key: cord-0305800-fjj0wygp authors: Galipeau, Y.; Siragam, V.; Laroche, G.; Marion, E.; Greig, M.; McGuinty, M.; Booth, R.; Durocher, Y.; Cuperlovic-Culf, M.; Bennett, S. L.; Crawley, A. M.; Giguere, P.; Cooper, C.; Langlois, M.-A. title: Relative Ratios of Human Seasonal Coronavirus Antibodies Predict the Efficiency of Cross-Neutralization of SARS-CoV-2 Spike Binding to ACE2 date: 2021-07-16 journal: nan DOI: 10.1101/2021.07.16.21260079 sha: d98679924d0124ceca9bff7cfb89518f699461f3 doc_id: 305800 cord_uid: fjj0wygp Antibodies raised against highly prevalent human seasonal coronaviruses (sCoVs), which are responsible for the common cold, are known to cross-react with SARS-CoV-2 antigens. This cross-reactivity prompts questions about their protective role against SARS-CoV-2 infections and COVID-19 disease severity. However, the relationship between sCoV exposure and SARS-CoV-2 correlates of protection have not been clearly identified. Here we performed a cross-sectional analysis of cross-reactivity and cross-neutralization to three SARS-CoV-2 antigens using pre-pandemic serum from four different groups: pediatrics and adolescents (<21 yrs of age), persons 21 to 70 yrs of age, persons older than 70 yrs of age, and persons living with HCV or HIV. We find that antibody cross-reactivity to SARS-CoV-2 antigens varied between 1.6% and 15.3% depending on the cohort and the isotype-antigen pair analyzed. We also demonstrate a broad range of neutralizing activity (0-45%) in pre-pandemic serum that interferes with SARS-CoV-2 spike attachment to ACE2. While the abundance of sCoV antibodies did not directly correlate with neutralization efficiency, by using machine learning methodologies, we show that neutralizing activity is rather dependent on the latent variables related to the pattern ratios of sCoVs antibodies presented by each person. These were independent of age or sex, and could be accurately predicted by comparing the relative ratios of IgGs in sera directed to NL63, 229E, HKU-1, and OC43 spike proteins. More specifically, we identified antibodies to NL63 and OC43 as being the two most important predictors of latent variables responsible for protection, and 229E as being the least weighted. Our data support that exposure to sCoVs triggers various cellular and immune responses that influence the efficiency of SARS-CoV-2 spike binding to ACE2, and may impact COVID-19 disease severity through various other latent variables. Four endemic human sCoVs (229E, OC43, NL63, and HKU1) are highly prevalent worldwide and cause common and recurrent respiratory infections (1-3). A retrospective prevalence study of acute respiratory infections in France found that OC43 is the most prevalent sCoV followed by NL63, HKU1, and finally 229E (4) . While nearly every adult has been exposed and demonstrates humoral responses to one or several of these sCoV (5, 6) , immunity to each specific sCoV wanes over time. A 35- year longitudinal study revealed that reinfections by a same sCoV were regularly observed at 6 and 9 months post-infection, but most were most frequently observed after 12 months (3) . These findings highlight that sterilizing immunity to these viruses is short-lived and that the relative ratios of antibodies against any one of the four sCoVs is highly variable and dependent on the most recent exposure. This study also indicates that exposure to one sCoV does not offer complete protection against infection by the other viruses, but it is unclear how prior immunity affects disease severity. Seasonal CoVs generally share an overall low degree of sequence similarities with SARS-CoV-2 proteins and only the alphacoronavirus NL63 also utilizes ACE2 as an entry receptor (7, 8) . Nevertheless, conserved epitopes on the spike S2 domain are believed to be responsible for antibody and T-cell cross reactivity between sCoVs and other human betacoronaviruses such as SARS-CoV-1, SARS-CoV-2 and MERS (9) (10) (11) (12) . Furthermore, the nucleoprotein (N) is one of the most conserved antigens across all human CoVs (13) . In fact, antibodies against sCoV antigens impede the specificity of serological tests against SARS-CoV-2 when full length S and N antigens are used in the assay (reviewed here: (14) ). While neutralizing activity directed against the SARS-CoV-2 S protein, more specifically the receptor binding domain (RBD), have been extensively described, non-RBD neutralizing epitopes have also been identified on both the S1 and S2 subunits (15) (16) (17) (18) . Furthermore, antibodies can exhibit non-neutralizing effector functions that restrict SARS-CoV-2 infection by binding to viral protein epitopes expressed on the surface of infected cells, including S and N antigens (19) (20) (21) (22) . While antibody and T-cell responses of SARS-CoV-2 convalescent individuals have been well characterized (22) (23) (24) , the influence of pre-existing immunity, both humoral and cellular, from exposure to sCoVs is not yet clearly established. While several recent studies have investigated the impact of prior immunity to sCoVs on SARS-CoV-2 infection, neutralization, and disease severity (1, 6, 9, 11, [25] [26] [27] [28] [29] , there is some discordance in the findings. Some studies have shown weak or no cross-protective antibodies in pre-pandemic blood of donors in vitro (1, 6, 29) , while several other studies have demonstrated various lines of evidence for antibody and T cell cross-neutralization, and even protection against COVID-19 disease severity (9, 23, 27, (30) (31) (32) . Here, we assessed the seroprevalence of sCoVs, along with the neutralization activity and cross-reactivity to three SARS-CoV-2 antigens, in pre-COVID-19 pandemic sera across four study groups. Our results provide insight into population subgroup variations in the seroprevalence of the four sCoVs, antibody isotype-specific cross-reactivity to SARS-CoV-2 