key: cord-0328684-09cb217i
authors: Van Ert, Hanora A.; Bohan, Dana W.; Rogers, Kai; Fili, Mohammad; Rojas Chávez, Roberth A.; Qing, Enya; Han, Changze; Dempewolf, Spencer; Hu, Guiping; Schwery, Nathan; Sevcik, Kristina; Ruggio, Natalie; Boyt, Devlin; Pentella, Michael A.; Gallagher, Tom; Jackson, J. Brooks; Merrill, Anna E.; Knudson, C. Michael; Brown, Grant D.; Maury, Wendy; Haim, Hillel
title: Limited variation between SARS-CoV-2-infected individuals in domain specificity and relative potency of the antibody response against the spike glycoprotein
date: 2021-08-05
journal: bioRxiv
DOI: 10.1101/2021.08.04.455181
sha: 5697cbad67348cb5ad8a70d1da02db5e306bc842
doc_id: 328684
cord_uid: 09cb217i

The spike protein of SARS-CoV-2 is arranged as a trimer on the virus surface, composed of three S1 and three S2 subunits. Infected and vaccinated individuals generate antibodies against spike, which can neutralize the virus. Most antibodies target the receptor-binding domain (RBD) and N-terminal domain (NTD) of S1; however, antibodies against other regions of spike have also been isolated. The variation between infected individuals in domain specificity of the antibodies and in their relative neutralization efficacy is still poorly characterized. To this end, we tested serum and plasma samples from 85 COVID-19 convalescent subjects using 7 immunoassays that employ different domains, subunits and oligomeric forms of spike to capture the antibodies. Samples were also tested for their neutralization of pseudovirus containing SARS-CoV-2 spike and of replication-competent SARS-CoV-2. We observed strong correlations between the levels of NTD- and RBD-specific antibodies, with a fixed ratio of each type to all anti-spike antibodies. The relative potency of the response (defined as the measured neutralization efficacy relative to the total level of spike-targeting antibodies) also exhibited limited variation between subjects, and was not associated with the overall amount of anti-spike antibodies produced. Accordingly, the ability of immunoassays that use RBD, NTD and different forms of S1 or S1/S2 as capture antigens to estimate the neutralizing efficacy of convalescent samples was largely similar. These studies suggest that host-to-host variation in the polyclonal response elicited against SARS-CoV-2 spike is primarily limited to the quantity of antibodies generated rather than their domain specificity or relative neutralization potency. IMPORTANCE Infection by SARS-CoV-2 elicits antibodies against various domains of the spike protein, including the RBD, NTD and S2. Different infected individuals generate vastly different amounts of anti-spike antibodies. By contrast, as we show here, there is a remarkable similarity in the properties of the antibodies produced. Different individuals generate the same proportions of antibodies against each domain of the spike protein. Furthermore, the relationship between the amount of anti-spike antibodies produced and their neutralization efficacy of SARS-CoV-2 is highly conserved. Therefore, the observed variation in the neutralizing activity of the antibody response in COVID-19 convalescent subjects is caused by differences in the amounts of antibodies rather than their recognition properties or relative antiviral activity. These findings suggest that COVID-19 vaccine strategies that focus on enhancing the overall level of the antibodies will likely elicit a more uniformly efficacious protective response.

immunoassays that use RBD, NTD and different forms of S1 or S1/S2 as capture antigens to 48 estimate the neutralizing efficacy of convalescent samples was largely similar. These studies 49

suggest that host-to-host variation in the polyclonal response elicited against SARS-CoV-2 50 spike is primarily limited to the quantity of antibodies generated rather than their domain 51 specificity or relative neutralization potency. 52

