key: cord-1054206-21q5lqhs
authors: Sasson, J. M.; Campo, J. J.; Carpenter, R. M.; Young, M. K.; Randall, A. Z.; Trappl-Kimmons, K.; Oberai, A.; Hung, C.; Edgar, J.; Teng, A. A.; Pablo, J. V.; Liang, X.; Yee, A.; Petri, W. A.; Camerini, D.
title: Diverse Humoral Immune Responses in Younger and Older Adult COVID-19 Patients
date: 2021-01-13
journal: nan
DOI: 10.1101/2021.01.12.21249702
sha: 918e5f80b7670caecc1a094a4402c0f4c0e9e30d
doc_id: 1054206
cord_uid: 21q5lqhs

We sought to discover links between antibody responses to SARS-CoV-2 and patient clinical variables, cytokine profiles and antibodies to endemic coronaviruses. Serum from patients of varying ages and clinical severity were collected and used to probe a novel multi-coronavirus protein microarray containing SARS-CoV-2 proteins and overlapping protein fragments of varying length as well as SARS-CoV, MERS-CoV, HCoV-OC43 and HCoV-NL63 proteins. IgG, IgA and IgM antibody responses to specific epitopes within the spike (S), nucleocapsid (N) and membrane proteins (M) were higher in older adult patients. Moreover, the older age group displayed more consistent correlations of antibody reactivity with systemic cytokine and chemokine responses when compared to the younger adult group. A subset of patients, however, had little or no response to SARS-CoV-2 antigens and disproportionately severe clinical outcomes. Further characterization of these serosilent individuals with cytokine analysis revealed significant differences in IL-10, IL-15, IP-10, EGF and sCD40L levels when compared to seroreactive patients in the cohort.

Abstract: 15 We sought to discover links between antibody responses to SARS-CoV-2 and patient clinical 16 variables, cytokine profiles and antibodies to endemic coronaviruses. Serum from patients of 17

varying ages and clinical severity were collected and used to probe a novel multi-coronavirus 18 protein microarray containing SARS-CoV-2 proteins and overlapping protein fragments of 19

varying length as well as SARS-CoV, MERS-CoV, HCoV-OC43 and HCoV-NL63 proteins. 20

IgG, IgA and IgM antibody responses to specific epitopes within the spike (S), nucleocapsid (N) 21 and membrane proteins (M) were higher in older adult patients. Moreover, the older age group 22 displayed more consistent correlations of antibody reactivity with systemic cytokine and 23 chemokine responses when compared to the younger adult group. A subset of patients, however, 24 had little or no response to SARS-CoV-2 antigens and disproportionately severe clinical 25

outcomes. Further characterization of these serosilent individuals with cytokine analysis revealed 26 significant differences in IL-10, IL-15, IP-10, EGF and sCD40L levels when compared to 27 seroreactive patients in the cohort. 28 29 Introduction: 30 With cases continuing to rise in the United States as we pass the annual mark since the beginning 31 of the pandemic, the scientific community continues to strive to further characterize the immune 32 response to SAR-CoV-2 infection. COVID-19 leads to a wide range of clinical responses, 33

varying from minor symptoms, effective immune response and viral clearance to major 34 respiratory compromise, significantly uncoordinated immune response and subsequent death (1). 35

Defining antibody responses, both qualitatively and quantitatively, is necessary in characterizing 36 illness severity, assessing treatment strategies and understanding long-term protection after 37 vaccine administration. 38 39

The antibody response to SARS-CoV-2 infection consists of a rise in immunoglobulin-M (IgM), 40 and a simultaneous or nearly synchronous rise in immunoglobulin-G (IgG) within the first 14-20 41 days of infection, plateauing on average about 6 days after seroconversion (2,3). Antibodies 42 targeting the nucleocapsid (N) protein, a 488 amino acid (aa) SARS-CoV-2 internal structure that 43 functions in compaction and protection of the viral RNA genome, and spike (S) protein, a 1273 44 aa protein that functions in fusion of viral to host cell membranes by binding to cellular 45 receptors, have been implicated as the dominant antibodies through the course of infection (4-6). 46

Correlation of antibody levels to severity of disease in previous studies have yielded mixed 47 results, owing to the heterogeneity of immune responses seen in COVID-19 infection (4,7). 48

