exhibited considerable inter-individual variation among the test subjects following in vitro stimulation, when frequencies of antigen-reactive IgG+ ASC are reported in the following, they are expressed as a percentage of the total IgG+ ASC compartment, i.e., “% antigen-reactive IgG+ ASC” .
3.2. Equal overall assay performance of pre- and post-COVID-19 PBMC
SARS-CoV-2 exposure, unlike the endemic seasonal influenza and EBV infections also studied here, is unique in as much that the exposure can be verified beyond measuring antibody reactivity. As the first laboratory confirmed case of SARS-CoV-2 virus infection in the United States occurred on January 20, 2020 , PBMC collected before November 1, 2019 have been verifiably derived from subjects who could not have been exposed to SARS-CoV-2, and thus must be immunologically naïve to this virus. It is a matter of debate, however, how much T and B cell cross-reactivity exists in such pre-COVID-19 era subjects elicited through prior exposure(s) to endemic coronavirus strains [58–61], and possibly other antigens . It is also remains unclear whether such pre-existing, cross-reactive immunity contributes to defining the severity of the primary SARS-CoV-2 infection. Also, unlike for influenza and EBV, not only exposure itself but even the time-point of the primary SARS-CoV-2 infection can be verified by PCR testing. For this study we therefore could compare B cell reactivity to SARS-CoV-2 antigens in two cohorts that are highly defined with respect to immune exposure to this virus: PBMC collected from individuals who could not have been infected (the “pre-COVID-19” cohort), and those who have been collected after PCR-verified SARS-CoV-2 infection during the first wave, in 2020, caused most likely by the original “prototype” Wuhan-Hu-1 strain (the “COVID-19” cohort). Notably, all PBMC were collected before vaccination became available.
PBMC of the test subjects were cryopreserved according to a protocol that maintains B cell functionality . As shown in Suppl. Figure 6A, the viability of the PBMC was comparable for both cohorts after thawing. So was the functionality: the number of total IgG+ ASC, while showing the expected inter-individual variation within each group was comparable among the two cohorts (Suppl. Figure 6B). When tested in two separate experiments, the frequencies of total IgG+ ASC reproduced well for the individual donors (Suppl. Figure 6C) suggesting that the inter-donor variability seen is inherent to each donor, and does not represent an assay variable. Suppl. Figure 6D depicts the correlation between PBMC viability after five days of polyclonal stimulation for each donor in both cohorts plotted versus the total IgG+ ASC frequency (that is, “what percentage of viable cells were IgG+ ASC”). While, as expected, there was no direct correlation between overall viability of the PBMC and total IgG+ ASC frequency, the cells from both cohorts behaved alike. Collectively, these observations suggest that if any difference is seen in antigen-reactive frequencies among these cohorts that it cannot be attributed to the freezing conditions or the duration of time the cells were stored under liquid nitrogen. (The comparison of influenza- and EBV-reactive B cell frequencies between the two cohorts below, in Fig. 4, and the inter-assay reproducibility of both total and antigen-reactive IgG+ ASC frequencies, presented in Suppl. Figures 6C and 7, respectively, will further substantiate this claim).
3.3. Exquisite specificity of SARS-CoV-2 antigen-reactive B cell ImmunoSpot® assays
With the exception of a single donor, PBMC from all other subjects in the COVID-19 cohort (n = 25), collected following convalescence from SARS-CoV-2 infection, displayed S1-reactive IgG+ ASC after polyclonal stimulation, albeit in frequencies that spanned three orders of magnitude, ranging between 0.01–1.15% of all IgG+ ASC (Fig. 2A). In contrast, less than 5 S1-reactive IgG+ ASC were detected in any of the pre-COVID-19 era donors, even at the highest cell number tested (3 x 105 PBMC/well). Importantly, at this high cell input such S1-reactive IgG+ ASC were often too numerous to accurately count for several COVID-19 donors (Suppl. Figure 2).
Testing for Bmem cell reactivity against the SARS-CoV-2 nucleocapsid (NCAP) protein reproduced the exquisite specificity seen for the S1 antigen (Fig. 3A). As with the S1 antigen, NCAP-reactive ASC occurred in frequencies less than 5 SFU per 3 x 105 PBMC for all pre-COVID-19 era donors tested. In stark contrast, IgG+ ASC with reactivity against seasonal influenza and/or EBV antigens were readily detectable in pre-COVID-19 donors (Fig. 4 and Suppl. Table 4). This absence of detectable NCAP-reactive IgG+ ASC in pre-COVID-19 donors is of particular interest, firstly, because the NCAP protein expressed by novel SARS-CoV-2 virus strains share more sequence conservation and predicted immunogenic epitopes with circulating seasonal coronaviruses than the Spike (S1) protein and as such is more likely to cross-react with B cells triggered by seasonal coronaviruses [64, 65], and second, because establishing immunity to NCAP is gaining importance for the immunodiagnostic of SARS-CoV-2 infection in COVID-19 vaccinated (i.e., Spike antigen immunized) individuals.