Infection by SARS-CoV-2 elicits antibodies against various domains of the spike protein, 54

including the RBD, NTD and S2. Different infected individuals generate vastly different amounts 55 of anti-spike antibodies. By contrast, as we show here, there is a remarkable similarity in the 56 properties of the antibodies produced. Different individuals generate the same proportions of 57 antibodies against each domain of the spike protein. Furthermore, the relationship between the 58 amount of anti-spike antibodies produced and their neutralization efficacy of SARS-CoV-2 is 59 highly conserved. Therefore, the observed variation in the neutralizing activity of the antibody 60 response in COVID-19 convalescent subjects is caused by differences in the amounts of 61 antibodies rather than their recognition properties or relative antiviral activity. These findings 62

suggest that COVID-19 vaccine strategies that focus on enhancing the overall level of the 63 antibodies will likely elicit a more uniformly efficacious protective response. 64

cells that express on their surface fusion-competent spike trimers by transfection with an 114 expression plasmid that encodes the full-length protein. Samples were also tested by ELISAs, in 115 which recombinant soluble dimeric forms of the RBD, NTD or the complete ectodomain of 116 S1/S2 (designated Ecto) were used as the capture antigens. The Ecto protein was generated by 117 abrogating the furin cleavage site at spike positions 682-685 (3). Binding of antibodies in serum 118 or plasma to these antigens was measured using a secondary antibody specific for the human 119 kappa light chain, which detects isotypes IgG, IgM and IgA. In addition, we tested the samples 120 with commercial immunoassays that detect IgG against the S1 subunit (Ortho Vitros), S1/S2 121 subunits (DiaSorin Liaison IgG) and a trimeric soluble form of S1/S2 (DiaSorin TrimericS IgG). 122

To quantify non-spike-targeting antibodies elicited against SARS-CoV-2, we used the Roche 123 assay that measures total antibodies against the nucleocapsid protein of SARS-CoV-2. Given 124 that our study focused on quantitative relationships between antibody levels and neutralization 125 efficacies, we excluded from the analyses all samples that were negative for SARS-CoV-2 126 antibodies in at least 5 of the 8 immunoassays. Our final test set was composed of 85 samples 127 (57 serum and 28 plasma). The Ortho test was only performed with the 57 serum samples due 128 to assay incompatibility with plasma. 129

The RBD, NTD and Ecto ELISAs, as well as cbELISA showed a normal distribution of 130 their log 10 -transformed values (see Fig. 1B and results of a Shapiro-Wilk test in Supp. Fig.  131 Interestingly, a strong association was observed between the content of antibodies against the 138 across all IC 50 thresholds. The highest performance was observed for the Ortho, Liaison, Ecto 210 and RBD assays, followed by NTD, TrimericS and cbELISA (Fig. 3C) . Since the Roche assay 211 did not achieve a precision of 0.9, the areas above the curve could not be computed. We then 212 calculated the area above the curve when the required precision was set at levels ranging 213 between 0.75 and 0.95. For most precision requirement levels in this range, the lowest 214 performance was observed for the Roche assay, followed by cELISA, with modestly better 215 performance for the TrimericS and NTD (Fig. 3D) . All other assays exhibited similar 216 performance across the different precision requirements. To determine statistical significance of 217 the differences between performance of any two assays, we performed a permutation-based 218 test (see Materials and Methods section). Briefly, for each pair of assays compared, we 219 measured the area above the curve and calculated the difference. We then permuted for each 220 patient sample the immunoassay identifiers, the area above the curve was recalculated for both 221 immunoassays and the difference determined. The fraction of the times the difference was 222 greater using the permuted values relative to the unpermuted values was calculated as the P 223 value. Significant differences for a one-sided test (P values lower than 0.05) were observed 224 between the cbELISA and all other spike-based assays. The NTD and TrimericS assays 225 showed moderate differences from other assays; however, they were not significant at the 95% 226 confidence level (Fig. 3E) . Therefore, the ability of cbELISA (i.e., the full-length membrane-227 bound form of spike) to predict neutralization was significantly lower than that of all other assays 228 that apply isolated domains of the protein as capture antigens. 229

To independently validate the above findings, we also measured neutralization titers for 230 24 of the serum samples using infectious SARS-CoV-2 under BSL-3 conditions, and correlated 231 those findings with immunoassays values. Virus-induced cytopathology was used to detect 232

infection. The dilution of serum at which cytopathic effects were observed in fewer than 50% of 233 the wells was determined, and data were fit to a regression model to calculate the precise IC 50 234 value. For three of the samples, the IC 50 was not achieved at the lowest dilution of the serum 235 used (1:40); the remainder showed a range of IC 50 values, with a median dilution of 1:212 (Fig.  236   4A) . A strong correlation was observed between the neutralization titers of the sera measured 237 using the replicative SARS-CoV-2 and the VSV-based pseudovirus that contains the spike 238 protein (Fig. 4B) . As expected, IC 50 values in the pseudovirus assay were higher than those 239 measured using infectious virus, since the former measures the dilution at which 50% of virus 240 infectivity is reduced whereas the latter assay measures the dilution at which more than 50% of 241 wells show complete neutralization of all input virus. 242

