key: cord-0930752-h73zq9qi
authors: Chen, Yuezhou; Zuiani, Adam; Fischinger, Stephanie; Mullur, Jyotsna; Atyeo, Caroline; Travers, Meghan; Lelis, Felipe J.N.; Pullen, Krista M.; Martin, Hannah; Tong, Pei; Gautam, Avneesh; Habibi, Shaghayegh; Bensko, Jillian; Gakpo, Deborah; Feldman, Jared; Hauser, Blake M.; Caradonna, Timothy M.; Cai, Yongfei; Burke, John S.; Lin, Junrui; Lederer, James A.; Lam, Evan Christopher; Lavine, Christy L.; Seaman, Michael S.; Chen, Bing; Schmidt, Aaron G.; Balazs, Alejandro Benjamin; Lauffenburger, Douglas A.; Alter, Galit; Wesemann, Duane R.
title: Quick COVID-19 Healers Sustain Anti-SARS-CoV-2 Antibody Production
date: 2020-11-03
journal: Cell
DOI: 10.1016/j.cell.2020.10.051
sha: e671c6c4c07c547266bc8ae6aee9ae4cafc2712d
doc_id: 930752
cord_uid: h73zq9qi

Antibodies are key immune effectors that confer protection against pathogenic threats. The nature and longevity of the antibody response to SARS-CoV-2 infection is not well defined. We charted longitudinal antibody responses to SARS-CoV-2 in 92 subjects after symptomatic COVID-19. Antibody responses to SARS-CoV-2 are unimodally distributed over a broad range, with symptom severity correlating directly with virus-specific antibody magnitude. Seventy-six subjects followed longitudinally to ∼100 days demonstrated marked heterogeneity in antibody duration dynamics. Virus-specific IgG decayed substantially in most individuals, whereas a distinct subset had stable or increasing antibody levels in the same timeframe despite similar initial antibody magnitudes. These individuals with increasing responses recovered rapidly from symptomatic COVID-19 disease, harbored increased somatic mutations in virus-specific memory B cell antibody genes, and had persistent higher frequencies of previously activated CD4+ T cells. These findings illuminate an efficient immune phenotype that connects rapid symptom clearance to differential antibody durability dynamics.

Coronavirus disease 2019 , caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is a major global threat. COVID-19 shows remarkable heterogeneity spanning from asymptomatic to lethal infections (Wu and McGoogan, 2020; Zhou et al., 2020; Zhu et al., 2020) . There is a critical need to understand the nature of the immune response to SARS-CoV-2 to shed light on requirements and likelihood for durable protective immunity in humans.

Antibodies are secreted effector molecules produced as dimers of immunoglobulin (Ig) heavy (H) and light (L) chain pairs from B-lineage cells and come in various IgH isotypes (e.g. IgM, IgG, IgA). Antibody responses to initial infection can reduce the probability of getting sick from the same pathogen more than once. Upon a firsttime infection, the antibody system can learn to better recognize the pathogen through a process of B cell clonal selection and somatic hypermutation (SHM), then produce these improved versions of antibodies in greater amounts to prophylax for a future encounter by the pathogen.

After primary infection or vaccination, IgG antibody production can be maintained and protect for decades as is the case for diphtheria, varicella-zoster, and measles (Amanna et al., 2007) . Durable antibody responses like these rely on coordinated T and B lymphocyte interactions within lymphoid tissue germinal centers (GCs). Activated B cells within GCs diversify Ig genes through SHM-producing Ig variants which then compete for limiting T follicular helper (T FH ) cell survival through coordinated and organized cellular interactions (Cyster and Allen, 2019; Mesin et al., 2016) . This competition matures the affinity of the antibodies produced by the B cells and facilitates differentiation of these GC-experienced B cells into long-lived plasma cells (LLPCs) and memory B cells, necessary cell types for sustained antibody production and efficient cellular recall responses (Balaz et al., 2019; Weisel and Shlomchik, 2017) . Memory B cells can more efficiently differentiate into antibody J o u r n a l P r e -p r o o f secreting plasma cells upon subsequent pathogen invasion, but pre-formed pathogen-specific antibodies produced from LLPCs represent an additional layer of immune function that can protect from initial invasion. B cells that are activated outside of GCs can also differentiate into memory B cells (Takemori et al., 2014) in addition to shorter-lived versions of antibody-secreting cells such as plasmablasts and short-lived plasma cells (SLPCs).

COVID-19 recovered subjects produce IgGs targeting viral nucleocapsid (N), spike (S), and the S receptor binding domain of spike (RBD), which is of particular relevance for their high likelihood of neutralizing capacity (Premkumar et al., 2020) . However, these antibodies are low magnitude in the majority of mild SARS-CoV-2 infections, with higher levels produced in more severe disease (Long et al., 2020a; Ma et al., 2020; . These low initial antibodies levels have been shown to decline in most individuals (Beaudoin-Bussieres et al., 2020; Grandjean et al., 2020; Isho et al., 2020; Iyer et al., 2020; Long et al., 2020b; Seow et al., 2020) .

While S-reactive antibodies from convalescent patients can potently neutralize SARS-CoV-2, they largely lack evidence of SHM (Ju et al., 2020; Robbiani et al., 2020; Rogers et al., 2020) . The low SHM in SARS-CoV-2reactive memory B cells and weak responses hint at low utilization of the GC process, consistent with reports of primarily extrafollicular (i.e. outside GC) immune responses (Woodruff et al., 2020) and dysregulated GC responses (Kaneko et al., 2020) in subjects with severe COVID-19. In this light, whether natural SARS-CoV-2 infection can lead to sustained antibody responses, and what may influence these responses are critical questions.