3.4. Discordance between SARS-CoV-2 antigen-reactive memory B cell frequencies and antibody levels
We also performed ELISA assays involving the same recombinant SARS-CoV-2 antigen preparations as applied above for ImmunoSpot® testing. Plasma from both cohorts were tested in serial dilutions alongside an IgG reference standard to establish for each subject the concentrations of S1 (Fig. 2B) and NCAP (Fig. 3B) antigen-reactive IgG levels. Judged as cohorts, the results were clear-cut and revealed significantly increased IgG reactivity against the SARS-CoV-2 S1 and NCAP coating antigens. However, the results were less clear-cut when judged at the level of individual subjects. Approximately half of the subjects in the COVID-19 cohort showed low levels of IgG reactivity against either the S1 or NCAP coating antigens while displaying elevated frequencies of antigen-reactive IgG+ ASC (derived from memory B cells) following in vitro stimulation (Figs. 2C and 3C). Thus, for immunodiagnostic purposes, detection of B cell memory appears to be a more sensitive and reliable indicator of SARS-CoV-2 infection history than measurements of antibody reactivity alone.
For both the SARS-CoV-2 S1 and NCAP antigens, plasma IgG levels and frequencies of antigen-reactive IgG+ ASC (derived from Bmem cells recirculating in blood) were poorly correlated. Specifically, the R2 values were 0.2535 for the S1 protein (Fig. 2C) and 0.0155 for NCAP (Fig. 3C), respectively. Therefore, in the context of SARS-CoV-2 immunity, plasma antibody levels were poor indicators of the corresponding antigen-reactive memory B cell pool sizes. These measurements were made on blood samples collected within months after recovery from COVID-19. As antibody titers are lower after mild- compared to severe SARS-CoV-2 infection [66, 67], and additionally wane over time [7–9], while Bmem cell frequencies are thought to be more stable over time, one might expect the discordance to grow as time passes, but longer-term longitudinal studies will be needed to address this point. Irrespective of the outcome of those future studies, however, the main message of this communication is likely to hold up: detection of Bmem cells themselves is likely to be a more sensitive and reliable indicator of immune exposure and memory potential than standard measurements of antibody reactivity alone, and antibody levels are poor indicators of memory B cells frequencies in any given individual.
3.5. Discordance between influenza virus and EBV antigen-reactive memory B cell frequencies and antibody levels
In the next set of data, we aimed to establish whether the findings reported above for SARS-CoV-2 are unique to this viral infection and/or the circumstances of our testing. The IgG+ B cell response induced by SARS-CoV-2 infection might be unique due to the immune evasion strategies of this virus . But even if that is not the case, the dissociation between plasma antibody levels and Bmem cell frequencies could just be a feature of (a) a primary B cell response seen (b) shortly after recovery from a mild infection with a virus that (c) typically is cleared within 2 weeks, as these all apply to the COVID-19 cohort tested above. Studying influenza-reactive B cell immunity might therefore provide further insights in this regard. The influenza virus is also cleared by the immune system (with the rare exceptions of severe infections) within 2 weeks post-infection. However, the time-point of the primary infection likely lies in the distant past, years or even decades ago, and re-infections causing secondary B cell responses can be assumed for most adults. As seen in Figs. 4A-C, the correlation between plasma antibody levels and Bmem cell frequencies were only marginal for three representative influenza virus strains, with R2 values of 0.1896 for CA/09 (H1N1), 0.0789 for TX/12 (H3N2), and 0.0913 for Phuket/13 (FluB) rHA antigens, respectively. These data suggest that the observations reported above for SARS-CoV-2 are neither unique to this virus, nor related to the early time-point of measurements, but might be a more universal feature of the B cell response following (at a minimum) respiratory tract infections.