We compared immunoassay values of the samples with their neutralization efficacies of 243 replicative SARS-CoV-2 (Fig. 4C) . Strong correlations were observed for all spike-based assays 244 (Supp. Fig. S5 ). Precision analyses using an IC 50 threshold of 1:400 demonstrated 245 considerable differences between performance of the assays (Fig. 4D) . Comparison of the 246 overall performance of the immunoassays across neutralization thresholds of 1:50 to 1:500 247 (using the area above the curve metric with a required precision of 0.9) showed a similar pattern 248 to the pseudovirus-based measurements (compare Fig. 4E and Fig. 3C ); the poorest 249 performance was observed for the Roche assay, followed by NTD and cbELISA. All other 250 assays performed similarly well. Comparison of assay performance at precision levels of 0.75-251 0.95 showed modest differences between cbELISA or NTD and all other spike-based assays 252 (Fig. 4F) ; however, these differences did not reach a significance level of 0.05 (Fig. 4G ) 253

Taken together, these results demonstrate that performance of immunoassays based on 254 RBD, S1, or monomeric and dimeric forms of S1/S2 to estimate the neutralization efficacy of 255 each sample was similar. Modestly lower predictive capacities are observed when NTD and the 256 full-length form of spike (as measured by cbELISA) are used as the capture antigens. Further, 257

comparison of the precision of the immunoassays to predict neutralization using pseudovirions 258 containing SARS-CoV-2 spike or replication-competent viruses yield roughly similar findings. 259 260

The above results show that different forms of spike used as capture antigens (NTD, 262 RBD, S1 or Ecto) can estimate neutralization with similar precision. Furthermore, the 263 relationship between the levels of NTD and RBD antibodies is relatively conserved in different 264 individuals; these antibodies compose the vast majority of the antibodies generated against 265 spike. We asked whether the neutralization efficacy increases with higher proportions of RBD-266 or NTD-targeting antibodies (relative to all spike-targeting antibodies). Comparison of the RBD-267

to-Ecto or NTD-to-Ecto ratios with the neutralization efficacy of the samples showed no 268 evidence for a relationship between these variables ( Fig. 5A and 5B) . Similarly, the RBD-to-269 NTD ratio was not associated with the neutralization efficiency of the samples (Fig. 5C) . These 270 findings indicate that convalescent samples with high neutralizing activity do not contain a 271 higher proportion of antibodies that target the RBD or NTD. 272

A large proportion of spike-targeting antibodies elicited by infection are non-neutralizing 273 (36, 37). A recent study has shown that infected and immunized hosts with high levels of spike-274 specific antibodies generate a significantly higher proportion of non-neutralizing antibodies than 275 individuals with lower levels of anti-spike antibodies (38). To explore this relationship in our 276 samples, we implemented a model to examine evidence for a variable ratio between 277 immunoassay values and neutralization efficacy. Two computational approaches were used; the 278 first looks for non-log-linearity in the relationship between neutralization and immunoassay tests, 279

whereas the second considers their rank-ratios and examines evidence for a systematic change 280 over the ranks of the immunoassay results. 281

To compare the variables and avoid a bias related to the dynamic ranges of the values, 282

we corrected the log 10 -transformed immunoassay and neutralization IC 50 values to the same 283 scale by adjustment to a range from 0.1 to 1. For each sample we calculated the ratio between 284 the immunoassay value and the IC 50 value (see analysis of the Ecto ELISA data in Fig. 5D) . 285

This ratio was compared between the 20 samples with the lowest immunoassay values and the 286 20 samples with the highest immunoassay values. Evaluation of these results did not find 287 significantly different ratios in the two groups (see P value for an unpaired T test in Fig. 5D) . A 288 similar lack of a significant difference was observed when the RBD and NTD were used as 289 capture antigens (Supp. Fig. S6) . However, the cbELISA results suggested a higher ratio (i.e., 290 a lower relative neutralization efficacy) for the samples with high antibody levels. 291