spike (S), nucleocapsid (N), the receptor binding domain (RBD) . We also assessed the neutralizing potential of antibodies in pre-pandemic serum that interfere with S binding to the ACE2 receptor. While neutralization of S binding to ACE2 was detected to various degrees, this activity remained low compared to the neutralizing potential resulting from SARS-CoV-2 infection. However, using several machine learning approaches, we were able to identify a functional dependence on prior sCoV exposure that includes both directly measured and latent variables. These enabled us to model neutralization via Gaussian process regression (GPR) of sCoV seroprevalence. In agreement to previous studies, we do not find a direct correlation between sCoV antibodies levels and neutralization. However, we do identify a strong predictive correlation between neutralization of S binding and relative ratios of the different sCoV antibodies. These data support the idea that latent variables associated with sCoV exposure have a predictive and protective role against SARS-CoV-2 infection and possibly disease severity. We acquired human serum and plasma samples, drawn prior to 2019, thus ensuring no previous exposure to SARS-CoV-2. We performed a cross-sectional profiling where a total of 580 patients were grouped into four cohorts based on age and viral infection. Cohorts were: pediatric samples of children and young adults less than 21 years of age (n=193), adults 21-70 yrs of age (n=273), adults greater than 70 years of age (n=64), and persons living with HCV or HIV (n= 50) of whom nine were followed longitudinally (Table 1 ). Nine patients received two longitudinal blood draws and sCoV seroprevalence was assessed at both time points (separated by 6-12 months) for a total of 589 samples. Each cohort was sex-balanced with mean age equivalence across sexes (Table 1) . We first investigated the prevalence of pre-exposure to the four endemic sCoVs in the pre-COVID-19pandemic cohorts (Fig.1) . Although the four sCoVs (HCoV-OC43, HCoV-229E, HCoV-HKU1, and HCoV-NL63) are prevalent worldwide (33) . However, there seems to be a pattern of reinfections by sCoVs that cycles every 12 months (3). We screened pre-COVID-19 serum samples for sCoV anti-S IgG (229E, OC-43, HKU-1, and NL63) using our high-throughput ELISA platform (34, 35) . As a reference to frame the dynamic range of our assays, SARS-CoV-2 anti-S data of confirmed negative (post-pandemic) (n=115) and SARS-CoV-2 PCR-confirmed positive samples (n=43) is presented (Table S1 ). We then compared the levels of IgG antibodies against sCoV S proteins in all four cohorts to the IgG antibody levels observed in . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint confirmed cases of SARS-CoV-2 ( Fig. 1) . Our results show that a majority of pre-pandemic samples were seropositive for IgG antibodies against OC43, NL63, and 229E S proteins ( Fig. 1 and Table 2 ). One major hurdle in estimating the seroprevalence of IgGs against the four sCoVs is the lack of a true negative reference population. To overcome this challenge, we used a small subset of pediatric samples aged from 1 to 2 years old and established the cut-off at two standard deviations of their mean (Fig. S1 ). The advantage of using these samples is that it limits the probability that these individuals were exposed to the virus due to their young age, although there is still the possibility of vertically-transferred IgG from the mother through breastmilk (36) (37) (38) . From these cut-offs, we were able to estimate a relative seroprevalence of sCoVs in the four cohorts. We found that children and adolescents and young adults less than 21 years of age had significantly lower 229E, OC43, and HKU1 sCoV seroprevalence than all other cohorts ( Fig. 1 and Table 2 ). For example, seroprevalence of HKU-1 in persons < 21 yrs of age was about 34.7%, which agrees with Dijkman et al. who found HKU-1 seroconversion in pediatric individuals to be around 36% (39) . Additionally, the total cohort weighted HKU-1 average seroprevalence of 60.7% is in agreement with Severance et al. which calculated the seroprevalence of HKU-1 at 59.2% (40) . Furthermore, this lower seroprevalence in pediatric samples corroborates previous observations where seroconversion from 229E and NL63 exposure occurs on average 3.5 years following birth (41) (Fig. S1 ). Across all cohorts, we found an overall size-weighted seroprevalence of 82.9% for OC43, 82.1% for 229E, and 74.6% for NL63 ( Table 2 ). Our calculated seroprevalence for OC43 and 229E is in line with previous studies (5, 40, 42) . To ensure the specificity of sCoV detection, we tested co-reactivity between the S proteins of each of the sCoV along with an additional antigen (RBD or N) on a subset of samples ( Fig. 2A) . It is expected that upon virus exposure and seroconversion, antibodies are detected against both S and N, and S and RBD antigens of the same virus. Antigen discordance in single-positive samples is likely the result of S crossreactivity to other sCoV. We then compared the proportion of samples that were dual-positive for S and N of OC43 and HKU-1, and S and RBD of 229E (Fig. 2B ). Positive concordance between the spikes and alternative antigens (i.e., N or RBD) for the four cohorts ranged between 67 -76% (Fig. 2B) . These results highlight that there is a relatively high level of cross-reactivity between the S proteins of sCoVs which compromises assay specificity. Therefore, these data indicate that an accurate determination of sCoVs seroprevalence should include more than one antigen to minimize the potential of S protein cross-reactivity. We next assessed the prevalence of IgA, IgM, IgG and IgE binding to SARS-CoV-2 antigens S, RBD, and N in pre-pandemic serum samples ( Fig. 3 and Tables 2 and S2-S7) using our high throughput serological assay (31) . Cut offs were set at 2 standard deviations (2 SD) above the mean values of the presumed negative . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint sample population. Among pre-pandemic cohorts, 4.6% and 4.9% of samples displayed IgG reactivity against the SARS-CoV-2 RBD and S antigens, respectively (Fig. 3G and 3H, and Table 2 ). In contrast, the N antigen appeared to be more cross-reactive, showing a higher number of samples that were positive for IgG cross-reactive antibodies (6.7% -15.3%) ( Fig. 3I and Table 2 ). Given the importance of N as a common target for serological assays (43) (44) (45) , efforts to accurately measure cross-reactivity to this antigen are critical to account for false positive results. This is especially important when considering that the N antigen of the four sCoVs has on average the highest level of sequence homology to the N of SARS-CoV-2, and that a recent infection by a sCoV could exacerbate such cross-reactivity through immunological imprinting (1, 14) . Regardless of their levels, anti-N antibodies are unlikely to offer protection against SARS-CoV-2 entry given that they do not neutralize the virus. Finally, the role of IgE antibodies in mediating immune responses against SARS-CoV-2 infection is currently unknown although IgE antibodies can play a major role in allergic reactions and inflammation (46) . Our results show that cross-reactive sCoV IgE antibodies are rare and are unlikely to have a detectable role in influencing SARS-CoV-2 infections (Fig. 3J -3L, and Table S7 ). We further investigated cross-reactivity of the IgG1, IgG2, IgG3, and IgG4 isotype subclasses against the three SARS-CoV-2 antigens (Fig. S2 , S3, and Tables S2-S7). Our analysis revealed that most of the IgG subclass antibodies in pre-pandemic sera only exhibited low levels of cross-reactivity with SARS-CoV-2 antigens. Detection of cross-reactive IgG3 antibodies is interesting, as it has been shown that SARS-CoV-2 S-specific IgG3 subclass levels increase with COVID-19 disease severity (47) . We further wanted to assess the false discovery rate (FDR) when two or three antigens are used for the analysis. For this, we performed a comparison of all the serology results across three antibody isotypes and three antigens to determine if a positive sample is positive for only one or multiple antigens and/or isotypes (Figs. 5, S3 and Tables 2 and S7). We found that in most cases, a positive sample is restricted to only one isotype and/or one viral antigen. This highlights the importance and value of considering simultaneously more than one antigen for SARS-CoV-2 serological testing. We calculated the FDR for each specific cohort disaggregated by individual antigen-isotype/subtype (Tables S2-S6) , and we also calculated the FDR for each combination of antigen and isotype/subtype (Tables 1 and S7 ). While the IgG FDR for our complete pre-pandemic cohort ranged from 5% to 11% when probing individual antigens, these values dropped to between 0.85% to 1.58% when probing two antigens, and was 0.51% when all three antigens are used to determine seropositivity (Table 2) . Previous studies have reported the presence of cross-reactive antibodies to SARS-CoV-2 S protein in prepandemic blood samples (1, 11, 26, 48, 49) . However, while generally limited information is available . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07. 16.21260079 doi: medRxiv preprint pertaining to levels of cross-neutralization of SARS-CoV-2, some information has been published with regards to the neutralization of S-pseudotyped viruses (1, 29). Given the importance of neutralizing antibodies that target the SARS-CoV-2 S trimer (26, (50) (51) (52) , we evaluated the relative efficacy of sCoV antibodies to inhibit SARS-CoV-2 S-ACE2 interactions using a protein-based surrogate neutralization ELISA (snELISA) assay (48) . Inhibition of S-ACE2 interaction varied from 0-45%, with a single patient showing inhibition level of 89% (<22 yr old, male; epilepsy), with median levels ranging between 18-23% ( Fig. 5 , Table 3 ). Very low neutralizing activity was previously observed by others in a different experimental system that measured infection by S-pseudotyped viruses (1, 29). By contrast, we identified a moderate yet distinct pattern of neutralization between cohorts of pre-pandemic samples. Next we asked if there was an association between sCoV antibody levels with the levels of neutralization observed. Using fuzzy c-means unsupervised clustering, we clustered samples into 10 distinct groups of human sera from 507 of our 580 patients from which we had sufficient material to monitor the degree of inhibition of SARS-CoV-2 S / ACE2 interaction and cross-reactivity with SARS-CoV-2 RBD, S, and N proteins ( Fig 6A) . Subjects stratified independently of age and sex or other prior viral infection (HIV & HCV) (Fig. 6A, Table S8 ). We found that these clusters defined a continuum of SARS-CoV-2 neutralization in patients ranging from 0% inhibition to 45% (Fig. 6B) . Interestingly, this serum sample also displayed against ACE2 that block spike binding could also theoretically contribute to the neutralization given that autoantibodies are detected in epilepsy (42) (43) (44) . Additional neutralization assays using S-pseudotyped viruses are on-going to confirm this observation. We then further sub-divided these patient clusters into three groups: Non-Responders, Partial Responders, and Responders based on their similarity to newly infected SARS CoV-2 patients (Day 1-9 from testing positive) with respect to SARS-CoV-2 neutralization (Table S1 ). Sera from 3% of all subjects (N=16, Cluster 3) did not neutralize binding of SARS CoV-2 S to ACE2 (Non-Responders, Fig. 6B ). Sixty eight percent of all subjects were Partial Responders, exhibiting significantly higher neutralizing activity than Non-Responders, yet significantly lower neutralizing activity than newly