Infection with EBV represents yet a fundamentally different immunological scenario. The primary infection occurs mostly during young adulthood (EBV infection is also called “student kiss fever”), and is systemic. Furthermore, unlike SARS-CoV-2 and influenza, EBV is not cleared from the body but persists lifelong with occasional re-activation episodes, and EBNA1 is expressed even in latency . Thus, the B cell response to EBNA1 can be considered a prototype for that occurring in the immune scenario of systemic antigen persistence. As seen in Fig. 4D, the correlation between EBNA1-reactive IgG antibody levels in plasma and IgG+ memory B cell frequencies was only marginal in healthy adult donors, with an R2 value of 0.1732. Also important for immune diagnostic purposes, like with SARS-CoV-2 and influenza, several subjects exhibited low levels of plasma antibody reactivity against the EBNA1 protein, a finding that when viewed in isolation could be suggestive of either a lack of virus exposure or for developing a weak B cell response to the exposure, yet many of these donors possessed Bmem cells in high frequency. Collectively, the findings made for EBNA1 further support the notion that monitoring the B cell memory compartment itself can be a more sensitive and reliable indicator of immune exposure and memory potential than standard measurements of antibody reactivity alone.
3.6 High reproducibility of total and antigen-reactive IgG+ ASC frequency measurements
While we had limited numbers of PBMC available for most subjects in the COVID-19 cohort, cells obtained from pre-COVID-19 era donors were available in sufficient quantities to assess the reproducibility of either total or antigen-reactive IgG+ ASC frequency measurements. We tested replicate aliquots of PBMC cryopreserved from the same blood draw in two separate experiments. Thus, except for the cryopreserved cell material being the same, all other steps of the testing process were independent assay variables, including (a) thawing and washing of the cryopreserved cells, (b) live/dead cell counting and adjusting the PBMC concentration to 2 x 106 cells/mL for the (c) subsequent polyclonal stimulation during the five day cell culture period, followed by renewed live/dead counting of the PBMC and re-adjusting their concentration for being plated into (d) the FluoroSpot assays (total or antigen-specific). The (e) visualization and (f) counting of the SFU also represent potential inter-assay variables. In this way, for each pre-COVID-19 era donor, frequencies for (g) total IgG+ ASC, as well as (h) the frequency of the antigen-reactive IgG+ ASC were determined, permitting to calculate “% antigen-reactive IgG+ ASC relative to total IgG+ ASC”. Suppl. Figure 6C summarizes the results obtained from repeat testing of total IgG+ ASC frequencies using replicate vials of PBMC while Suppl. Figures 7A-C depict the reproducibility in antigen-reactive IgG+ ASC frequencies established against the CA/09 rHA, TX/12 rHA and EBNA1 proteins. Collectively, these data showed that, in spite of the multitude of potential assay variables, the frequencies of ASCs reactive with these antigens reproduced closely.
3.7. Discussion of the mechanism underlying discordance and implications
Our current understanding of the fate decision checkpoints determining whether an activated B cell will join the Bmem cell compartment or differentiate into an antibody-secreting PC cell is the subject of several excellent reviews [1, 16, 70–73]. While many parameters can contribute to B cell fate decisions after antigen encounter, in the following we will focus on just two mechanisms that could account for, or contribute to the discordance between circulating antibody levels and antigen-reactive Bmem cell frequencies.
First, limitations on PC versus Bmem cell survival might account for our finding. PC can be classified into two types; those that are short-lived and contribute to antibody titers only acutely, and those that are long-lived PC (LLPC) and provide sustained antibody production for decades and potentially the lifetime of an individual [74–76]. Presently there is an incomplete understanding of what distinguishes the ability of a LLPC to survive relative to a short-lived ASC . Importantly, LLPCs are not intrinsically long-lived and instead their survival is dependent on the acquisition of a distinct transcriptional profile and their ability to access specialized and pro-survival niches such as those existing in the BM. Furthermore, there is likely only a finite number of suitable niches in such anatomical locations in which PCs can take up residence and acquire longevity through their intimate interactions with stromal cells and receipt of pro-survival cues [2, 77, 78]. As such, a plausible explanation for the reduced levels of antigen-reactive circulating antibody detected in many of the subjects investigated in this communication is that GC-derived ASCs were in fact generated as a consequence of virus infection, but failed to successfully take up long-term residence in the BM. Bmem cells, in contrast, do not need to compete for such niches for their survival. While in theory the ImmunoSpot® assay would be particularly well-suited to directly test this hypothesis by enumerating in individual subjects’ PC frequencies in BM  versus Bmem cell numbers in the blood, the BM compartment is not readily accessible for routine immune monitoring and thus assessment of LLPC in our donor cohort was not possible. While this mechanism might explain why antibody levels can be low in the presence of abundant Bmem cells, it does not account for the reverse scenario.