There is limited data, however, on antibodies to specific epitopes within these viral proteins and 49 their association with disease severity. 50 51

A large cohort study consisting of over 17 million patients identified common patient 52 characteristics and comorbidities as predictors of death from COVID-19. Among these, age was 53 found to be the strongest predictor of poor outcome (8). This, therefore, raises the question of the 54 differences in antibody response to infection between age groups. Prior studies have shown that 55 older age is associated with increased antibody response (9). Other studies suggest that older age 56 promotes uncoordinated interactions between the branches of the adaptive immune response 57 which ultimately leads to poor outcomes (10). This suggests that the wide range of clinical 58

presentations of COVID-19 could be attributed to multiple interactions between the components 59 of the adaptive response which are influenced by patient demographics and comorbidities. 60 61

Given the consistent circulation of endemic coronaviruses in the population, also known as 62 "common cold" coronaviruses, there is interest in the cross-reactivity of antibodies directed to 63 these viruses with SARS-CoV-2 and their subsequent effect on clinical outcomes of COVID-19 64

(11). The endemic human coronaviruses (HCoV) include alpha (HCoV-229E and HCoV-NL63) 65 and beta (HCoV-OC43 and HCoV-HKU1) subgroups, with the latter also made up of B 66

(containing SARS-CoV and SARS-CoV-2) and C (containing MERS-CoV) lineages (11). 67

Whether it be through cross-protection or antibody-dependent enhancement of infection, more 68 studies are needed to determine the immune interaction between responses to endemic 69 coronaviruses and how they affect disease severity from COVID-19. 70 71

We sought to fill some of these knowledge gaps through use of a novel multi-coronavirus protein 72 microarray with its ability to identify antibody responses to small epitopes using various sized 73 viral protein fragments of SARS-CoV-2. Serum from COVID-19 patients were exposed to these 74 arrays with subsequent correlation of relevant clinical data collected from medical records. This 75 microarray also allowed for correlation of the antibody response to SARS-CoV-2 to 76 coronaviruses of other subtypes and lineages. Overall, we looked to further characterize specific 77 antibody responses to SARS-CoV-2, their correlation with other known coronaviruses and their 78 association with patient clinical data. 79 80 . CC-BY-NC 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 January 13, 2021. 

SARS-CoV-2 protein fragments identify higher levels of antigenic reactivity in older adult 101 patients than younger adult patients 102

Both IgA and IgG antibodies showed higher reactivity to the full-length N protein in the older 103 age group when compared to younger age group (Fig. 1) negative controls. Identical regions of interest occurred within the N protein at aa 151-419, S1 aa 120 551-650 and M aa 1-50 in this cohort and were found to have significant differences in antibody 121 response in the older compared to younger patient groups in our cohort. Within the S2 protein, 122 similar reactive regions were detectable in positive patients as in Camerini et al although these 123

were not found to be significantly different between older and younger patients in this cohort: S2 124 aa 51-100, S2 aa 201-350 and S2 aa 451-480. There was, however, a significant difference 125 detected in the IgA response to S2 aa 451-550 between age groups. While this fragment was 126 found to be also reactive to IgG in this cohort, the difference in reactivity between age groups did 127 not reach significance (p=0.082). The IgM response in the S2 aa 251-300 region was also found 128

to have significant differences in reactivity between age groups. Notably, regions within the S2 129 protein have higher sequence homology between SARS-CoV-2 and the endemic HCoVs 130 compared to the S1 region. In comparison, antibody response to full-length N, S1, S2 and RBD 131 proteins by Milliplex analysis did not show significant differences between age groups (Fig. 1A ) 132

Antibody reactivity to antigenic regions were further stratified by ventilator status within the 133 older age group, but this did not reveal any significant differences in responses when analyzed 134 between the three groups (i.e. older ventilated, older non-ventilated and younger patients) ( Fig.  135  S1 ). 136 137 . CC-BY-NC 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 January 13, 2021. ; CC-BY-NC 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 January 13, 2021. 