To further explore whether the immunoassay-to-neutralization ratio shows any indication 292 of dependence on the immunoassay value, we examined the variability in this ratio by looking 293 for non-linearity in their log-relationship using all 85 samples. The null hypothesis tested was 294 that the log-scale relationship between these variables should be linear, which was tested by 295 considering a quadratic term for immunoassay results in a multiple linear regression (MLR) 296 model. While the data appeared well modelled directly on a log-10 scale, to eliminate concerns 297 about distributional assumptions, the regression coefficient was bootstrapped, and the 298 corresponding 95% confidence interval determined. We first analyzed the results of the Ecto 299 assay. As shown in Fig. 5E , an MLR slope value of 0 (i.e., lack of a quadratic effect, leaving a 300 linear increase in neutralization activity for a given increase in binding) lies within the 95% 301 confidence interval, so we fail to reject the null hypothesis that the variables follow a ratio 302 relationship. Similar analyses of the data from the NTD, RBD and cbELISA tests also failed to 303 show evidence at the 95% level to support a non-linear relationship between immunoassay 304 values and neutralization (Fig. 5F ). 305

We also applied a rank-based approach, whereby immunoassay and neutralization 306

values were transformed to their ranks (from 1 to 85). A simple linear regression (SLR) 307 coefficient was then fitted to the relationship between the immunoassay rank value and 308 immunoassay-to-neutralization rank-ratio, and bootstrapping was applied once more to produce 309 95% confidence intervals. The null hypothesis tested was that a slope of zero exists for this 310 relationship. Again, no evidence was observed to support the notion that the ratio between Ecto 311 values and neutralization varies across different levels of S1/S2-targeting antibodies (Fig. 5G) . 312

A similar bootstrapping analysis of the rank values for the RBD, NTD and cbELISA failed to 313 demonstrate a non-zero slope that would indicate a linear relationship between the two 314 variables (Fig. 5H) . 315

Given the sample size (n=85), the presence of a strong relationship between 316 neutralization fraction and antibody binding activity seems unlikely. Nevertheless, we do 317 observe negative non-significant coefficients for the quadratic effect of log-binding activity on 318 neutralization levels, and positive non-significant coefficients of for the linear relationship 319 between binding activity and the rank ratio of binding to neutralization (Fig. 5, G and H) . Both of 320 these results indicate the plausibility of a weak relationship between the neutralization ratio and 321 binding activity measures, in which higher binding activity could be associated with lower 322

proportional neutralization activity, but the magnitude of such an effect is likely to be limited. 323

Over the course of the COVID-19 pandemic, our understanding of the antibody response 326 against SARS-CoV-2 has evolved. Initial investigations suggested that most neutralizing 327 antibodies elicited by infection or vaccination target the RBD (9, 39). More recent studies have 328 shown a co-dominance of antibodies that target the RBD and NTD (25, 26) . Proteomic 329 deconvolution studies of the IgG repertoire in COVID-19 convalescent patients suggested that 330 the bulk of the neutralizing response targets epitopes outside the RBD (40) . To better 331 understand the target specificity of the response in different individuals, we analyzed the relative 332 level of antibodies against different domains, subunits and oligomeric forms of spike in COVID-333 19 convalescent samples. Our findings suggest the model shown in Fig. 6 . A polyclonal 334 antibody response is elicited in each infected individual against multiple domains of spike. High 335 variation is observed between individuals in the amounts of antibodies generated; however, 336 there is limited variation in the proportion of antibodies against the RBD and NTD (relative to all 337 anti-spike antibodies). Similarly, limited variation is observed in the relationship between the 338 amounts of antibodies against the RBD and NTD, with a ratio ranging between 1 and 3 in 78% 339 of subjects. Importantly, the relative potency of the response (i.e., the level of neutralizing 340 activity relative to the level of antibodies generated) is also constant in different individuals. 341