To address this, we conducted a longitudinal study of COVID-19 convalescent subjects. We quantified plasma IgG and IgM, as well as the stability of plasma IgG to multiple SARS-CoV-2 antigens among subjects with mostly mild disease over time. We found that the anti-SARS-CoV-2 antibodies were broadly distributed and correlated with symptom severity. While a majority displayed IgG decay, a distinct subset showed sustained levels of anti-SARS-CoV-2-specific IgG levels over the same time frame. This distinct subset showed shorter symptom duration, increased SHM in SARS-CoV2 S-reactive memory B cell antibody genes shortly after symptom resolution, and an increase in frequencies of previously activated CD4 + T cells. These findings suggest a distinct immunophenotype connecting symptomatic disease resolution kinetics and antibody durability dynamics for SARS-CoV-2.

We recruited subjects that recovered from COVID-19 between March 2020 and June 2020 in the Boston MA area. Each case was diagnosed based on symptoms consistent with COVID-19 and confirmatory laboratory testing (91 PCR test-confirmed and 1 antibody test-confirmed (Table S1 ). Five individuals were J o u r n a l P r e -p r o o f hospitalized, while all others recovered at home with mostly mild disease. The 1 st blood sample was collected following primary symptom resolution followed by repeated collections at monthly intervals. Plasma and peripheral blood cells were isolated from each blood sample for analysis.

Quantitative ELISA measurement of plasma anti-SARS-CoV-2 IgG and IgM to N, S, and RBD revealed a 3order magnitude range of virus-specific IgG ( Figure 1A) . Five of the 92 subjects showed no greater IgG prepandemic era controls, consistent with previously-reported positivity rates in mild cases (Harritshoej et al., 2020; Hou et al., 2020; Meyer et al., 2020; Rijkers et al., 2020) . Additionally, most subjects displayed levels of anti-SARS-CoV-2 IgM close to pre-COVID-19 era control plasma levels ( Figure 1B ). Spearman rank-order analysis found significant correlations of anti-SARS-CoV-2 IgG magnitude between both age and self-reported symptom severity, with r values between 0.31 and 0.41 ( Figure 1C ) with no correlations between initial virusspecific IgG level and body mass index (BMI) or symptom duration ( Figure 1C ). Scatter plots for age and severity with antibody magnitude illustrate that these features trend together ( Figures 1D-1I ). All additional correlation scatter plots and an illustration of the range of symptom severity scores are given in Figure S1 . The direct and highly significant correlation of self-reported symptom severity to anti-SARS-CoV-2 IgG magnitude supports the value of self-reported severity as a fair metric between individuals within this cohort as this is an established correlation (Long et al., 2020a; Robbiani et al., 2020) .

We also performed Luminex assays to measure reactivity of specific antibody isotypes (including IgM, IgG1, IgG2, IgG3, IgG4 and IgA) to SARS-CoV-2 antigens as well as antibody interactions with Fc receptors (FcRs) ( Figure 2 ). We compared anti-SARS-CoV-2 IgG and IgM levels between pre-pandemic negative control samples and the first 60 recruited subjects from the COVID-19 convalescent cohort (Figure 2A and B). The Luminex assay confirmed ELISA-based IgG observations and was superior to ELISA at differentiating the low anti-S IgM levels between controls and SARS-CoV-2 infected individuals ( Figure 2B ). We observed that 10 -15% of subjects were negative for anti-N, anti-S and anti-RBD IgG1, consistent with ELISA data ( Figure 2C ).

IgM measurements showed that 20 -49% of COVID-19 recovered subjects could not be distinguished from negative controls by the time of the 1 st blood draw ( Figure 2D ). Strong correlations between anti-N and anti-S or anti-N and anti-RBD IgG1 levels ( Figure 2E and 2F) suggest that some individuals may recover from COVID-19 without measurable antibodies. It is possible these individuals had a false-positive PCR test, but the criteria for the presence of symptoms plus the positive test make false positives less likely.

We used a machine learning latent variable modeling approach to discern covariation features most importantly associated with severity. We separated individuals into low (1-4) or high (5-10) severity groups and performed orthogonal partial least square discriminant analysis (OPLS-DA), in which measurement variance contributing to discrimination between these two categories is accounted for in one latent variable comprising a weighted combination of critical features selected by regularization from the Luminex and ELISA data ( Figure   J o u r n a l P r e -p r o o f 2G). The result suggests that anti-S1 IgG1, anti-RBD IgG titer, anti-N IgG2 and IgG3 measures are most importantly correlative of high symptom severity ( Figure 2H ). This model is 65% accurate at predicting symptom severity based on an area under the curve of the receiver-operator characteristic (ROC) curve analysis ( Figure 2I ). Correlation network analysis ( Figure 2J ) illustrates the additional antibody features that covary with the four key features, representing the most germane immune system processes more broadly.

The Luminex assay also included influenza HA and spike receptor binding domains from 3 cold-causing coronaviruses. Antibodies to these antigens generally did not predict COVID-19 disease outcomes ( Figure S2 ), although IgG to human coronavirus 229E did associate with COVID-19 symptom severity.

We explored antibody decay dynamics by quantifying anti-N, anti-S, and anti-RBD IgG for repeated plasma isolations ( Figure 3 ). Seventy-six subjects donated 2 nd and 3 rd monthly samples (median 39, range 13 -88) such that draw 3 was ~100 days following disease onset (median 109, range 83 -173) ( Figure 3A ). Virusspecific IgG decline occurred in most individuals. We quantified this decline by calculating the quotient of the 3 rd draw IgG level divided by the 1 st draw IgG level for each antigen and deemed this the antibody durability index ( Figure 3B ). Anti-N IgG declined in 88%, anti-S in 72%, and anti-RBD in 74% of the convalescent subjects by their 3 rd blood draw. The median antibody durability index values were 0.49, 0.65 and 0.61 for anti-N, anti-S and anti-RBD IgG, respectively. These findings are consistent with a decaying IgG response in the majority of COVID-19 convalescent subjects.