Signal intensity is shown by color gradients: IgG (grey to blue), IgA (grey to red), and IgM (grey to 148 green). Bar pairs shown with gold outline represent significantly differential antibody binding between 149 older and younger COVID-19 patients, defined as a mean log2 signal intensity ≥ 0.1 in at least one group 150 and a t-test p value ≤ 0.05. The regions of greatest reactivity for each protein are outlined in magenta. The

inner circle bands represent the responses to full-length S1, S2 and N and the receptor binding domain 2A). This demonstrated overall higher reactivity of antibody response among the older patients 161 compared to the younger patients. This was most striking in IgG responses but could also be seen 162

in IgA responses, although with varying consistency in reactivity levels between the two isotypes 163 (i.e. a patient with high IgG levels to a specific SARS-CoV-2 fragment did not always have 164 corresponding high levels of reactivity with IgA and visa-versa). However, antibody reactivity 165 displayed substantial heterogeneity within age and severity groups with some patients showing 166

little to no IgG, IgA and IgM response to all SARS-CoV-2 fragments. Notably, patients without 167 antibody reactivity to protein fragments did have reactivity to proteins of other HCoVs. The 168

serum IgG response to HCoV-OC43 and HCoV-NL63 appeared robust in most patients, but 169

serum IgA and IgM responses to these HCoVs had less signal intensity. 170 171

Linear models were created to observe correlations between patient clinical data and antibody 172 binding to SARS-CoV-2 fragments which were reactive in at least 10% of the population. After 173 adjustment for age, sex and ventilator status, a significant correlation was found between IgG 174 response and days from symptom onset (Fig. 2B) . A region of notable correlation was found in 175 the S1 aa 551-600, which was further supported by significant correlation seen with S1 aa 551-176 650 fragment and S1 aa 501-600 fragment. S2 aa 501-588 was additionally found to have 177 significant correlation with days from symptom onset. There was also a significant correlation 178 found between IgG antibody response and length of hospital stay (Fig. 2C) . As with days of 179 illness, this correlation was found to be most notable regarding the S1 aa 551-600 region, further 180 supported by significant correlation in the S1 aa 551-650 and S1 aa 501-600 fragments. 181

Additional correlation with length of hospital stay was also seen in the S2 aa 501-588 region. 182 183 . CC-BY-NC 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 January 13, 2021. ; https://doi.org/10.1101/2021.01.12.21249702 doi: medRxiv preprint 184 . CC-BY-NC 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 January 13, 2021. ; https://doi.org/10.1101/2021.01.12.21249702 doi: medRxiv preprint Correlations between antibody reactivity to antigenic protein fragments and cytokine/chemokine 219 levels in patient serum samples were then stratified by age group (Fig. 3) . This revealed the same 220 positive and negative correlations consistently represented in the older age group. In contrast, the 221 younger adult patient group demonstrated notable heterogeneity in its correlations with antibody 222 responses. Furthermore, correlations that were significantly positive or negative in the older age 223

group at times showed a reverse correlation in the younger age group. For instance, IL-10 was 224 significantly negatively correlated to IgG response most notably to the N and S2 fragments in the 225 older age group. This, however, was not consistent with the younger age group where 226 correlations between IgG response to these regions and IL-10 are, although variable, mostly 227

positive. However, similar correlations did occur regardless of age group, such as with the 228 significantly positive correlation seen between the IgG response to N aa 200-400 and levels of 229 IL-5. 230

. CC-BY-NC 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 January 13, 2021. ; https://doi.org/10.1101/2021.01.12.21249702 doi: medRxiv preprint 

IgG responses to SARS-CoV-2 S2 protein correlate with IgG responses to homologous 246 endemic human coronavirus S2 proteins 247

We further assessed the correlation of the IgG response to the S2 and N proteins of SARS-CoV-2 248

to that of HCoV-OC43 and HCoV-NL63 (Fig. 4) . There were strong linear correlations seen 249 between antibody reactivity to the S2 protein of SARS-CoV-2 and the S2 proteins of HCoV-250 OC43 and HCoV-NL63 regardless of age or ventilator status (Figs. 4, S3A and S3B ). This was 251 inconsistent with correlations observed between IgG responses to N proteins, which were weakly 252 correlated, in part due to individuals with little or no reactivity (log2 normalized signal intensity 253 < 1.0), that suggested a population with either delayed or negative seroreactivity, henceforth 254 "serosilent" individuals. This was further exemplified through density plots highlighting the 255 differences in bimodal antibody responses between responding and non-responding individuals. 256

