Predicting COVID-19 Severity with a Specific Nucleocapsid Antibody plus Disease Risk Factor Score

The COVID-19 pandemic has resulted in over two million deaths worldwide. Despite efforts to fight the virus, the disease continues to overwhelm hospitals with severely ill patients.

the phage-displayed epitopes (3 pools of n = 5 patients per pool). Nonspecific interactions were blocked (ChonBlock), and phage-displayed epitopes were added for ELISA. The resultant data were normalized by signal from the corresponding negative control (phage without a displayed epitope). Seven candidate epitopes from the pooled patients were further investigated with a larger number of individual patient samples (n = 28) (Fig. 2B). The strongest reproducible binding was observed for three epitopes from M (Ep6), N (Ep9), and S (Ep20) proteins. Additional COVID-19 plasma samples were profiled for binding to these three epitopes (n = 86 total) (Fig. 2B).
Only the Ep9 epitope from N protein demonstrated robust, statistically significant antibody binding in 27% of patients (n = 186) (Fig. 2B). Of these patients, 100 did not have corresponding health information and were not analyzed further in this report. To test non-phage-displayed epitopes, dose-dependent binding of antibodies to Ep9 fused to enhanced green fluorescent protein (eGFP-Ep9) or to full-length N protein demonstrated that anti-Ep9 (aEp9) IgGs bound its antigen with a 50% effective concentration (EC 50 ) of 3.22 nM (95% confidence interval [CI] = 2.49 to 4.14 nM). This experiment examined plasma samples with the highest IgG response against the N protein in the COVAM assay. Patients without aEp9 Abs had roughly the same level of binding to N protein as observed for aEp9 Abs binding to Ep9. However, such aEp9 Abs appeared to add to N protein binding by antibodies; an approximately 2-fold increase in apparent antibody binding levels for N protein was observed, if the patient also had aEp9 Abs (Fig. 2C). In patients for whom longitudinal samples were available, the highest levels of aEp9 Abs were observed at days 1 to 14 post-symptom onset (n = 11) and were detectable within 6 days (Fig. 2D). In four of these patients, aEp9 Abs persisted after day 14.
Cross-reactivity of aEp9 Abs against orthologous epitopes from other coronaviruses. Next, the cross-reactivity of aEp9 Abs was examined with Ep9 orthologs from four phylogenetically related coronaviruses known to infect humans (Fig. S2A). Specifically, plasma with aEp9 Abs (n = 3) and pooled plasma from healthy individuals (n = 5) were assayed. The Ep9 epitopes from SARS-CoV-2 and SARS have 90% amino acid sequence homology. Unsurprisingly, this high degree of similarity resulted in a cross-reactive Ep9 epitope, and a strong antibody response was observed to Ep9 epitopes from both viruses (Fig. 2E). The coronaviruses MERS, HKU-1, and NL63 have 52%, 43%, and 8% sequence homology to SARS-CoV-2 Ep9, respectively (Fig. S2B). These more distantly related orthologs exhibited no cross-reactivity with the aEp9 Abs. Furthermore, no response was observed to Ep9 in pooled plasma from healthy individuals. , or E (D) proteins illustrate our epitope design (colored). Sequence Ep13* has the mutation D614G, which increases the fitness of SARS-CoV-2 (51)(52)(53). The depicted structural models were derived from an S protein X-ray structure (PDB: 6VXX) (24) or computation modeling of N, M, and E proteins (Protein Gene Bank: QHD43423, QHD43419, and QHD43418, respectively) (54). Table S1 provides sequences, sources, and rationale for epitope design.