The differential, affinity-based selection of mutated B cell subclones along the PC cell versus the Bmem cell differentiation pathways might account for our findings. The GC is an anatomical site in which antigen-activated B cells can undergo multiple rounds of cell division and progressively acquire somatic hypermutations in their BCR’s IgH/IgL. Perhaps most relevant to the interpretation of the data presented here is the well-established notion that the fate decision of a given GC B cell subclone is determined based on the affinity of its BCR for the eliciting antigen [1, 16]. Namely, GC B cell subclones endowed with a low-affinity BCR for the antigen stop proliferating (especially at later stages of the response when antigen becomes limiting) due to insufficient interactions with follicular T helper (TFH) cells, and instead exit the GC reaction and become long-lived Bmem cells. In contrast, GC subclones possessing a high-affinity BCR are instructed by TFH cells to undergo further rounds of proliferation and acquire additional somatic hypermutations [80, 81]. As antigen become increasingly more limiting in the GC reaction, the pressure for selection of high-affinity GC B cell subclones becomes even more stringent while the remainder of B cells with lower affinity for the antigen are shunted towards the Bmem cell pool. This affinity-based selection process, often referred to as affinity maturation, ultimately leads to the differentiation of PC producing antibody with an exquisitely high-affinity for the eliciting antigen and Bmem cells of high, but primarily lower affinity . Consequently, GC-derived PC and Bmem cells represent two related, yet non-overlapping, antigen-reactive repertoires with divergent cell fate decisions based on their BCR’s affinity for antigen.
Therefore, it is assumed that B cell responses that give rise to GC reactions build “two walls of protection against pathogens” . The first wall, encompassed by pre-formed antibodies possessing high affinity afford the advantage of providing instant defense against the homologous pathogen, but on the downside, these antibodies can potentially interfere with the ability to respond to new variants that are antigenically similar [83, 84]. The second wall of defense is comprised by Bmem cells that not only can participate in anamnestic responses against the homologous virus, but also contribute clonally expanded, class-switched, and semi-affinity matured precursor cells that may be reactive with emerging variants of the pathogen and which can undergo further affinity maturation to fine tailor the B cell response to such variants [85, 86].
The above notion of divergent affinity-based selection underlying alternative PC cell versus Bmem cell repertoires has primarily arisen from studies of murine models involving, by necessity, minimalist approaches like the use of BCR-transgenic mice and model antigens . To what extent they are applicable to human B cell responses against viruses in large remains unverified. These murine studies were further enabled by reliance on variable-reducing test conditions, including the homogenous genetic background, age, and sex of the mice studied, their controlled specific-pathogen free environment, defined antigen exposure, and through unfettered access to the lymphoid tissues in which B cell immune responses evolve. As indicated previously, human PC residing in the BM are not readily accessible for general immune monitoring purposes. While human Bmem cells can be readily accessed via blood draws, the ability to study them systematically in larger cohorts awaited an enabling technology. To our knowledge the data communicated here represent the first systematic study comparing the frequency of virus-reactive Bmem cell frequencies with circulating antibody levels in sizable human cohorts. The basic findings we report here, that antibody levels and Bmem cell frequencies are frequently discordant, may not come as a surprise in face of the murine literature, however, the extent of inter-individual variation between serum antibody and Bmem cell frequencies we observed in our human cohorts was, to a degree, unexpected. All of our PBMC donors were healthy young or middle-aged adults and all the B cell responses we studied were induced by, and directed against, viruses that these adults successfully controlled. And yet, in some of these individuals, memory B cell frequencies were remarkably high while antibody titers were surprisingly low, and vice versa. We postulate here that there might be an evolutionary pressure behind developing such discordant phenotypes.
With fundamental pros and cons for pre-existing, antibody-afforded versus Bmem cell-mediated protection, there could exist an evolutionary advantage for endowing different individuals in the population with the propensity to preferentially rely on one or the other immune defense strategy. While with the viruses we studied here either strategy is compatible with successful defense, it can be envisioned that with some pathogens the fate decisions of B cells could have profound implications on the outcome of the host.
3.8. Concluding remarks
Our data also have fundamental implications for immune monitoring in general. Measuring serum antibodies only, as it is commonly done in clinical trials and in the clinic, provides information only on the abundance and efficacy of the “first wall” of antibody-mediated defense. A better understanding of the “second wall” of defense, encompassing the Bmem potential against both homo- and heterotypic viruses, in contrast, requires direct assessment of the Bmem cells themselves. In support of this notion, we found that elevated frequencies of antigen-reactive IgG+ Bmem cells is a more sensitive measure for revealing past SARS-CoV-2 virus exposure than antibody positivity (see Figs. 2C and 3C, and manuscript in preparation). Towards the practical end, we show the feasibility of high-throughput assessment of antigen-reactive Bmem cells enabling its inclusion into the routine immune monitoring portfolio.