Notably these individuals were responsive to S2 and N proteins of endemic HCoVs and appeared 257

to be non-responsive to solely SARS-CoV-2 proteins (Fig. 4) . 258 259 . CC-BY-NC 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 January 13, 2021. ; https://doi.org/10.1101/2021.01.12.21249702 doi: medRxiv preprint 

Positive correlations between S2 responses to SARS-CoV-2 and endemic HCoVs were not 272 observed for IgM, largely owing to the limited IgM response seen to HCoV-OC43 and HCoV-273 NL63 as discussed earlier (Fig. S4A) . While similar correlations as with IgG were seen when 274 . CC-BY-NC 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 January 13, 2021. ; https://doi.org/10.1101/2021.01.12.21249702 doi: medRxiv preprint comparing IgA responses to S2 proteins of endemic HCoVs, this did not extend to N proteins, 275 again, due to the limited IgA response in serum (Fig. S4B) . When comparing the antibody 276 response to individual SARS-CoV-2 fragments to endemic coronaviruses, there were diffusely 277 positive correlations seen between S2 fragments with S2 proteins of endemic coronaviruses, 278 most notably to HCOV-OC43 (Fig. S5 ). 279 280

Given the notable separation seen in non-responding individuals compared to the rest of the 281 cohort, cytokine profiles were then assessed in these patients. As the majority of these 282 individuals were in the older age group, cytokines/chemokines from the older age group were 283 compared (Fig. 5) . We identified four serosilent individuals to both immunodominant SARS-284

CoV-2 proteins, N and S2 (N-, S2-), with an additional two patients that were seronegative to S2 285 protein (S2-) but seroreactive to N protein (N+) (Fig. S6) . Analysis of cytokine/chemokine 286 responses to all six patients revealed significantly higher IL-10, IL-15 and IP-10 in serosilent 287

individuals when compared to the rest of the older age group. It also displayed significantly 288 lower levels of EGF and sCD40L in serosilent individuals (Fig. 5) . 

Discussion: 301 While this study revealed epitopes seen within SARS-CoV-2 proteins, which corresponded to the 302 concurrent study by Camerini et al, it additionally revealed how reactivity differed between age 303 groups and how these same regions correlated with clinical and laboratory data. Although 304 differences between antibody responses did not reach significance when separated by severity 305 (i.e., those ventilated vs those who did not require ventilation), there were significant differences 306 seen between age groups, with higher antibody reactivity apparent in the older age group. We 307 . CC-BY-NC 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 January 13, 2021. ;

were additionally able to see correlations in antibody response to SARS-CoV-2 antigenic 308 fragments and serum cytokine/chemokine profiles along with correlations in reactivity to 309 endemic HCoVs. 310 311

Among the regions of interest highlighted in our study was S1 aa 551-600 region, an area just 312 past the RBD at the C-terminus of the S1 protein, which has also been noted in other recent 313 studies involving SARS-CoV-2 epitope mapping (12,13). In our study, IgG reactivity to this 314 region was not only higher in the older COVID-19 patient cohort but was significantly correlated 315 with hospital length of stay and days of illness. While differences in reactivity did not reach 316 significance regarding ventilator status in this small cohort, the correlations found in this study 317

suggest unfavorable clinical outcomes associated with large levels of antibodies to this region. 318

Previous studies have also shown a correlation between IgG responses to S1 protein and days of 319 illness, which likely can be attributed, at least in some part, to reactivity within this region (14). 320 321