While the majority of convalescent subjects showed decline, some individuals showed stable or enhanced antibody production over the same time period. We examined this by grouping subjects based on their ~100day (i.e. draw 3-defined) antibody durability indices into "sustainer" (durability index ≥1) and "decayer" (durability index < 1) categories. We noted that most individuals qualifying as IgG sustainers with respect to one antigen also sustained production of IgG specific to the other antigens ( Figure 3C and Table S2 ). Plotting quantitative ( Figure 3D ) as well as draw 1-normalized ( Figure 3E ) antibody levels over the first 3 blood draws showed the trend of stable/increasing antibody levels in sustainers and decreasing antibody levels in decayers.

Plotting the same data as days post symptom onset demonstrated a similar diversity of timing of blood draws between the groups ( Figure S3A and B). In addition, sustainers and decayers do not differ substantially in timing of blood draws, either with respect to symptom onset or symptom resolution ( Figure S3C -F).

Of the 76 individuals who completed 3 longitudinal blood draws, 72 seroconverted. For these 72 individuals, we explored associations between subject and disease characteristics and sustained antibody production ( Figure 3F ) using Spearman correlation analysis. Both anti-S and anti-RBD IgG durability indices significantly correlated inversely with disease symptom duration, with r values of -0.28 and -0.27, respectively. The anti-N IgG durability index showed no significant relationship with symptom duration, suggesting a unique effect of the J o u r n a l P r e -p r o o f anti-S and anti-RBD antibodies. Full scatter plots for the correlation between anti-S ( Figure 3G ) and anti-RBD ( Figure 3H ) durability indices and COVID-19 symptom duration illustrate the association between a shorter disease course and more durable antibody responses. Anti-S IgG durability also correlated inversely with symptom severity (r = -0.27). No significant correlations were found between age or BMI and antibody durability ( Figures 3F and S4A -S).

Direct comparison of symptom duration showed that anti-S IgG sustainers had significantly shorter symptom duration (median 9.5, range 2-34) than anti-S IgG decayers (median 15.5, range 1-49) ( Figure 3I and 3K). Anti-RBD IgG sustainers also had significantly shorter symptom duration (median 10, range 2-17) compared to anti-RBD IgG decayers (median 16, range 1-49). Similar analyses for age, BMI, and symptom severity did not reveal significant differences ( Figure S4T and S4U). Anti-S IgG and anti-RBD IgG sustainers had draw 1 anti-S IgG and anti-RBD IgG spanning the range of levels observed for the entire cohort ( Figure 3D and S4V ). In addition, direct comparisons of initial antibody levels among sustainers and decayers showed no significant differences ( Figure 3J and L) and we saw no significant relationship between initial IgG level to N, S or RBD and disease symptom duration ( Figure 1C ). Despite similar initial blood draw levels, sustainers showed a trend towards greater anti-S IgG ( Figure 3J ) and significantly more anti-RBD IgG ( Figure 3L ) compared to decayers in draw 3, consistent with the notion that a unique feature of sustainers is anti-virus IgG maintenance in contrast to initial magnitude. Together these data indicate that relatively sustained antibody production occurs within individuals with a diverse range of initial IgG magnitude and shortened disease course. While it is unknown what level of measurable functional antibody activity correlates with protection, we explored potential functional consequences of differences in antibody decay dynamics between draw 1 and 3 ( Figure S5 ). A major mechanism of antibody neutralization of SARS-CoV-2 is inhibition of binding of S to angiotensin-converting enzyme 2 (ACE2), the receptor for SARS-CoV-2 (Hoffmann et al., 2020) . ACE2-binding inhibition assays showed that while initial inhibition activity was lower for sustainers than decayers, they became indistinguishable by draw 3 ( Figure S5A and S5B). Relatedly, sustainers had higher ACE2-inhbition durability indices ( Figure S5A and S5B), indicating that overall stability of antibody responses correlated with more stable functional activity. We also measured stability of SARS-CoV-2 neutralization activity using an automated high-throughput pseudovirus neutralization assay as well as conventional pseudovirus neutralization assays ( Figure S5C ). Neutralization titer positively correlated with age and disease severity, similar to overall antibody levels ( Figure S5D ). In contrast to ELISA antibody measurements and ACE2inhibition levels, 50% neutralization titers (NT 50 ) were tightly clustered among all subjects in the 1 st and 3 rd samples with no evidence of differences in magnitude or durability dynamics between sustainers and decayers at these time points ( Figure S5F -I). The constrained ability to distinguish subjects based on differences in NT 50 levels over time was likely due to the small range of values approaching limit of detection, in contrast to other antibody measures ( Figure S5E ). In this regard, substantial NT 50 decline between draws 1 and draws 3 was observed only within individuals with high initial neutralization titers ( Figure S5C ).

To further explore features of the swift-healing antibody-sustainer phenotype, we characterized CD4 + ( Figure   4 ) and CD8 + ( Figure S6 ) T cell populations using a previously established T cell phenotyping strategy (Mathew et al., 2020) . We observed that sustainers had a higher frequency of memory (CD45RA -) CD4 + T cells in both 1 st draw and in their 3 rd draw several months later. No significant differences in CD8 + T cell populations were observed. These data suggest that those that heal quickly from mild COVID-19 harbor differences in CD4 + T cell subsets that persist well past disease resolution.

We also FACS-sorted S-specific single memory B cells and sequenced their Ig genes to assess SHM. We confirmed that S-specific memory cells were not plasmablasts based on CD20 and CD38 expression ( Figure   S7A -C). We selected 12 sustainers and 13 decayers based on similar initial antibody levels ( Figure S7J and K). Clinical and antibody features of these subjects showed they are representative of each group ( Figure S7D -Q) and they had similar frequencies of S-specific memory cells ( Figure 5A and B). We observed significantly higher IgH V gene segment (V H ) mutations in the sustainer clones isolated from the 1 st blood draw ( Figure 5C ).