Thus, the domain specificity and relative inhibitory activity of the response is conserved among 342 individuals, with the main variation being the total amount of the antibodies produced. forms of the protein as the capture antigen; RBD exhibits a similar predictive capacity to that of 355 S1 or S1/S2, with only modestly lower performance for NTD. Thus, inclusion of S2 or 356 trimerization of the protein to mimic the native form of spike does not improve the ability to 357 estimate the amount of neutralizing antibodies. In fact, the poorest performance was observed 358 for the full-length, membrane-bound form of the protein measured by cell-based ELISA. The 359 lower predictive capacity of the cbELISA may result from detection of non-neutralizing 360 antibodies that may recognize the native form of spike (37). Alternatively, differential post-361 translational processing of spike in the HOS cells (relative to the human embryonic kidney 293T 362 cells used to produce the recombinant proteins for these assays) may affect antigenicity of this 363 protein (45). 364

We were surprised to discover that subjects with different amounts of spike-specific 365 antibodies contained a constant level of relative potency. Such results contrast with a recent 366 study by Amanat et al., which suggested that convalescent samples that contain high amounts 367 of spike antibodies (as measured by the Mount Sinai Laboratory COVID-19 ELISA IgG Antibody 368 Test) contain a higher proportion of non-neutralizing antibodies that target the full-length 369 ectodomain of spike (38). It should be noted that in their calculations, the authors analyzed the 370 immunoassay-to-IC 50 ratios using the raw values obtained in these tests. Unfortunately, such an 371 approach can introduce a bias if the dynamic ranges of the two variables differ, which may 372 impact the results of the analysis. To address this potential bias, we performed our calculations 373 using ranks and values that were corrected to the same scale. Both approaches showed similar 374 results, whereby the relative potency is constant in different samples, regardless of the amount 375 of anti-spike antibodies generated. Future studies will reveal whether the target specificity of 376 antibodies with neutralizing activity is also constant in different individuals and independent of 377 the robustness of the response. Such studies are of particular importance in vaccinated 378 individuals, to accurately quantify and characterize specificity of the antibody fractions that can 379 protect from infection. 380

All blood donors were screened following the FDA guidance instructions under an 384 institutional review board approved protocol (IRB #202003554). The consent signed by all 385 donors allowed the use of blood samples for research purposes. Donors were identified and 386 screened following FDA guidelines at the time they enrolled. Two study groups were assessed. 387

The first is composed of 57 convalescent serum samples from subjects that had either been 388 confirmed by reverse transcription polymerase chain reaction (RT-PCR) to be SARS-CoV-2 389 positive from a nasopharyngeal swab (n=51) or had signs or symptoms of COVID-19 and were 390 found to be positive by serological testing (n=6). All donors except one had relatively mild 391 COVID-19 symptoms; this donor was hospitalized for one day due to palpitations. Donor 392 screening was performed at least 10 days after resolution of symptoms. At the time of plasma 393 collection, serum samples were collected in serum separator tubes and allowed to clot for at 394 least 30 minutes. Serum was then isolated, aliquoted and stored at -80°C until use. The second 395 study group is composed of convalescent plasma collected from women hospitalized for 396 delivery, who had previously been infected by SARS-CoV-2, as confirmed by a SARS-CoV-2-397 positive PCR (n=7) or positive serology test (n=21). Samples were collected in EDTA-containing 398 tubes, aliquoted and frozen until use at -80°C. using polyethyleneimine (PEI), as previously described (46). Proteins were harvested in 293S 415

ProCDM and purified using Protein A beads. Eluted products were dialyzed against phosphate 416 buffered saline (pH 7.4). All proteins were analyzed by SDS-PAGE and visualized by silver 417 staining to verify their molecular weight and purity. 418 419 ELISA using RBD, NTD and S1/S2 as capture antigens 420

The RBD, NTD and Ecto recombinant proteins were used as capture antigens in an 421 enzyme-linked immunosorbent assay (ELISA). Briefly, proteins were suspended in PBS at a 422 concentration of 25 nM (2 μg/mL of NTD, 1.37 μg/mL of RBD and 5 μg/mL of Ecto) and 423 incubated overnight in protein-binding 96-well plates (PerkinElmer). The next day, wells were 424 washed once with blocking buffer, composed of 140 mM NaCl, 1.8 mM CaCl 2 , 1 mM MgCl 2 , 25 425 mM Tris pH 7.5, 20 mg/ml BSA and 1.1% nonfat dry milk. Serum or plasma samples were 426 diluted 1:500 (vol:vol) in blocking buffer, added to the wells and incubated for 45 min at room 427 temperature. Samples were then washed four times with blocking buffer and a horseradish 428 peroxidase (HRP)-conjugated secondary antibody that targets the kappa light chain of human 429 IgG1 was added (diluted 1:1200 in blocking buffer). After incubation for one hour at room 430 temperature, samples were washed 5 times with blocking buffer and 5 times with washing buffer 431 (140 mM NaCl, 1.8 mM CaCl 2 , 1 mM MgCl 2 and 25 mM Tris pH 7.5). HRP enzyme activity was 432 measured by light emission using Supersignal West Pico Plus chemiluminescence detection 433 reagents with a Synergy H1 microplate reader. 434 435