We further highlighted three regions within the S2 protein: S2 aa 51-100, S2 aa 201-350 and S2 322 aa 451-480, which have also been implicated in recent epitope mapping studies (12,13). While 323 small sample size may have limited our ability to see true differences in IgG reactivity within 324 these regions with respect to severity and age, we were able to see a correlation between 325 reactivity within the S2 aa 201-350 and hospital length of stay, highlighting clinical implications 326

to having antibodies to this region. 327 328

When antibody reactivity to SARS-CoV-2 was correlated with cytokine/chemokine profiles in 329 these serum samples, we found consistent correlations existed within the older patient group 330 while large variations occurred in younger COVID-19 patients. This suggests a clinically 331 unfavorable cytokine/chemokine profile that correlates with higher antibody reactivity, which 332 more commonly occurs in older patients. IL-5, a type 2 (Th2) cytokine shown by Lucas et al to 333

correlate with severe COVID-19 disease, was shown to have a significantly positive correlation 334

to antibody response to the N aa 200-400 region, which may further suggest poor outcomes 335 related to Th2 responses (15). Additionally, IL-10 had significant negative correlations to 336 antibody responses to S2 and N proteins in the older age group, which is consistent with its 337 known anti-inflammatory properties. Interestingly IL-10 has been implicated in numerous other 338 viral, bacterial, and protozoal infections whose clinical outcomes were observed to be time-339 dependent of peak IL-10 production and its ability to cause either inhibition of effective 340 pathogen clearance or prevention of excessive immune response to foreign infectious antigens 341 (16). 342 343 We also found strong correlations between the IgG response to SARS-CoV-2, HCoV-NL63 and 344

HCoV-OC43 S2 proteins, which have also been noted in other recent epitope studies (12,13). 345

This has been attributed to considerable sequence homology observed between S2 proteins of 346 SARS-CoV-2 and endemic HCoVs, particularly to the more closely related endemic 347 betacoronaviruses (HCoV-OC43 and HCoV-HKU1). Associations found in this study suggest 348 either cross-reactivity of newly produced antibodies to SARS-CoV-2 with other HCoV antigens 349

in the array or cross-reactivity in which preexisting antibodies to other coronaviruses can 350

recognize SARS-CoV-2 antigens. While this is difficult to determine without analysis of patient 351 serum prior to infection, we see evidence of both phenomena occurring in our cohort. Lack of a 352 concomitant serum IgM response observed in this cohort to endemic HCoVs along with lack of 353 . CC-BY-NC 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 January 13, 2021. ; observed correlation between the IgM response to S2 proteins between them suggests a 354 preexisting, boosted IgG rather than new, acute antibody production. However, the magnitude of 355 reactivity to S2 protein and the positive correlation of anti-S2 IgG between SARS-CoV-2 and 356 endemic HCoVs suggests likely a component of new antibody reactivity to some epitopes due to 357 significant immune activation. Preexisting, cross-reacting antibodies to SARS-CoV-2 would 358 allow an opportunity for cross neutralization of SARS-CoV-2 antigens and raises the possibility 359 of improved clinical outcomes in these patients. This may further explain why children, who are 360 known to have more consistent exposures to endemic HCoVs, may be more protected from 361

severe COVID-19 infection (12). However, of note, in our analysis, correlations of antibodies to 362 S2 proteins between SARS-CoV-2 and endemic HCoVs were apparent regardless of age or 363 ventilator status, possibly suggesting less of an influence on clinical outcomes. 364 365

There was also notably absent antibody reactivity to SARS-CoV-2 proteins among a subset of 366 individuals in this cohort. Given that these patient samples were collected at a single time point, 367

it is difficult to know if these represent patients with no response throughout the entire illness 368 course or are individuals in which antibody levels were slow to respond. Wajnberg et al found 369

the latter in assessment of longitudinal samples, noting that in addition to a slow antibody 370 response to SARS-CoV-2 infection, peak titers were lower than patients with a more robust 371 initial response (17). This may be clinically relevant, as patients with low and slow antibody 372 response may be those likely to benefit most from SARS-CoV-2 antibody treatment regimens. 373 374 We then further characterized serosilent patients by looking at their clinical characteristics and 375 immune response. Although conclusive analysis is limited by sample size, two out of the four 376 serosilent (N-, S2-) older adults required ventilation, three were admitted to the ICU and two 377

were the only deceased patients in the study, suggesting negative clinical outcomes associated 378 with minimal antibody response in these patients. In contrast to severe COVID-19 disease linked 379 with high antibody response as discussed above, serosilent patients suggest an alternative 380 immunologic profile to infection providing an additional avenue for poor clinical outcomes. 381