We found that 19.4% of draw 1 sustainer clones had greater than 15 mutations (top 10 th percentile) in contrast to 4.2% of decayer clones ( Figure 5D ). We also found that sustainer-derived clones with less than 15 mutations also had significantly higher draw 1 mutation frequency ( Figure S7R ). The difference in mutation frequency between sustainers and decayers collapsed by the time of the 3 rd blood draw ( Figure 5C and D) as both groups gained significantly more mutations by draw 3. Light chain V gene (V L ) mutation largely paralleled the V H results ( Figure 5E and F). No differences in V H gene segment usage were observed between the groups ( Figure 6G ). These data suggest that increased virus-specific memory B cell SHM early after recovery may be a unique sustainer feature and that continued evolution of anti-SARS-CoV-2 memory B cells occurs more globally later on in convalescence.

Sixty-eight subjects returned for a 4 th blood draw, of which 64 showed SARS-CoV-2 seroconversion ( Figure 6 ).

The 4 th draw samples allowed us to test whether (i) antibodies remain durable in sustainers and (ii) whether this durability translates to greater antibody levels at ~145 post symptom onset ( Figure 6A ). For this analysis, we continued to divide subjects based on their antibody durability index calculated for blood draw 3 (i.e. draw 3-defined sustainers and decayers) ( Figure 6B ) and measured overall antibody levels, ACE2 binding inhibition, and pseudovirus neutralization for the fourth draw blood samples ( Figure 6C and D). For these draw 3-defined groups, we calculated draw 4 durability indices by dividing a subject's 4 th draw value by their 1 st draw measure.

After a relatively short period of ~1 month between draws 3 and 4, absolute magnitudes of anti-S and anti-RBD IgG had become significantly greater for sustainers compared to decayers. The total antibody durability indices were also significantly higher for sustainers compared to decayers. Additionally, anti-S IgG sustainers had J o u r n a l P r e -p r o o f significantly greater neutralizing antibody at this timepoint ( Figure 6C ) and anti-S and anti-RBD sustainers showed significantly higher ACE2 binding inhibition durability indices in draw 4. We reanalyzed our cohort by grouping subjects as sustainers or decayers based on ratios of draw 4 IgG levels to those in draw 1 (i.e. draw 4-defined sustainers and decayers) to compare to draw 3-defined groups ( Figure   7A ). We found considerable overlap in draw 3-and draw 4-defined sustainers ( Figure 7B ). The key clinical correlation -reduced symptom duration in sustainers -was also observed for draw 4-defined sustainers ( Figure 7C and D). We also observed similar differences in total antibody levels, neutralization and durability indices for draw 4-defined sustainers ( Figure 7E and F) as we did for draw 3-defined sustainers ( Figure 6C and D). Together, these data strengthen the conclusion that sustained production of virus-specific IgG is associated with short disease duration. The data also indicate that despite a large range of initial antibody levels, IgG sustainers end up having greater antibody magnitudes compared to decayers as a whole.

The data presented here show that antibody responses to SARS-CoV-2 in mild disease are broadly distributed following infection and declined substantially in most individuals over time. However, some individuals sustained antibody levels over the same time frame. These IgG sustainers had shorter disease courses despite similar distribution of initial anti-SARS-CoV-2 IgG levels, and their anti-S memory B cells harbored increased levels of SHM shortly after disease resolution. Heterogeneity of sustained and declining antibody titers has been documented in common cold coronaviruses. A small study showed that some individuals after virus challenge produced near peak levels of anti-coronavirus IgG 50 weeks out from virus challenge, while others declined near background signal levels between 10 and 20 weeks (Callow et al., 1990) . A lack of sustained immunity to seasonal coronaviruses has also been shown epidemiologically (Edridge et al., 2020; Kiyuka et al., 2018) . While antibody durability is known to be influenced by virus type (Amanna et al., 2007) as well as host and environmental factors (Hagan et al., 2019) , why certain individuals produce longer-lived antibody responses while others do not is not fully understood.

Anti-SARS-CoV-2 antibody decline towards baseline as seen in many individuals is consistent with a dominant role for short-lived plasmablasts and/or SLPCs as a transient antibody source. This, paired with low diseaseproximal memory B cell SHM in most individuals is consistent with dominant extrafollicular antibody responses in IgG decayers (Woodruff et al., 2020) perhaps due to impaired GCs as seen in individuals with more severe disease (Kaneko et al., 2020) . Why SARS-CoV-2 infection may be associated with predominantly extrafollicular antibody responses and dysregulated GCs remains to be fully elucidated. It is possible that the same pro-inflammatory mediators contributed by innate and adaptive immune cells that drive severe disease in some cases (Lucas et al., 2020) are the same that dysregulate the GC processes as has been shown in severe malaria (Ryg-Cornejo et al., 2016) and intracellular bacterial infections (Popescu et al., 2019) . A similar process could underly GC inhibition in severe COVID-19 (Kaneko et al., 2020) and may underlie declining J o u r n a l P r e -p r o o f antibody levels due to largely extrafollicular B cell activation and/or suboptimal GC responses in most individuals with mild COVID-19 as well.

Low SHM antibodies can be potent neutralizers (Ju et al., 2020; Robbiani et al., 2020; Rogers et al., 2020) , suggesting that the human pre-immune antibody repertoire is not deficient in specificities to neutralizing targets on SARS-CoV-2 spike (Brouwer et al., 2020) . SARS-CoV-2 humoral responses are therefore not constrained by gaps in the repertoire but may be limited by poor mobilization of B cells into the LLPC compartment to produce high titer sustained responses as can be seen in other infections (Amanna et al., 2007) , consistent with potential suppression of optimal GC responses and LLPC differentiation in COVID-19. Notably, SHM levels among sustainer and decayer memory B cells continue to accumulate SHM several months following resolution of infection consistent with recent work showing increased plasma avidity to SARS-CoV-2 in convalescent individuals over time (Piccoli et al., 2020) . Antigen from pathogens can be retained for many months by follicular dendritic cells in secondary lymphoid tissues allowing affinity maturation to proceed in the absence of an active infection. It is possible that a return to a normal inflammatory milieu after resolution of COVID-19 restores GCs that may have been suppressed only during active disease, and selection on retained viral antigen facilitates late accumulation of SHM in decayers. A non-mutually exclusive alternative is that memory B cells that lack SHM may be counter-selected over time independent of continued GC output (Takahashi et al., 1998 ).