Binding of serum antibodies to SARS-CoV-2 spike expressed on HOS cells was 437 measured using a previously-described cell-based ELISA system (22, 24 The DiaSorin Liaison SARS-CoV-2 S1/S2 IgG chemiluminescence assay detects IgG 452 against spike subunits S1 and S2. Samples were analyzed according to the manufacturer's 453 to 8 replicate wells for each dilution and cytopathic effects were evaluated over the next 5 days. 475

The median tissue culture infectious dose (TCID 50 ) was used to quantify virus titer, which 476 describes the dilution of the virus at which fewer than half of the replicate wells show cytopathic 477

To measure neutralization, serial two-fold dilutions of the serum samples (ranging from 479 1:40 to 1:2,560) were prepared in DMEM/FCS 2%. Virus was added to the diluted serum at a 480 final concentration of 25 TCID 50 per well. Samples were incubated at room temperature for one 481 hour and added to Vero-E6 cells seeded the day before in 96-well plates (1.5  10 4 cells per 482 well). Six replicate wells were used for each dilution. Cells were then cultured for 4 days at 37°C 483 until infectivity was evaluated. The number of wells in which intact monolayers were present 484 was assessed using an inverted light microscope. The 50% neutralizing titer (IC 50 Twenty-four hours after transfection, cells were infected with a stock of VSV pseudovirus that 498 encodes the firefly luciferase gene in place of the native VSV-G glycoprotein gene and contains 499 the glycoprotein of Lassa virus (5). Six hours later, infected cultures were washed twice with 500 phosphate buffered saline (PBS, pH 7.4) to remove input pseudovirions, and fresh DMEM/FCS 501 2% was added. Media was collected at 24-and 48-hours after infection, the supernatants were 502 filtered through 0.45 μm pore-sized membranes and centrifuged at 5,380  g for 16 hours at 503 4°C. The pellet was resuspended in PBS and centrifuged through a 20% sucrose cushion at 504 134,000  g for 2 hours at 10°C. Pellets containing the pseudoviruses were resuspended in 505 PBS and stored at -80°C until use. 506

For neutralization assays, two-fold serial dilutions of the serum samples were prepared 507 in DMEM/5% FCS, ranging between 1:40 and 1:2,560. Viruses were added to the diluted serum 508 at a concentration calculated to yield between 100,000 and 200,000 relative light units (RLUs) of 509 luciferase activity per well. These values were determined to be within the linear range of virus 510 input versus luciferase activity measured. Vero-E6 target cells were seeded the day before 511 infection in 96-well white opaque flat-bottomed plates (1.5  10 4 cells per well). The virus-serum 512 or virus-plasma mixture was incubated for one hour at 37°C and added to the wells. Six 513 replicate wells were used for each condition. Samples were then incubated for 24 h at 37ºC, 514 after which the media were removed and 35 µl of Passive Lysis buffer (Promega) was added to 515 each well. Luciferase activity was recorded as a measure of viral infection, as previously 516 described (24). Briefly, 100 µl of luciferin buffer containing 15 mM MgSO 4 , 15 mM KPO 4 (pH 517 7.8), 1 mM ATP, and 1 mM dithiothreitol was added to each well, followed by 50 µl of 1 mM d-518 luciferin potassium salt (Syd Laboratories). Luminescence was detected using a Synergy H1 519