The protein microarray COVAM analysis is a high-throughput serological test for SARS-CoV-2 Ab cross-reactivity with a panel of 61 antigens from 23 strains of 10 respiratory tract infection-causing viruses (22). In this assay, each antigen was printed onto microarrays, probed with human plasma, and analyzed with an ArrayCam imager. COVAM distinguishes between IgG and IgM Abs binding to the full-length N protein ( Fig. S3 and S4, respectively). Thus, the COVAM analysis complemented the phage ELISA by expanding the scope of antigens surveyed and adding Ab serotype information. The ELISA and COVAM data both demonstrated that aEp9 Abs were highly FIG 2 Mapping COVID-19 patient antibody responses with phage-displayed SARS-CoV-2 epitopes. (A) This phage ELISA with the indicated epitopes (x axis) examined plasma pooled from patients (n = 3 pools of 5 patients each, 2 technical replicates). STOP4 is the phage negative control. (B) The epitopes with the highest signals were then further examined by ELISA with plasma from individual patients (n as indicated). (C) This ELISA measures dose-dependent binding of aN IgGs from plasma pooled from five aEp9-positive patients and five non-Ep9, aN-positive patients to eGFP-Ep9 (dashed line), eGFP negative control (eGFP-FLAG, dotted line), or fulllength N protein (fl-N, solid line). The indicated concentrations of Ep9 or fl-N were immobilized on microtiter plates, and binding of pooled patient plasma (1:100) was detected using a-Fc IgG-HRP Abs (1:10,000). Pooled patients were matched by similar aN IgG binding signal in COVAM analysis (inset). Nonlinear lines of best fit for binding saturation are represented. Statistical comparisons of B max , Hill slope, and EC 50 between groups determines that binding of aEp9 IgGs to fl-N or eGFP-Ep9 and that of non-Ep9, aN IgGs to fl-N are significantly different (P , 0.0001). Error bars represent 6SD. The data demonstrate that the EC 50 value of aEp9 Abs is equal to the cumulative EC 50 of all other aN Abs in patients lacking the aEp9 Abs. In the presence of the aEp9 Abs, the apparent binding levels of aN Abs against fl-N approximately double. (D) With samples from individual patients (designated P# and by color) collected at the indicated times, aEp9 Abs were measured. The subset of patients shown here comprises all samples for which longitudinal data were available. (E) Phage ELISA with samples from patients with strong aEp9 Ab responses (two from the longitudinal study and one from the patient population) examines cross-reactive binding to Ep9 or Ep9 orthologs from the indicated coronaviruses (x axis, 3 technical replicates). The arrow on the y axis and gray line (B and D) represents the negative control used for normalizing the data. Error bars represent SEM (A, B, C, and E) or range of two measurements (D). specific for lineage B betacoronaviruses and were unlikely to be found in patients before their infection with SARS-CoV-2.
More severe disease and poorer outcomes for aEp9 patients. Direct comparison of data with full-length N protein from COVAM and Ep9 phage ELISA (n = 40 patients assayed with both techniques) revealed five unique categories of patients (Fig. 3A). To enable this comparison, raw data from each assay were normalized as a percentage of the negative control. Category 1 consists of patients without Abs to the N protein. The next categories included patients with IgMs (category 2) or IgGs (category 3) binding to N protein, but not Ep9, termed non-Ep9 aN Abs. Category 4 included patients with aEp9 Abs (both IgMs and IgGs). Category 5 patients had exclusively IgG aEp9 Abs. The aEp9 Abs were found only in patients with IgMs or IgGs against full-length N protein from the COVAM assay; the COVAM analysis thus independently corroborated the phage ELISAs (Fig. 3A).
Interestingly, the patients with aEp9 Abs suffered more prolonged illness and worse clinical outcomes compared to patients with non-Ep9 aN Abs or no aN Abs. In this study, severe COVID-19 cases were defined as resulting in death or requiring admission to the intensive care unit (ICU) or intubation. The fraction of severe COVID-19 cases was 2.5 times higher in aEp9 Abs patients than non-Ep9 aN Abs patients (Fig. 3B, yellow panel); the differences in proportions of severe and nonsevere aN-positive patients with or without aEp9 Abs were statistically significant (P , 0.030, Fisher's exact test). Patients without aN Abs (category 1) had less severe symptoms. The aEp9 Ab patients also had longer durations of symptoms and hospital stays relative to patients with non-Ep9 aN Abs or no aN Abs ( Fig. 3C and D). A larger data set of patient plasma analyzed by phage ELISA confirmed this conclusion (P , 0.0013, Fisher's exact test) Our data further demonstrated that asymptomatic COVID-19 patients (n = 3) also tested negative for aEp9 Abs (Table S2). The data also revealed early seroconversion of aEp9 IgGs ( Fig. 3E) but not aEp9 IgMs (Fig. 3F).