An important implication of our data is the possible existence of an efficient SARS-CoV-2 virus "handler" phenotype, defined as individuals who experience swift COVID-19 resolution associated with relatively sustained anti-SARS-CoV-2 IgG production. Based on the requirement of GCs for LLPCs and SHM accumulation, this suggests that the sustainer phenotype may include relatively more optimal coordination of lymphocyte interactions in physiologic GC responses. In this regard, GC dysregulation and short-lived antibody responses may not be inevitable outcomes of COVID-19. Elevated percentages of CD4+ effector memory T cells in those that heal quickly and sustain antibody production adds support to a distinct immunophenotype connected to more rapid resolution of disease. Whether the increased percentage of effector memory T cells in sustainers is a reflection of increased disposition to CD4+ effector memory differentiation, an indicator of past experience, or whether it is related to COVID-19 sequelae in sustainers remains to be uncovered.

The degree to which the efficient healer phenotype is due to intrinsic host differences or prior immune priming also warrants further investigation. Sustainers could represent a subset of individuals with pre-existing memory lymphocytes from a seasonal coronavirus infection that happens to cross-react with SARS-CoV-2. Reactivation of cross-reactive memory cells could prime GC reactions for more optimal function and a recall response may help explain the limited disease duration of sustainers. It is also possible that stimulation of pre-existing memory B cells in this context may be more prone to generate LLPCs. This scenario would also be consistent with the observation of higher early SHM levels in sustainers compared to decayers. Arguing against a priming J o u r n a l P r e -p r o o f model is that analysis of other human coronaviruses did not show correlations with COVID-19 symptom duration. However, because timing of potential priming events may matter, and antibodies to common human coronaviruses may not be sustained, analysis beyond antibody levels will be required to further address this.

As discussed above, increased SHM does not appear to be required for potent antibody neutralization function.

In this light, we think it unlikely that the elevated early SHM in sustainers had a direct causal role in quicker symptom clearance, although we have not ruled this out. Instead, we propose that elevated SHM and durable antibody production in sustainers are sequalae of a superior wholistic immune process that overall was responsible for swifter healing. As a part of the immune process in sustainers, we posit that reduced suppression of optimal GC responses occurred during the original infection resulting in superior LLPC differentiation. Why sustainers heal more quickly and whether this is related to why they were able to sustain virus-specific IgG production, potentially by shielding GCs from infection-mediated dysregulation/suppression, remain to be determined. Understanding the specific drivers of diminished versus enhanced antibody durability in cohorts such as this may provide insights into correlates of protection to SARS-CoV-2 and may elucidate targetable pathways to enhance efficient clearance and more sustained antibody protection.

While population diversity was our aim, most volunteers for this study were adults, female, and Caucasian.

While we have confidence in the biologic conclusions, it will be important for future studies to understand the nature of antibody responses in children and adults from a diversity of backgrounds and races. In addition, while the mild and moderate cases represented a large portion of SARS-CoV-2 symptomatology (Oran and Topol, 2020; Wu and McGoogan, 2020) , severe and asymptomatic cases are not represented here. In this context, our work may also provide a framework to address whether the features of increased SHM coupled with quicker symptom resolution and relatively sustained anti-SARS-CoV-2 antibody responses are seen in severe and asymptomatic cases and whether this extends to seasonal coronavirus biology. Whether continued virus replication plays a role in driving sustained responses also remains to be determined, but we feel this is less likely as it does not easily explain the quick healing correlate. In addition, future work will be enhanced by including sample collection before and during COVID-19 allowing direct insights into pre-immune correlates of the sustainer and decayer phenotypes.

Readiness to A.G.S., B.C., G.A., and D.R.W. and Fast Grant funding for COVID-19 science (to D.R.W).

Foundation. Enid Schwartz is also acknowledged. We thank Sudeshna Fisch and Reem Abbaker for support in patient recruitment and sample collection, and Losyev Grigoriy for flow cytometry cell sorting. We thank Noah B. Whiteman and Bruce D. Walker for manuscript review, Graphical abstract generated with BioRender.com. Student's t test was used for significance testing of differences in total antibody levels following a log transformation. All other significance testing used the Mann-Whitney U test. ns is not significant, *p < 0.05, **p < 0.01, ***p<0.001, and ****p < 0.0001. 

total antibody levels represent the positivity cutoff described above and for neutralization the limit of detection.

(F) Plots are as described for (E), with sustainers and decayers as defined by draw 4 anti-RBD IgG. Student's t test was used for significance testing of differences in total antibody levels following a log transformation of the values. All other significance testing for differences in symptom duration used the Mann-Whitney U test. ns is not significant, *p < 0.05, **p < 0.01, ***p<0.001, and ****p < 0.0001. and decayers (black, n=49) as well as anti-RBD (B) sustainers (n=17) and decayers (n=52) described in Figure 3 . The ACE2 inhibition durability index was calculated by dividing the 50% inhibitory concentration (IC 50 ) titer in draw 3 by the IC 50 titer in draw 1 for subjects with titers exceeding negative controls. The dashed line at 2 0 (i.e., 1) indicates stable ACE2-inihibition ability across draws 1 and 3. (C) 50% neutralization titers (NT 50 ) at draws 1 and 3 from an automated, high-throughput (green circles, draw 1 n = 86 and draw 3 n = 55 ) or conventional pseudovirus neutralization assay (gold circles, draw 1 n = 91 and draw 3 n = 76), with the dashed line indicating limit of detection. (D) Spearman correlation analysis correlating conventional draw 1 NT 50 values (Neut. Titer, n = 91) or NT 50 durability index (Neut. durability index, n = 64) and clinical parameters displayed in a grid. For each correlation, the r value is given and significance level are given. Red color intensity indicates strength of positive correlation, intensity of blue indicates strength of negative correlation.