Hybrid reader (BioTek Instruments). 520 521

For each immunoassay, we obtained the curve that describes the required percentile of 523 samples for each neutralization threshold to yield a precision of 0.9. The area above the curve 524 was then determined, which describes all percentile-neutralization threshold combinations that 525 yield a precision level higher than the minimum precision of interest (here, 0.9). This metric thus 526 captures the precision of each assay across multiple neutralization thresholds. To test for 527 significant difference between the area above the curve for any two immunoassays, we used a 528 permutation test. The null and alternative hypotheses for a one-sided test can be stated as: where and describe the area above the curve for immunoassays , and , respectively, and 532 is the total number of immunoassays tested. To test the above hypothesis, we first log-533

transformed immunoassay values and standardized them to a scale of 0 to 1: 534

is the vector of values for immunoassay . The difference between the area above the 535 curve for and was then calculated, denoted as . We then performed a permutation test 536 whereby we permuted for each patient sample the immunoassay identifiers and the area above 537 the curve was recalculated for each immunoassay. This process was repeated 1,000 times ( = 538 1, 2, … , 1000). The difference between the areas above the curves for each iteration of the 539 permutation test was defined as . The instances that the permuted value of was greater 540 than or equal to the non-permuted was calculated and expressed as a fraction of the number 541 of iterations performed, which was defined as the P value for testing the null hypothesis. 542 543

In the absence of a universal gold standard, log 10 -transformations appeared reasonable 545 to capture immunoassay values and neutralization activity. Under our null hypothesis, a change 546 in binding activity, log 10 ( ), should be associated with a linear increase in neutralization, log( ) = 0 + 1 log( ) + . A simple way of detecting departures from this model is to look for 549 curvature in the effect of log ( ): log( ) = 0 + 1 log( ) + 2 log( ) 2 + . Any evidence that 2 550 is nonzero will show departure from the hypothesized relationship; for example, if higher values 551 of binding activity produce a diminished change in neutralization efficacy, we would expect 2 to 552 be negative. We therefore fit a multiple linear regression with the outcome variable of log-IC 50 553 and each of the log-scale immunoassay variables in turn as . To avoid any problematic 554 assumptions about the distribution of the error term , the MLR was fit under a bootstrapping 555 procedure, in which 50,000 repeated samples were taken to produce a bootstrap distribution of 556 the parameter estimates. This was used to compute non-parametric 95% confidence intervals 557 for the 2 quadratic effects. 558

In addition to this MLR approach using log-transformed assay values, we conducted a 559 series of rank-based analyses. Rather than focusing on the ratio-relationship directly, we 560 hypothesized that high neutralization values (relative to the sampling distribution) should 561 correspond to high binding values (relative to the sampling distribution), in such a way that the 562 rank-ratios, , should follow a distribution with mean not depending on the binding rank, . 563

This was investigated via a bootstrapped simple linear regression with the rank-ratio of binding 564 to neutralization as the outcome, and the binding rank as the single explanatory variable. Under 565 the null hypothesis, the slope parameter for the binding rank, 1 , should be equal to zero. We 566 again performed 50,000 repeated samples to produce bootstrap distributions and corresponding 567 non-parametric confidence intervals for 1 . 568

We thank all blood donors that contributed samples to this study and Julie Kurt from the 571

Gynecology and Obstetrics at the University of Iowa for assistance in coordinating these 573 studies. We also thank Michelle Sexton of the Iowa State Hygienic Laboratory and Dr. Louis 574

Katz of the Mississippi Valley Regional Blood Center for assistance in conducting the 575 immunoassays. This work was supported by the Department of Pathology at the University of 576

Iowa. DWB was supported by NIH T32 AI007511. KR and HVE were supported by NIH 577 T32GM007337. HVE and NR were supported by NIH R01AI134733 and R21 AI144215 to WJM. 578

The funders had no role in study design, data collection and interpretation, or the decision to 579 submit the work for publication. All corresponding authors had full access to all the data in the 580 study and had final responsibility for the decision to submit the manuscript for publication. 581 Figure S4 . Immunoassay percentiles required to predict neutralization at the 699 indicated thresholds with a precision of 0.9. The shaded area describes the combination 700 between neutralization thresholds and sample immunoassay percentiles that allow prediction 701 with a precision of 0.9 or higher. The Roche test did not achieve a precision of 0.9, and thus a 702 value could not be computed for this assay. 