Among the cytokine differences discovered between serosilent and responsive patients, IL-10 382 was implicated as one of the most differential, with serosilent individuals displaying significantly 383

higher levels compared to the rest of the older patient cohort. This is again congruent with 384 known influences of IL-10 as discussed above and highlights its potential role in the serosilent 385 patient cytokine profile (16). We additionally found a significant decrease in sCD40L in 386 serosilent patients compared to the rest of the group, which is consistent with sCD40L's known 387 ability to promote B cell proliferation, differentiation and immunoglobulin production (18). 388 389

A substantial limitation in our study was the small sample size which likely limited our ability to 390 detect relationships between epitopes, cytokines, and clinical outcomes. This further limited in 391 our ability to statistically analyze and classify our serosilent samples and therefore were 392 categorized subjectively based on full length IVTT N and S2 protein reactivity. As this small 393 cohort is meant for hypothesis generation, a larger cohort is needed to further validate our 394 findings. Our study is also limited to epitopes produced in Escherichia coli which restricts our 395

ability to see epitopes that require eukaryotic post-translational modification such as 396

glycosylation. This is particularly relevant in regards to the spike protein, which exists as a 397 trimer on the virion surface and undergoes conformational changes during viral entry into cells 398 (6). As addressed in Camerini et al we also observed similar limitations in response to S1 399 . CC-BY-NC 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 January 13, 2021. ; fragments produced in vitro, which perhaps was influenced in part by prokaryotic production of 400 IVTT proteins. However, we were able to detect an area in S1 which is notable in its correlations 401 and outcomes as discussed above. 402 403

404 Patient sample and clinical data collection 405

Patients who tested positive for COVID-19 by PCR at the University of Virginia Medical Center 406 had residual routine lab specimens collected into a biorepository. Serum samples of patients of 407

varying age and severity were collected from April 2020 until July 2020. Blood collected in 408

EDTA was centrifuged at 1300 x g for 10 minutes, then plasma was aliquoted and stored at -409 80°C until testing. Thirty of these serum samples were provided to Antigen Discovery Inc. to be 410 exposed to protein microarrays as described below. 411 412

Clinical data including patient medical history, lab work and clinical course were collected from 413 the electronic medical record using honest brokers with unique study numbers to ensure 414 confidentiality (Tables 1, S2 30 aa protein fragments and overlapping 13-20 aa peptides from SARS-CoV-2 (WA-1), SARS-427

CoV, MERS-CoV, HCoV-NL63 and HCoV-OC43. Purified proteins and peptides were obtained 428 from BEI Resources. All these coronavirus proteins were produced in Escherichia coli except the 429 SARS-CoV-2 and SARS-CoV S proteins, which were made in Sf9 insect cells and the SARS-430

CoV-2 RBD, made in HEK-293 cells. Other proteins and protein fragments were expressed using 431 an E. coli in vitro transcription and translation (IVTT) system (Rapid Translation System,  432 Biotechrabbit, Berlin, Germany) and printed onto nitrocellulose-coated glass AVID slides (Grace 433

Bio-Labs, Inc., Bend, OR, USA) using an Omni Grid Accent robotic microarray printer (Digilabs,  434 Inc., Marlborough, MA, USA regression was used to model associations between antibody and patient information obtained 481

from electronic records. Antibody responses to individual reactive antigens (n=52) were modeled 482 as dependent variables, and the following variables were modeled as independent variables: sex, 483

age category, requirement of a ventilator, days symptomatic prior to sample collection, length of 484 hospital stay, admission to the ICU, maximum required supplemental oxygen category, 485 comorbidity score, maximum body temperature during while admitted, body-mass index (BMI), 486 maximum CRP, maximum ferritin, maximum D dimer, minimum lymphocytes, maximum AST 487 . CC-BY-NC 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 January 13, 2021. ;

The trinity of COVID-19: immunity, 522 inflammation and intervention

Antibody responses to 525 SARS-CoV-2 in patients with COVID-19

Profiling Early Humoral 527 Response to Diagnose Novel Coronavirus Disease (COVID-19)

COVID-19): A 531 Post hoc Analysis of LOTUS China Trial

535 Phosphoregulation of Phase Separation by the SARS-CoV-2 N Protein Suggests a 536

Biophysical Basis for its Dual Functions

Distinct conformational 539 states of SARS-CoV-2 spike protein. Science (80-)