(E) Box and whisker plots illustrating differences in the distributions of values in each antibody measure dataset for the COVID-19 cohort. N IgG, S IgG and RBD IgG are IgG levels as measured by ELISA (n = 92 for draws 1 and 3). ACE2 inh. is the IC 50 titer for ACE2-binding inhibition assay (n = 69 for draw 1, n = 68 for draw 3). HT neut. is NT 50 value as measured by the high-throughput neutralization assay (n = 86 for draw 1, n = 55 for draw 3). Conv. neut. is NT 50 measured by a conventional pseudovirus neutralization assay (n = 91 for draw 1, n = 76 for draw 3). Each value was log transformed and divided by the mean value for that measure. A broader distribution indicates higher variance in the distribution. (F-G) Analysis of differences in high-

Significance testing for all comparisons used the Mann-Whitney U test. Significance is reported in the panels, ns is not significant, *p < 0.05, **p < 0.01, ***p<0.001, and ****p < 0.0001. and decayers (black, n = 12) were analyzed by flow cytometry to ascertain whether spike-binding CD19 + cells (IgM -IgD -CD27 + IgG + Spike + ) are memory cells or plasmablasts in subjects analyzed for SHM. (A) Flow plots illustrating the gating strategy to determine the spike-binding CD19 + cell type are shown on top, with sequential gating shown left to right. IgM -IgD -CD27 + IgG + Spike + cells were analyzed for CD20 and CD38 expression to identify memory cells (CD20 + CD38 int/-) and plasmablasts (CD20 -CD38 Hi ). Bottom plots show a second gating approach to confirm that plasmablasts (CD20 -CD27 + CD38 Hi ) could be identified among PBMCs from the same subjects using this antibody panel. (B) Quantitation of the proportion of IgM -IgD -CD27 + IgG + Spike + memory cells (Mem.) or plasmablasts (PB) among the sustainers and decayers. (C) Quantitation of total plasmablasts as a proportion of all live CD19+ cells for sustainers and decayers. (D-Q) Comparison of (D) symptom duration, (E) age, (F) BMI, (G) severity, (H) timing of initial blood draw relative to symptom onset, (I) timing of third blood draw relative to symptom onset, (J) initial anti-S IgG level, (K) initial anti-RBD IgG level, (L) anti-S durability index, (M) anti-RBD durability index, (N) ACE2-inhibition durability index, (O) conventional neutralization durability index, (P) timing of initial blood draw relative to symptom onset, and (Q) timing of third blood draw relative to symptom onset between the sustainers (red, n = 12) and decayers (black, n = 13) included in the analysis of S-specific memory B cell IgH sequences ( Figure 6 ). (R) Dot plots (left) and box and whisker plots (right) showing mutation numbers per sequence in the VH of sorted S+ single memory B cells with less than 15 V H mutations from sustainers (red, n=12) and decayers (black, n=13). Significance testing for initial antibody levels used Student's t-test on log transformed data. Significance testing for all other comparisons used the Mann-Whitney U test. Significance is reported in the panels, ns not significant, *p < 0.05, **p < 0.01, ***p<0.001, and ****p < 0.0001.

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Duane R. Wesemann. dwesemann@bwh.harvard.edu

No unique materials were generated for this study.

All primary data reported in this study are available at Mendeley Data (DOI: 10.17632/3dxv7mzb76.1)

This study and protocol were approved by the Partners Institutional Review Board. Participants were recruited Blocking was performed for >2 hours at room temperature. During the block step, plasma was thawed at room temperature, combined with an equal volume of 2% Triton in PBS (Thermo Fisher) and incubated for 20 min at room temperature to inactivate enveloped viruses. Dilutions of plasma and standards were made using 1% BSA and 0.05% Tween-20 in PBS as diluent. Block solution was discarded and plates washed with 0.05% Tween-20 in PBS. Plasma dilutions were then applied to the plates and they were placed at 4 o C overnight.

Following the plasma incubation step, plates were washed three times with 0.05% Tween-20 in PBS. For quantitation of plasma IgG or IgM reactive to SARS-CoV-2 antigens, a 2x dilution series was produced from an initial 100x plasma dilution for each sample. A 2x dilution series of either a pooled COVID-19 convalescent plasma or RBD-binding monoclonal IgG1, CR3022, were included on each plate as controls. A standard curve was produced by non-linear regression with Graphpad software from the control values. The Ig level for each unknown sample was determined by interpolation for a single dilution with an OD405 405nm falling in the mid-range of the standards. Where a polyclonal standard was used, antibody levels are reported in reference to the standard as arbitrary units (AU). Where CR3022 was used a standard, antibody levels are reported as CR3022 concentration equivalents (mAb µg/mL equivalents).

Pre-COVID19 plasma from four healthy subjects were included as negative controls on each ELISA plate.

These pre-COVID19 plasma were collected before 10/01/2019.

A custom multiplexed Luminex assay was used to measure relative titer of antigen-specific subclasses, isotypes and Fc-receptors (FcRs), as previously described (Brown et al., 2012) . Briefly, antigen was covalently coupled to carboxyl-beads using EDC (Thermo Fisher) and Sulfo-NHS (Thermo Fisher). Antigen used in the Farber) as well as S1 (Sino Biological) and S2 (Sino Biological) subunits of S. Antigen-coupled beads were blocked, washed with PBS-Tween, resuspended in PBS and stored at 4˚C. Plasma was diluted (1:500 for IgG1, 1:1000 for all Fc-receptors, and 1:100 for all other isotype/subclass readouts). Immune complexes were formed by overnight incubation of diluted plasma and antigen-coupled beads at 4˚C, shaking at 700 rpm. The following day, plates were washed with an automated plate washer (Tecan) with 0.1% BSA 0.02% Tween-20.