Characterization and application of monoclonal antibodies against N protein of 721

Antigenicity of the SARS-CoV-2 Spike Glycoprotein

A Multibasic Cleavage Site in the 725

Spike Protein of SARS-CoV-2 Is Essential for Infection of Human Lung Cells

Cryo-EM structure of the 2019-nCoV spike in the prefusion 729 conformation

CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically 733

Proteolytic 735

Activation of SARS-CoV-2 Spike at the S1/S2 Boundary: Potential Role of Proteases 736 beyond Furin

Neutralizing Antibodies against SARS-CoV-2 Identified by High-Throughput Single

Sequencing of Convalescent Patients' B Cells

A noncompeting pair of human neutralizing antibodies block COVID-19 745 virus binding to its receptor ACE2

Diverse immunoglobulin gene usage and convergent 749 epitope targeting in neutralizing antibody responses to SARS-CoV-2

Potent neutralizing antibodies from COVID-19 patients define multiple 757 targets of vulnerability

neutralizing antibodies directed against spike N-terminal domain target a single 762 supersite

Convergent antibody responses to

SARS-CoV-2 in convalescent individuals

Evidence of escape of SARS-CoV-2 variant B.1.351 from natural 778 and vaccine-induced sera

Rapid isolation and profiling of a diverse panel 784 of human monoclonal antibodies targeting the SARS-CoV-2 spike protein

SARS-CoV-2 specific antibody and neutralization 789 assays reveal the wide range of the humoral immune response to virus

A neutralizing human antibody binds to the N-terminal domain 794 of the Spike protein of SARS-CoV-2

Neutralizing and 798 protective human monoclonal antibodies recognizing the N-terminal domain of the 799 SARS-CoV-2 spike protein

Structure-based development of 802 human antibody cocktails against SARS-CoV-2

The Case for S2: The Potential 804

Benefits of the S2 Subunit of the SARS-CoV-2 Spike Protein as an Immunogen in 805 Fighting the COVID-19 Pandemic

2020. S Protein-Reactive IgG and Memory B Cell Production 808 after Human SARS-CoV-2 Infection Includes Broad Reactivity to the S2 Subunit

Infection during Pregnancy in a Rural Midwest All-delivery Cohort and Associated 813 Maternal and Neonatal Outcomes

The lipid membrane of HIV-1 stabilizes the viral envelope glycoproteins 816 and modulates their sensitivity to antibody neutralization

The High Content of Fructose in Human Semen Competitively Inhibits 819

Broad and Potent Antivirals That Target High-Mannose Glycans

Induction of a Tier-1-Like 822

Phenotype in Diverse Tier-2 Isolates by Agents That Guide HIV-1 Env to Perturbation-823

The plasmablast response to SARS-CoV-2

mRNA vaccination is dominated by non-neutralizing antibodies that target both the NTD 829 and the RBD

Potent neutralizing antibodies 833 against multiple epitopes on SARS-CoV-2 spike

Characterization of 100 sequential SARS-CoV-2 836 convalescent plasma donations

Anti-spike, Anti-nucleocapsid and Neutralizing 839

Antibodies in SARS-CoV-2 Inpatients and Asymptomatic Individuals

SARS-CoV-2 neutralization and serology 843 testing of COVID-19 convalescent plasma from donors with nonsevere disease

Neutralizing Antibody and Soluble ACE2 Inhibition of a Replication-849

Competent VSV-SARS-CoV-2 and a Clinical Isolate of SARS-CoV-2

Sex, age, and hospitalization drive antibody responses in a COVID-19 855 convalescent plasma donor population

A vesicular stomatitis virus replicon-based bioassay 857 for the rapid and sensitive determination of multi-species type I interferon

Levels in Coronavirus Disease 2019 Convalescent Patients

Comparative Performance of 867

Prevalent, protective, and convergent IgG recognition of SARS-CoV-2 901 non-RBD spike epitopes

Performance of six SARS-CoV-2 904 immunoassays in comparison with microneutralisation

Assays Estimate Highly Variable SARS-CoV-2 Neutralizing Antibody Activity in 909

Correlation of SARS-CoV-2 neutralizing 911 antibodies to an automated chemiluminescent serological immunoassay

Evaluation of Three SARS CoV-2 IgG Antibody Assays and 915

Correlation with Neutralizing Antibodies

Proteolytic processing of the human 917 immunodeficiency virus envelope glycoprotein precursor decreases conformational 918 flexibility

 656  30  31  12  15  675  231  1  11  169  495  3  124  725  14  599  25  729  261  325