infected COVID-19 patients (Fig. 6B ). Twenty-nine percent of subjects were Responders, exhibiting significantly higher neutralizing activity than Non-Responders and comparable activity to that of newly infected COVID-19 patients (Fig. 6B) . We asked what pattern of sCoV reactivity against their respective S proteins discriminated these groups. We found no statistical difference in the levels of antibodies against OC43-S, 229E-S, HKU-1-S, or NL63-S ( Fig. 6C) . Moreover, the vast majority of cross-reactive SARS-CoV-2 antibodies were directed against the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint N antigen (Tables 2 and S7 ). These data suggested that absolute abundance of sCoV anti-S IgG antibodies did not dictate neutralization. To identify critical features underlying neutralization response, we used Pearson's correlation analysis to determine major linear correlations as well as signed distance correlation to obtain information about linear, non-linear and indirect, and more distant associations. Through these analyses, we were able to discern a pattern of seroprevalence to sCoVs that closely associated with the capacity to neutralize S / ACE2 interactions. First, the percent inhibition of S / ACE2 binding in the Responder group positively associated with higher levels of SARS-CoV-2 cross-reactive IgG Spike and IgG RBD antibodies (at a Pearson's correlation above 0.2 and p<0.05), as expected (Fig. 6D ), and additionally with abundances of IgM RBD, IgM S and IgA S antibodies (at a distance correlation level above 0.2 and p<0.05) (Fig. 6E ). Of these five neutralization correlates, abundances of OC43 S IgG associated with higher levels of SARS-CoV-2 crossreactive IgG RBD and IgA NP. Additionally, levels of NL63 S and HKU-1 S IgG positively associated with levels of cross-reactive IgG NP (Fig. 6E ). At lower distance correlation level of 0.17 there was also direct correlation between NL63 and the percent inhibition (not shown). Conversely, abundances of anti-229E S IgG negatively correlated with both SARS-CoV-2 IgM RBD (Pearson's correlation, distance correlation, Fig. 6D ) and IgG RBD (distance correlation, Fig. 6E ). Finally, HKU-1 S IgG levels negatively correlated (via distance correlation) with IgM S and IgG RBD (Fig. 6E ). Taken together, these data suggest that NL63 S and, to a lesser extent, OC43 S antibody levels positively associate with correlates of percent inhibition and 229E and HKU-1 negatively associate with correlates of percent inhibition. These data are consistent with Selva et al. who have shown that elderly individuals with higher sCoV exposure show increased levels of cross-reactive SARS-CoV-2 IgA and IgG responses, while children show elevated SARS-CoV-2 IgM and is associated with less frequent exposures (53) . Here, we show that these responses link to OC43 S abundance. In our analysis, cross-reactive SARS-CoV-2 IgM levels negatively correlated with 229E and HKU-1 anti-S antibodies. Thus, a higher abundance of these antibodies is associated with a decreased neutralizing probability following exposure to these two sCoVs according to our model. Percent inhibition is in turn correlated with anti-SARS-CoV-2 IgM (IgM RBD and IgM Spike), IgA Spike and IgG (IgG Spike and IgG RBD), possibly indicating a different source of protection in our responder group members belonging to different age groups. These results are in excellent agreement with the further statistical analysis of the feature significance for classification, regression to percent neutralization determined using Relieff. Although there was no statistically significant difference in the levels of antibodies against sCoVs between the three groups ( Fig. 6C ), Relieff selection of the most significantly different features related to the continual value of percent neutralization demonstrated that . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. To test this hypothesis computationally, we asked whether we could use only sCoV S reactivity to predict neutralization of the SARS-CoV-2 S /ACE2 interaction (Fig. 6F ). We used a Gaussian process regression (GPR) method to explore the functional relationship between sCoV S antibody levels and the percent inhibition of ACE2. The GPR model, using sCoV antibody levels as predictors, provided excellent agreement with percent inhibition of ACE2 for all three patient groups (Non-responders, Partial Responders and Responders, Fig. 6G ). Because GPR optimizes variable projection, weights, as well as covariance as a function of the latent variables, these results indicate that sCoV antibody measurements, when combined with latent variables, (i.e., variables that are not measured but can be determined as functions of measured values), dramatically improve prediction of percent inhibition of ACE2 interaction. GPR analysis also provides a confidence interval of this prediction, i.e., error level of the prediction (indicated in dashed lines in Fig. 6F ). The significance of a variable in the developed GPR model, measured as the weight of the predictor indicates major significance of NL63 in modeling followed by OC43 and HKU-1. In this analysis predictor weight of 229E is significantly lower. This is once again in agreement with the correlation analyses demonstrating 229E s indirectly negatively correlated with percent inhibition (Fig. 6E ). Taken together, these data link the capacity of pre-pandemic sera to neutralize SARS CoV-2 S / ACE2 binding to both specific patterns of elevated antibody levels against NL63 S and OC43 S as well as additional unknown factors, yet to be identified, conferred by this specific pattern of cross-reactive antibodies to SARS-CoV-2 S (Fig. 6E ). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint Overall, our data has highlighted that there is a very large range in antibody levels against sCoVs. Interestingly, measuring seroprevalence of sCoVs using a single antigen is prone to specificity challenges due to cross-reactivity between sCoV antigens. We show that only 67% to 76% of spike-positive samples were also positive when probed with a second antigen from the same virus (Fig. 2B ). This indicates that we are over-reporting seroprevalence of individual sCoVs. Additionally, while there is no statistical difference in prevalence between adults and elderly individuals, there appears to be a modest decrease in prevalence in younger individuals for 229E, OC43 and HKU-1. A recent paper by Selva et al. also observed that children have fewer exposures to sCoVs than older adults (53) . Exactly how high levels of antibodies against sCoVs or recency of infection by sCoVs may influence COVID-19 disease severity remains unclear. However, our data highlights relatively high levels of cross-reactivity against certain SARS-CoV-2 antigens, with N being the most frequently detected at 11% overall prevalence, while S displayed about 5% prevalence ( Table 2) . These values are comparable to another study that showed 16.2% cross reactivity to N and 4.2% for S (1). Small differences between our two studies can be accounted for by cohort composition that affects certain antigen/isotype pairs disproportionately (Fig. 3, S2 and tables S2-S6). Although some studies have indicated that no detectable SARS-CoV-2 neutralization activity was detected in pre-pandemic blood samples using various types of assays (1, 29, 53) , the snELISA that we used was able to discern small variations from 0% to 45% in S-to-ACE2 binding inhibition. Although snELISAs are not capable of providing the same complete assessment of whole virus entry neutralization as measured when using live replicative virus, this methodology does enable an isolated and focused assessment of one parameter of the neutralization which is the binding of the viral spike to its host receptor. Small variations at this level can have profound implications during an actual infection in a human host that could perhaps be missed using cell lines to assess neutralization with live virus, the current gold standard. Spikepseudotyped lentiviruses are also useful surrogates to study neutralization and viral entry, but lentiviruses bud from the cell surface and likely harbor a different lipid and host molecule compositions in their envelope than SARS-CoV-2 that mostly egress through exocytosis and possibly also through the lysosomal pathways (54, 55) . Given these differences in the biochemical and immunological composition of the envelopes of lentiviruses and betacoronaviruses, neutralization data obtained with pseudotyped lentiviruses should not be considered as definitive. Nevertheless, in agreement with previous studies, we also did not identify a direct link between the abundance of sCoV S antibodies and neutralization of S/ACE2 interactions (Fig. 6C ). However, our data do indicate that cross-reactive IgG against S and RBD of SARS-CoV-2 have a positive correlation with neutralization efficiency, and also to a lesser degree, IgM against S and RBD, and IgA against S ( Fig. 6D and 6E ). An even more interesting observation is how NL63 correlates . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint with cross-reactive N IgGs (Fig. 6D) , a similar observation was also made by another group (30) . While N antibodies are unlikely to be neutralizing, their Fc effector function, their role in CD8+ T cell responses and /or the cytokines and immune factors involved in their production may all have an indirect influence on other antiviral pathways in a live infected host. Despite a small number of studies reporting undetectable cross-neutralization or unlikely protection by sCoVs antibodies (1, 29, 53) , there is a large body of evidence that supports a role for prior exposure to sCoVs with milder COVID-19 disease severity and/or detectable immunity to SARS-CoV-2 (6, 9, 12, 27, 31, 32, 56, 57) . Using machine learning approaches, we present here data that reconcile these differences by providing evidence that it is the relative pattern of prior sCoV antibody levels in sera, and not total levels, that predict neutralization intensity levels. Additionally, GPR analysis shows that latent variables, that have not been identified in this work but that can be inferred from the measures of sCoVs, provide a model that accurately predicts the percent inhibition of S / ACE2 binding. While evidence continues to grow in support of the protective outcome of prior exposure to sCoVs, NL63 in particular, it is well understood that humoral responses are only one arm of the immune system. With latent factors being here identified to exert an influence over S/ACE2 binding, an understanding of how sCoV may diminish the severity of COVID-19 will require a broader investigation. Evidence that crossreactive T cells also play a role in protection is also gaining momentum (31) . In future work, it will be critical to evaluate if our GPR analysis holds true in predicting COVID-19 severity by way of sCoV antibody ratios. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint Additional experimental procedures are described in detail in the Supplementary Material section. Use of human samples for this study was approved by the University of Ottawa Ethics Review Board: Certificates H-04-20-5727, H-04-21-6643 and H-07-20-6009. The current study collected pre-pandemic serum or plasma samples separated into four cohorts for a total of 580 enrolled patients. The cohorts selected for inclusion in this study are described in (Table 1) Table S1 . . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint The ELISA procedure is a modified method of a recent published study (16) . Antibody responses in prepandemic serum samples of all four cohorts were tested for the presence of antibody isotypes (IgA, IgM, IgG, and IgE) and IgG subtypes (IgG1, IgG2, IgG3, and IgG4) using this ELISA binding assay. SARS-CoV-2 antigens S, RBD, and N were diluted in sterile 1X PBS (Multicell #311-010-CL) to 2ug/mL and used to coat a 96-well plates (VWR #62402-959) using 50uL and incubated on a 4 o C shaker overnight. The next day, the plates were washed three times with PBS-T and were subsequently blocked with buffer (PBS-T + 3% non-fat milk powder) for one hour. The patient serum samples were diluted 1/50 in dilution buffer with a final volume of 100uL. Accordingly, the control calibration antibodies were prepared with appropriate dilutions to generate the calibration curve (anti-COVID Ig). The dilutions for each curve were as follows: total IgG and IgG1 1/5000 followed by 1:2 serial dilutions; IgM, IgA, and IgG2 1/4000 followed by 1:2 serial dilutions; and IgG3 and IgG4 1/10,000 followed by 1:2 serial dilutions. Blocking solution on ELISA plates was removed after one hour followed by the addition of 100uL of the diluted serum samples and control antibodies to the appropriate wells. The plates were incubated with samples for two hours at room temperature (RT) on a shaker (700rpm). After two hours, plates were washed three times with 200uL PBS-T followed by the addition of secondary-HRP antibodies (1:3000) diluted in dilution buffer. 50uL of the appropriate diluted secondary antibody-HRP was added to each well and the plates were incubated at RT for one hour of shaking (700rpm). After one hour of incubation, the plates were washed four times and developed using OPD tablets (Sigma, P9187) dissolved in 20mL of water for infection (WFI) (Thermo Fisher A1287301). After ten minutes incubating in the dark, 50µl of 3M HCl (Fisher #7647-01) was added to each well to stop the reaction and the optical density (OD) at 490nm was measured using a BIO-TEK Power Wave XS2 Plate Reader. Wells filled with dilution buffer in place of serum represented background controls and were subtracted from the patient serum values. To establish consistent and unbiased cut-offs, a systematic approach for each combination of antigen and isotype was used. Using the blank subtracted values from the serological assays a preliminary cut off was is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint off value were considered positive (ratio to cut-off equal or higher than 1) and values under were considered negative (ratio to cut-off lower than 1). To establish the seasonal coronavirus cut-offs, a set of pediatric samples aged between 1-2 years old was used as the reference population due to the high prevalence of seropositivity in adults. The same two pass exclusion strategy at 2SD of the mean was used on that reference group as described above. To evaluate potential neutralization activity against SARS-CoV-2 S protein, we carried out a surrogate neutralization ELISA based on (25) . This neutralization assay was performed using 384-well Immuno plates (Thermofisher, 460372) coated with full SARS-CoV-2 S protein diluted in PBS-T (30uL/well of 2ug/mL), followed by an overnight rotating incubation at 4 0 C. The next day, plates were washed two times with 50μl of PBS-T and further incubated with 40uL blocking buffer for one hour on a plate shaker at a speed of 400 rpm. After incubation the plates were washed with PBS-T followed by the addition of pre-COVID-19 serum patient samples diluted 1:4 in dilution buffer (20 μL/well). The plates were further incubated for one hour and washed three times with PBS-T. Biotinylated ACE2 was diluted in dilution buffer to 0.05ng/uL, followed by 20 μL being added to each well (1 ng/well) and a one hour incubation (400 rpm). After incubation the plates were washed three more times with PBS-T followed by incubation with 20uL of 5ng/uL Streptavidin-Peroxidase polymer diluted in dilution buffer (Sigma#S2438) for 1 hr at 400 rpm. After incubation the plates were washed three times with 60uL of PBS-T. 20uL of SuperSignal ELISA Pico Chemiluminescent substrate diluted 1:2 in PBS was added to each well to measure the luminescence (RLU) signal. Neutralizing activity was determined by calculating the % inhibition against SARS-CoV-2 S proteins using the following equation: Data analysis was conducted using GraphPad Prism v9, R and Matlab 2021a (Matworks Inc.). Univariate analysis was performed using one-way or two-way ANOVA where appropriate. Welch corrections were applied to all analyses. Post hoc tests were Games-Howell's multiple comparisons when group sizes were greater than 50 or Dunnett T3 when less than 50. Sample clustering was performed using fuzzy c-means (FCM) clustering method (fcm function running under Matlab). FCM allows each feature to belong to more than one group by providing a degree of membership, "belonging", to each cluster through maximizing proximity between similar features and distance between dissimilar features. FCM is based on the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. Abbreviations: SD, standard deviation . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint Table 3 . Descriptive statistics of the protein-based neutralization assay by cohorts. Median values and range (minimum and maximum values) of the percentages of the inhibition of SARS-CoV-2 spike / ACE2 was calculated by cohorts from the protein based surrogate neutralization assay (snELISA). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. Demographics of the pandemic cohorts are in Table 1 . Dotted lines represent the cut-off values calculated from seronegative pediatric samples (Fig. S1 ). Statistics were one-way ANOVA with Welch correction and Games Howell post hoc multiple comparisons test for pairwise differences. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. The cut-off was calculated at 2 standard deviation using pediatric samples aged 1-2 years as a negative group. Statistics were one-way ANOVA with Welch correction, post-hoc tests were Games-Howell's multiple comparison tests *p<0.05, **p<001, ***p<0.001, ****p<0.0001. B) Concordance between using the respective sCoV spike and alternative to evaluate seroprevalence. Percentages of positive concordance between S and alternative antigens are indicated in each Venn diagram. Venn diagram have been generated using BioVenn online. Supplementary Tables S3 to S7 . Statistics were one-way ANOVA with Welch correction . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint and Games Howell post hoc multiple comparisons test for pairwise differences in A-I where n=193 (<21 yrs of age), n=273 (21-70 yrs of age), n=64 (>70 yrs of age), n=59 (HCV & HIV cohort (50 persons of which 9 were sampled twice 6 or 12 months apart). In panels J-L, serum volumes were sufficient to analyze n=53 (<21 yrs of age), n=68 (21-70 yrs of age), n=12 (>70 yrs of age), n=28 (HCV & HIV cohort). Given sample sizes, statistics were one-way ANOVA with Welch correction and post-hoc tests were Dunnett T3 multiple comparison tests, *p<0.05, **p<001, ***p<0.001, ****p<0.0001. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07. 16.21260079 doi: medRxiv preprint Each cluster (with the exception of cluster 6 composed of a single patient with robust neutralizing activity) was compared to control samples constituting of serum of infected individuals collected 1 to 9 days after testing positive by PCR for SARS-CoV-2 (green asterisks) or to Non-Responders (cluster 8, black asterisks). Statistics were one-way ANOVA with Welch correction for unequal standard deviation, posthoc Dunnett's T3 *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001 green vs SARS-CoV-2, black vs cluster 3. Subjects that failed to inhibit SARS-CoV-2 S/ACE interactions were dubbed Non-Responders (black, Cluster 3). Subjects that exhibited significantly higher neutralizing activity than Non-Responders yet significantly lower neutralizing activity than COVID-19 patients (green) were dubbed Partial Responders . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 16, 2021. Total n = 507*; *One subject from HCV&HIV cohort was weakly clustered in multiple clusters and was not associated with a single group. This sample was excluded from the analysis. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint To determine if the pre-COVID-19-pandemic sera contained IgGs against the four sCoVs, the S proteins of 229E and NL63 were purchased from creative diagnostics (DAGC134-HCoV-229E; DAGC133-NL63), while OC43 and HKU-1 S proteins were produced for the study (see Antigen Production below). Using our High Throughput Serological Platform (Hamilton Microlab Star, Biotek 405Ls) we automated the Elisa protocol described in this paper with slight modifications. The serological assay was performed in 384 high binding plates (Thermo Fisher, 460372) using a volume of 12.5 uL to coat each well. Plates were blocked using 80uL of blocking buffer per well. The concentration of antigen, antibodies, and sample diluents were kept as above, although the volume was reduced to 10uL per well. To develop the plates a Luminescent substrate (Thermo Fisher, 37069) was used and RLU measured by a Biotek Neo2 plate reader. To produce the RBD of SARS-CoV-2 a plasmid encoding the RBD (MN908947) containing amino acid 319-541 with an N-terminal secretory protein sequence was generously given to us by Dr Florian Krammer (Mount Sinai, NYC). The 229E RBD (P15423) was cloned in pCAGGs plasmid with a hexahis Tag (C-term) with secretory signal (N-term) and transfected into 293F cells maintained in Freestyle 293 expression media (Thermo Fisher, 12338018) at 37 o C, 7% CO2, with shaking (125rpm). A total of . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint ExpiFectamine (Thermo Fisher, 14525). Three days post-transfection the cells were spun down at 4000g for 20min at 4 o C and supernatants filtered through a 0.22um stericup vacuum filter (Millipore Sigma, S2GPU10RE). The filtered supernatant was purified using a Ni-NTA resin (Qiagen, 30210), and washed four times with a washing buffer containing 20nM imidazole. The protein of interest was then eluted with three column volumes of the elution buffer containing 234mM of imidazole. The eluted volume was concentrated and buffer exchanged for PBS using a 10kDa Amicon filter (Millipore Sigma, UFC901008). Concentration was measured and integrity verified by SDS-Page. The spike ectodomain (SARS2: MN908947; OC43: AAT84362.1; HKU1: Q0ZME7.1) cDNA constructs with furin site mutated, two stabilizing prefusion proline mutations, the human resistin as trimerization partner and a C-terminal FLAG-(His)6 (SARS2) or FLAG-Twin Strep Tag-(His)6 (OC43 and HKU1) tag were cloned into pTT241 vector and transfected in CHO2353 to generate stable pools. SARS2 spike was purified by IMAC as described previously while OC43and HKU1 spikes were Purified nucleocapsid protein was buffer exchanged in DPBS using a Centripure P100 column, sterilefiltered through 0.2 μm membrane, aliquoted and stored at -80°C. All purification steps were performed at room temperature. Integrity and purity of the purified nucleocapsid was analyzed by SDS-PAGE and analytical size-exclusion high performance liquid chromatography (SEC-HPLC). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 16, 2021. ; https://doi.org/10.1101/2021.07.16.21260079 doi: medRxiv preprint Epidemiology and clinical presentations of the four human coronaviruses 229E, HKU1, NL63, and OC43 detected over 3 years using a novel multiplex real-time PCR method Epidemiology and Clinical Symptoms Related to Seasonal Coronavirus Identified in Patients with Acute Respiratory Infections Consulting in Primary Care over Six Influenza Seasons Prevalence of Antibodies to Four Human Coronaviruses Is Lower in Nasal Secretions than in Serum Cross-reactive serum and memory B-cell responses to spike protein in SARS-CoV-2 and endemic coronavirus infection Coronavirus biology and replication: implications for SARS-CoV-2 Serologic cross-reactivity of SARS-CoV-2 with endemic and seasonal Betacoronaviruses. medRxiv Severe Acute Respiratory Syndrome Coronavirus 2-Specific Antibody Responses in Coronavirus Disease Patients Cross-reactive Antibody Response between SARS-CoV-2 and SARS-CoV Infections Epitope-resolved profiling of the SARS-CoV-2 antibody response identifies cross-reactivity with endemic human coronaviruses Development and Evaluation of a Multiplexed Immunoassay for Simultaneous Detection of Serum IgG Antibodies to Six Human Coronaviruses Humoral Responses and Serological Assays in SARS-CoV-2 Infections Mapping Neutralizing and Immunodominant Sites on the SARS-CoV-2 Spike Receptor-Binding Domain by Structure-Guided High-Resolution Serology N-terminal domain antigenic mapping reveals a site of vulnerability for SARS-CoV-2 Extremely potent human monoclonal antibodies from COVID-19 convalescent patients Neutralization potency of monoclonal antibodies recognizing dominant and subdominant epitopes on SARS Correlates of protection against SARS-CoV-2 in rhesus macaques Longitudinal analysis of humoral immunity against SARS-CoV-2 Spike in convalescent individuals up to 8 months post-symptom onset Heterogeneous antibodies against SARS-CoV-2 spike receptor binding domain and nucleocapsid with implications for COVID-19 immunity Targets of T Cell Responses to SARS-CoV-2 Coronavirus in Humans with COVID-19 Disease and Unexposed Individuals SARS-CoV-2-reactive T cells in healthy donors and patients with COVID-19 SARS-CoV-2-specific T cell immunity in cases of COVID-19 and SARS, and uninfected controls Lack of cross-neutralization by SARS patient sera towards SARS-CoV-2 Cross-Sectional Evaluation of Humoral Responses against SARS-CoV-2 Spike Recent endemic coronavirus infection is associated with less-severe COVID-19 A systematic review of antibody mediated immunity to coronaviruses: kinetics, correlates of protection, and association with severity Absence of Severe Acute Respiratory Syndrome Coronavirus 2 Neutralizing Activity in Prepandemic Sera From Individuals With Recent Seasonal Coronavirus Infection Antibody responses to endemic coronaviruses modulate COVID-19 convalescent plasma functionality Characterization of pre-existing and induced SARS-CoV-2-specific CD8(+) T cells Selective and cross-reactive SARS-CoV-2 T cell epitopes in unexposed humans Recently discovered human coronaviruses SARS-CoV-2 Seroconversion in Humans: A Detailed Protocol for a Serological Assay, Antigen Production, and Test Setup Persistence of serum and saliva antibody responses to SARS-CoV-2 spike antigens in COVID-19 patients Kinetics of maternal antibodies against rubella and varicella in infants Passive transmission and persistence of naturally acquired or vaccineinduced maternal antibodies against measles in newborns Maternal antibodies: clinical significance, mechanism of interference with immune responses, and possible vaccination strategies The dominance of human coronavirus OC43 and NL63 infections in infants Development of a Nucleocapsid-Based Human Coronavirus Immunoassay and Estimates of Individuals Exposed to Coronavirus in a Human Coronavirus NL63 and 229E Seroconversion in Children Accurate serology for SARS-CoV-2 and common human coronaviruses using a multiplex approach Preliminary Identification of Potential Vaccine Targets for the COVID-19 Coronavirus (SARS-CoV-2) Based on SARS-CoV Immunological Studies SARS-CoV-2 S1 and N-based serological assays reveal rapid seroconversion and induction of specific antibody response in COVID-19 patients A comparison of four serological assays for detecting anti-SARS-CoV-2 antibodies in human serum samples from different populations Immunoglobulin E in health and disease Sensitive detection of total anti-Spike antibodies and isotype switching in asymptomatic and symptomatic individuals with COVID-19 A simple protein-based surrogate neutralization assay for SARS-CoV-2 Kinetics of viral load and antibody response in relation to COVID-19 severity Potent neutralizing antibodies against multiple epitopes on SARS-CoV-2 spike Rapid isolation and profiling of a diverse panel of human monoclonal antibodies targeting the SARS-CoV-2 spike protein Systems serology detects functionally distinct coronavirus antibody features in children and elderly de Supplemental Methods SARS-CoV-2 ELISA Antigens and Antibodies To evaluate the functional antibody responses of sCoV antibodies against different SARS-CoV-2 proteins The following secondary and control antibodies were used during ELISA experiments. Secondary antibodies: Anti-human IgG-HRP (NRC anti-hIgG#5-HRP fusion), anti-human IgA-HRP (Jackson ImmunoResearch, 109-035-011), anti-human IgM-HRP (Jackson ImmunoResearch Anti-SARS-CoV-2 S CR3022 Human IgG1 (Absolute Antibody Ab01680-10.0), Anti-SARS-CoV-2 S CR3022 Human IgG2 (Absolute Antibody Ab01680-11.0), Anti-SARS-CoV-2 S CR3022 Human IgG3 (Absolute Antibody Ab01680-12.1), and Anti-SARS-CoV-2 S CR3022 Human Ig4 MHSSALLCCLVLLTGVRA) followed by a Twin-StreptagII -(His)6 -FLAG tag was synthesized by Genscript with codon-optimization for expression in CHO cells.A biotin acceptor peptide sequence (BAP: GLNDIFEAQKIEWHE) was added in-frame at the C-terminus of ACE2. The cDNA was cloned into pTT5 and ACE2-BAP cDNA was expressed by transient gene expression in CHO 55E1 cells as described above with the addition of 5% (w:w) of pTT5-BirA (E. coli biotin ligase) expression plasmid. Clarified culture supernatant harvested at 8 days post-transfection was purified by IMAC on nickel Sepharose excel as described above following manufacturer's instructions. Pooled Strep-Tactin eluate (buffer-exchanged into DPBS) was stored at -80°C The authors would like to thank Anne-Claude Gingras for helpful comments on the manuscript. M.