Antigen-specific antibody titers were detected with PE-coupled secondaries (Southern Biotech). For FcRbinding, FcRs with an AviTag (Duke Protein Production Facility) were biotinylated with BirA500 kit (Avidity). For detection of FcR binding, FcRs were labelled with PE before addition to the immune complex. Fluorescence was detected using an Intellicyt iQue and analyzed using Forecyt software.

Prior to multi-variate analysis, each Luminex and ELISA-derived feature was box-cox transformed, centered and scaled to unit variance. OPLS-DA was performed on a feature set reduced via elastic net regularization and variable selection. Elastic net minimizes the number of features needed in the model without forfeiting model performance (Zou and Haste, 2005) . All ElasticNet parameters were optimized using a tuneLength = 10 and leave-one-out cross validation. The final reduced feature set included variables selected in 80% or more of 100 rounds of the elastic net algorithm. The R 'ropls' package was used to implement OPLS-DA (orthI =1; PredI=1). An AUC-ROC curve was generated to evaluate the performance of the model using the python 'sklearn.metrics' package. Pairwise spearman's correlation tests were performed comparing each feature in the dataset to the four features included in the final OPLS-DA model. Correlations with coefficients greater than 0.75 and p < 0.1 were inputted into Cytoscape to generate correlation networks. To divide symptoms scores into low and high, we chose 1-4 for low, and 5-10 for high due to the fact that the distribution was bimodal with one peak at 3 and the other at 5, so 4 is a natural dividing point.

PBMCs were isolated by Ficoll (GE Lifesciences) separation from blood. B cells were enriched from total PBMCs using anti-CD19-conjugated magnetic beads (Miltenyi). Flow-through cells were used to analyze T cell populations. Following enrichment, B cells were incubated with 2 µg/mL flag-tagged S protein (Genscript) on ice for 30 min. Cells were then washed and incubated with both APC and PE conjugated anti-Flag antibodies to double stain S-binding B cells. Surface-marker targeting stains were simultaneously applied with the antiflag antibodies. DAPI -IgM -IgD -IgG + CD27 + Spike + cells were single sorted in 96 well plates containing 4 µL/well of ice-cold lysis buffer (0.5x PBS containing 10mM DTT, and 4 U RNAseout). Plates were sealed with AlumaSeal 96 film (Sigma), and immediately frozen on dry ice before transfer to a -80 o C freezer. The RT-PCR and sequencing of Ig transcripts was performed as described (Tiller et al., 2008) . Briefly, single cell RNA was J o u r n a l P r e -p r o o f reverse transcribed with 200 ng random hexamer primer (Thermo Fisher), 1 µL 10mM of dNTP (Promega), 0.7 µL 100 mM DTT, 0.1 µL RNAseout, 0.5% v/v Igepal CA-630 (Sigma), and 50 U superscript III reverse transcriptase (Thermo Fisher). The reverse transcription reaction was performed in a thermocycler using the following program: 42 o C for 10 min, 25 o C for 10min, and 50 o C for 60 min. Subsequently, 2 µL of cDNA was used to amplified Ig sequences by separate heavy, κ light chain and λ light chain-targeting nested PCRs. Each reaction was performed in a 20 µL volume containing 250 nM primer mix, 250 nM dNTP mix (Promega) and 0.5 U HotStar Taq DNA polymerase (Qiagen). Two rounds of amplification were performed for 35 cycles at 94 o C for 30s, 58 o C (IgH/Igκ) or 60 o C (Igλ) for 30 s, 72℃ for 55 s (1st PCR) or 45 s (2nd PCR). The PCR products from the second PCR were sent for Sanger sequencing with reverse primers. The sequences were trimmed based on quality score provided by the service provider (GENEWIZ) and aligned by IgBlastn v.1.16.0 with IMGT annotated germline sequences. To determine somatic hypermutations, productive and in-frame IGHV nucleotide sequences were aligned to their closest germline sequences (Robbiani et al., 2020) .

Inhibition of RBD binding to ACE2 by plasma was measured using a commercial kit supplied by Genscript as described by the manufacturer. In brief, RBD conjugated to HRP was incubated with dilutions of plasma from 1:5, 1:10, 1:30, 1:90, 1:270, to 1:810 in supplied dilution buffer for 30 minutes at 37 o C. 100 µL of plasma and RBD-HRP solutions were then applied to a microtiter plate coated with recombinant ACE2 and incubated at 37 o C for 1 hour. The plates were then washed 4 times with the supplied wash solution and 100 µL of TMB applied to each well. Plates were then incubated in the dark at 25°C for 15 minutes and development quenched using 50 µL of stop solution. Absorbance at 450 nm was then read using a microplate reader (Biotek Synergy H1). Positive and negative controls were provided with the kit. Percent of inhibition was calculated by

(1 -OD value of sample/average OD value of negative controls) x 100 %. Non-linear regression was performed using Graphpad Prism software to determine the 50% inhibitory concentration (IC 50 ) titer. IC 50 values derived from curves with goodness of fit less than 0.7 or IC 50 values less than 1 in initial and follow up blood draws were excluded from further analysis.

Neutralizing activity against SARS-CoV-2 pseudovirus was measured using a single-round infection assay in 293T/ACE2/TMPRSS2 target cells. Pseudotyped virus particles were produced in HEK293T cells (ATCC) by co-transfection of three plasmids: pHDM-SARS2-Spike-delta21, pLenti CMV Puro LUC (w168-1), and psPAX2 (Addgene). The HEK293T cell line stably overexpressing the human ACE2 cell surface receptor protein and TMPRSS2 was kindly provided by Dr. Marc Johnson (University of Missouri School of Medicine). For neutralization assays, serial dilutions of patient plasma samples were performed in duplicate followed by addition of pseudovirus. Plates were incubated for 1 hour at 37 o C followed by addition to target cells (2x10 4 /well). Wells containing cells and pseudovirus (without sample) or cells alone acted as positive and negative infection controls, respectively. Assays were harvested on day 2 using Promega One-Glo luciferase J o u r n a l P r e -p r o o f reagent and luminescence detected with a Biotek Synergy H1. Titers are reported as the plasma dilution that inhibited 50% of infection (NT 50 ), which was determined by non-linear regression using Graphpad Prism.