Strong correlation between disease severity and comorbidities in patients with aEp9 Abs. We compared risk factors, clinical parameters, and disease outcomes among patients with aEp9 Abs (n = 23) ( Fig. 4A and Fig. S5). A disease risk factor score (DRFS) was developed to evaluate the relationship between clinical preconditions and disease severity in patients with aEp9 Abs. The DRFS quantified a patient's age, sex, and preexisting health conditions associated with COVID-19 disease severity and mortality. Risk factors include hypertension, diabetes, obesity, cancer, and chronic conditions of the following kinds: cardiac, cerebrovascular, and kidney (26)(27)(28)(29). Using the age score from the Charlson comorbidity index (30) yields a patient's DRFS as DRFS = R (number of risk factors) 1 (age score), where each risk factor was valued as either 0 or 1 if absent or present, respectively. The DRFS of patients with aEp9 Abs strongly correlated with COVID-19 disease severity (Pearson's r = 0.72, P value , 0.0001, and R 2 = 0.52) (Fig. 4A). The correlation in patients without aEp9 Abs was weak (r = 0.30, P value = 0.089, R 2 = 0.018) (Fig. 4A). Among patients with aEp9 Abs (n = 23), a DRFS of $3 determined disease severity with 92.3% sensitivity (1/13 false negatives) and 80% specificity (2/10 false positives) (Fig. 4B). In the entire study cohort (n = 86), patients with aEp9 Abs and a DRFS of $3 (n = 11) have severe disease with a high degree of specificity (96.7%) and a sensitivity of 44%. Notably, DRFS predicted disease severity only for patients with aEp9 Abs (n = 23), and patients without such Abs (n = 63) had no correlation with disease outcomes.
Examining key contributors to high DRFS, the presence of aEp9 Abs correlated with more severe disease in patients who have hypertension, diabetes, or age of .50 years. Correlation of disease severity with DRFS in patients with aEp9 Abs. The data depict a significant correlation between DRFS and disease severity in patients with aEp9 Abs (blue) but not in patients lacking aEp Abs (gray). In aEp9 patients, a DRFS threshold of 3.0 can predict severe disease (red). Two-tailed, parametric t tests were conducted to compare nonsevere and severe disease outcomes of patients with and without aEp9 Abs, where **** indicates P , 0.0001. The error bars represent SD with the indicated n. (C) The color-indicated risk factors (diabetes, hypertension, and age score) are depicted on the x axis as the fractions of patients in each disease severity category (y axis). Numbers indicate total patients (n) without aEp9 Abs (left) or with aEp9 Abs (right). The prevalence of risk factors (colors) increases with disease severity in patients with aEp9 Abs but not in patients without these Abs. (D) Patients with aEp9 Abs and DRFS of $3 are predisposed to increased COVID-19 severity and poorer outcomes.
Such correlation was not observed for patients lacking aEp9 Abs (Fig. 4C). Such risk factors were prevalent at roughly the same percentages in the two populations of patients (Table S2). Thus, these risk factors were particularly acute for patients with aEp9 Abs.
High levels of inflammatory cytokine and tissue damage markers in patients with aEp9 Abs. COVID-19 patients can have elevated serum concentrations of .20 inflammatory cytokines and chemokines (31). However, information on the cytokine levels and the association with tissue damage and worse COVID-19 outcomes has been inconsistent (31)(32)(33). For patients with IL-6 concentrations measured in plasma, patients with (n = 8) or without (n = 11) aEp9 Abs were compared. Interestingly, the comparison uncovered a strong positive sigmoidal association between IL-6 and AST unique to patients with aEp9 Abs (R 2 = 0.968, Spearman's r = 1.0, P value , 0.0001, n = 8) (red line, Fig. 5A); correlation of IL-6 and AST in patients with aEp9 Abs remained strong even after removal of the data point at the highest IL-6 concentration. Conversely, a slight negative trend was observed in patients lacking aEp9 Abs (Spearman's r = 20.575, P value = 0.0612, n = 13). Thus, the presence of aEp9 Abs could disambiguate the sometimes-contradictory association of IL-6 with disease severity.