Magnitude and kinetics of anti-SARS-CoV-2 antibody responses and their relationship to 542 disease severity

Factors 545 associated with COVID-19-related death using OpenSAFELY

Sex, age, and 548 hospitalization drive antibody responses in a COVID-19 convalescent plasma donor 549 population. medRxiv Prepr Serv Heal Sci

Antigen-Specific Adaptive Immunity to SARS-CoV-2 in Acute COVID-19 and 553 Associations with Age and Disease Severity

A systematic review of antibody mediated immunity to coronaviruses: kinetics, 557 correlates of protection, and association with severity

Preexisting and de 560 novo humoral immunity to SARS-CoV-2 in humans. Science (80-)

Viral epitope profiling 563 of COVID-19 patients reveals cross-reactivity and correlates of severity. Science (80-)

OPEN SARS -CoV -2 S1 and N -based 566 serological assays reveal rapid seroconversion and induction of specific antibody response 567 in COVID -19 patients

Longitudinal analyses 570 reveal immunological misfiring in severe COVID-19

IL-10: The Master Regulator of Immunity to 572 Infection

Robust 574 neutralizing antibodies to SARS-CoV-2 infection persist for months. Science (80-)

Controlling the false discovery rate: a practical and powerful 578 approach to multiple testing

circlize implements and enhances circular 580 visualization in R

Complex heatmaps reveal patterns and correlations in 582 multidimensional genomic data

Friendly M. Corrgrams: Exploratory displays for correlatigon matrices

and troponin lab levels, and the base 2 log-transformed measurements from the Milliplex serum 488 analysis. Due to the moderate sample size of the study, not all independent variables were 489 modeled simultaneously. Three "base" variables were used to adjust the effect estimates of all 490 other independent variables in separate 4-variable models; these base variables were sex, age 491 category and requirement of a ventilator. Adjustment for the false discovery rate was performed 492using the "p.adjust" function in R (19). To select variables associated with SARS-CoV-2-specific 493 antibodies, linear mixed effects regression (LMER) was used to model all antibody responses 494 against SARS-CoV-2 reactive antigens with random intercepts at the sample level and antigen 495 level to adjust for repeated measures. Similar to the approach with OLS regression, LMER 496 models used the same 3 base variables to fit separate models for all other fixed effects variables. 497All coefficients were returned from models fit using restricted maximum likelihood (REML). To 498 generate P-values for LMER models, the models were refit using maximum likelihood (ML) and 499 compared by ANOVA against null models with the coefficient removed using ML. Cytokines 500 and chemokines that were significantly associated with antibody levels in LMER models were 501 correlated with SARS-CoV-2 reactive antigens using Pearson's correlation coefficient. Clinical 502 patient variables were associated with cytokine levels using OLS regression, similarly to 503 antibody models. Correlation between SARS-CoV-2 S2 and N proteins with HCoV-OC43 and 504HCoV-NL63 S2 and N proteins was assessed using Pearson's correlation coefficient. Samples 505 were categorized as "serosilent" if full-length S2 IgG responses were less than 1.0 normalized 506 signal intensity. Differences in median log2 cytokine levels between serosilent and seroreactive 507 subjects was assessed using Wilcoxon's rank sum test. Data visualization was performed using 508 the circlize (20), ComplexHeatmap (21), ggplot2 and corrplot (22)packages in R. The p-values 509presented for full-length and overlapping fragments of SARS-CoV-2 proteins were not adjusted 510for the false discovery rate, because the measurements are not independent and an appropriate 511 method of p-value correction was not to our knowledge available for the extent of dependence in 512 the antibody measurements. As expected, there were high levels of colinearity in the antibody 513 response to overlapping fragments of different sizes in the reactive regions of SARS-CoV-2 514proteins. Although unadjusted p-values were used in these comparisons, the concordance of 515 fragment antibody binding and differential immunoreactivity in the independent study reported 516 concurrently in Camerini et al lends confidence that the responses reported are unlikely to be due 517 to chance. The full results from linear models are included in Supplemental File 1. However, 518 further studies will be able to validate these findings. 519 520