Samples with NT 50 derived from curves with goodness of fit less than 0.7 were excluded due to technical error unless the lowest dilution (i.e. 1:30) returned values less than 50% neutralization, in which case the NT 50 were included as the detection limit (i.e, 30). Subjects with NT 50 values no more than 30 (the detectable limit) were excluded from durability index calculations.

Lentiviral vector-based pseudovirus neutralization assays were performed as previously described (Crawford et al., 2020) . Briefly, HEK293T cells were transiently transfected with a combination of pHAGE2-CMV-Luciferase-IRES-ZsGreen-W transfer vector and four separate helper plasmids encoding HIV-GagPol, Rev, Tat and the SARS-CoV-2 spike protein. Pseudovirus was collected and frozen for subsequent neutralization assays. Neutralization assays were performed by mixing the indicated dilution of plasma with pseudovirus for one hour, prior to the addition of ACE2-expressing 293T target cells (a gift of Michael Farzan). After incubation at 37C with 5% CO 2 for 60 hours, luciferase signal was measured using a SpectraMAX L instrument. Nonlinear regression was performed using Graphpad Prism software to determine NT 50 . NT 50 values derived from curves with goodness of fit less than 0.7 were excluded from further analysis.

GraphPad Prism 8 software was used for all data analyses excluding multivariate analysis, which is described in detail above. Lognormality of antibody titration data was confirmed by Shapiro-Wilk test and Student's t-test was used for comparisons of log-transformed antibody levels. Single comparisons between other metrics were performed using Mann-Whitney U test and multiple comparisons were performed using Kruskal-Wallis test. For single variate correlation analyses involving continuous and categorical data, Spearman correlation analysis was performed. Measured p values are presented with relevant datasets or described in the text. J o u r n a l P r e -p r o o f

Duration of humoral immunity to common viral and vaccine antigens

Inhibition of Mevalonate Pathway Prevents Adipocyte Browning in Mice and Men by Affecting Protein Prenylation

Decline of humoral responses against SARS-CoV-2 spike in convalescent individuals

Potent neutralizing antibodies from COVID-19 patients define multiple targets of vulnerability

Highthroughput, multiplexed IgG subclassing of antigen-specific antibodies from clinical samples

The time course of the immune response to experimental coronavirus infection of man

Protocol and Reagents for Pseudotyping Lentiviral Particles with SARS-CoV-2 Spike Protein for Neutralization Assays

B Cell Responses: Cell Interaction Dynamics and Decisions

Humoral Response Dynamics Following Infection with SARS-CoV-2. medRxiv

Antibiotics-Driven Gut Microbiome Perturbation Alters Immunity to Vaccines in Humans

Comparison of sixteen serological SARS-CoV-2 immunoassays in sixteen clinical laboratories

SARS-CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven Protease Inhibitor

Detection of IgM and IgG antibodies in patients with coronavirus disease 2019

Persistence of serum and saliva antibody responses to SARS-CoV-2 spike antigens in COVID-19 patients

Persistence and decay of human antibody responses to the receptor binding domain of SARS-CoV-2 spike protein in COVID-19 patients

Human neutralizing antibodies elicited by SARS-CoV-2 infection

Loss of Bcl-6-Expressing T Follicular Helper Cells and Germinal Centers in COVID-19

Human Coronavirus NL63 Molecular Epidemiology and Evolutionary Patterns in Rural Coastal Kenya

Antibody responses to SARS-CoV-2 in patients with COVID-19

Clinical and immunological assessment of asymptomatic SARS-CoV-2 infections

Longitudinal analyses reveal immunological misfiring in severe COVID-19

Serum IgA, IgM, and IgG responses in COVID-19

Deep immune profiling of COVID-19 patients reveals distinct immunotypes with therapeutic implications. Science

Validation of a commercially available SARS-CoV-2 serological immunoassay

Prevalence of Asymptomatic SARS-CoV-2 Infection: A Narrative Review

Mapping Neutralizing and Immunodominant Sites on the SARS-CoV-2 Spike Receptor-Binding Domain by Structure-Guided High-Resolution Serology

TNF-alpha Contributes to Lymphoid Tissue Disorganization and Germinal Center B Cell Suppression during Intracellular Bacterial Infection

The receptor binding domain of the viral spike protein is an immunodominant and highly specific target of antibodies in SARS-CoV-2 patients

Differences in antibody kinetics and functionality between severe and mild SARS-CoV-2 infections

Convergent antibody responses to SARS-CoV-2 in convalescent individuals

Isolation of potent SARS-CoV-2 neutralizing antibodies and protection from disease in a small animal model

Severe Malaria Infections Impair Germinal Center Responses by Inhibiting T Follicular Helper Cell Differentiation

Longitudinal evaluation and decline of antibody responses in SARS-CoV-2 infection

In situ studies of the primary immune response to (4-hydroxy-3-nitrophenyl)acetyl. V. Affinity maturation develops in two stages of clonal selection

Generation of memory B cells inside and outside germinal centers

Efficient generation of monoclonal antibodies from single human B cells by single cell RT-PCR and expression vector cloning

SARS-CoV-2 neutralizing antibody responses are more robust in patients with severe disease

Memory B Cells of Mice and Humans

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

Characteristics of and Important Lessons From the Coronavirus Disease 2019 (COVID-19) Outbreak in China: Summary of a Report of 72314 Cases From the Chinese Center for Disease Control and Prevention

A pneumonia outbreak associated with a new coronavirus of probable bat origin

A Novel Coronavirus from Patients with Pneumonia in China

Regularization and variable selection via the elastic net

We thank study volunteers. This study was supported by NIH grants (T32 AI007245 to J.F, T32 GM007753 to B.M.H. and T.M.C. AI146779 to A.G.S., AI007306 to J.M., AI007512 to A.Z., T32 AI007306 to Y.C., AI142790