DISCUSSION
This study introduces a two-step test as a prognostic for predicting COVID-19 disease severity and its worst outcomes. Specifically, aEp9 Abs can effectively predict severe disease (specificity, 83.6%). However, combining presence of aEp9 Abs with DRFS of $3 provides much higher specificity (96.7%) for predicting severe disease. Previously, aN IgGs have been recognized as a focal site for an antibody response (18,19,21,34) and associated with disease severity and poor outcomes (11,34,35).
The present investigation expands on previous reports that recognize various regions of the RNA binding domain of N protein as focal sites for anti-SARS-CoV-2 antibody response. For example, the phage display-based VirScan identified an epitope region spanning residues 141 to 196 and microarrays further isolated peptides Abs with increasing IL-6 levels leading to poor outcomes. We hypothesize that patients with aEp9 Abs could benefit from IL-6 inhibition early in the disease, such as monoclonal antibody drugs targeting IL-6 or its receptor (IL6R), to disrupt a cytokine storm and reduce severe outcomes.
including residues 134 to 171, 155 to 171, 153 to 190, and 153 to 171 (18,19,21). The above investigations, however, do not find correlations between any of these epitopes and disease severity. Our results are confirmed by observations from a patient cohort in Singapore, which identify an epitope (residues 153 to 170) very similar to Ep9 (residues 152 to 172) and show a correlation between antibody response against the epitope and pneumonia and the tissue damage markers (CRP and lactate dehydrogenase [LDH]) (20). In our investigation, we examine in-depth patient clinical histories, test results, disease outcomes ranging from asymptomatic to fatal, and longer longitudinal profiling post-symptom onset, to determine the association of a larger subset of markers and risk factors. Such data allow calculation of the DRFS. Together with the presence of aEp9 Abs, patient DRFS allows early discrimination of severe from nonsevere disease outcomes. Additionally, fine epitope mapping demonstrates that aEp9 Abs strongly and uniquely correlate with COVID-19 disease severity relative to other aN Abs.
We hypothesize that the underlying mechanism relating aEp9 Abs to increased disease severity involves an overzealous immune response. Specifically, we observe early seroconversion and strong early upregulation of aEp9 IgGs (Fig. 3E). Similar IgG observations have been correlated with poor viral neutralization and clearance, resulting in increased COVID-19 severity (10,35,36). Also, high levels of IL-6 are observed for aEp9-positive patients with increased levels of the tissue damage marker AST; this correlation does not exist for patients lacking aEp9 Abs (Fig. 5A). The sensitivity to IL-6 concentration before AST-monitored organ damage suggests anti-IL-6 therapeutics could be an effective tool for management in the early and rapidly progressive stages of respiratory distress for aEp9-positive patients (31,(37)(38)(39)(40)(41). Since binding to N protein by aEp9 antibodies is unlikely to enhance uptake of SARS-CoV-2, an antibody-dependent enhancement mechanism could invoke antigen uptake by macrophages. This mechanism could stimulate complement activation and the cytokine storm observed here as elevated IL-6 response. Further investigation is required to determine the basis for increased disease severity in aEp9 patients.
The data demonstrate that aEp9-positive patients with a DRFS of $3 are 13.42 times (likelihood ratio) more likely to have severe COVID-19 disease symptoms within the study cohort (n = 86). The presence of aEp9 without DRFS is less effective as a prognostic (likelihood ratio of 3.17). Despite its high specificity (96.7%), the sensitivity of this two-step test is 44% (n = 86). However, this test could predict a subset of patients with a specific immune response (i.e., early IgG response and IL-6-dependent immune hyperactivity), and could suggest targeted treatment options (e.g., targeting IL-6 and its pathways).
Importantly, aEp9 Abs appear early in the course of disease. Thus, such a prognostic could outperform traditional markers for the cytokine storm such as IL-6, which appears 6 to 8 days after symptom onset (31,39); all plasma samples collected from aEp9-positive patients (n = 7, Fig. 2D) between 1 and 6 days post-symptom onset demonstrate detectable levels of aEp9 IgG ($2-fold over negative control). Early detection of aEp9 Abs in patients could be used to triage and treat COVID-19 prior to the onset of its most severe symptoms; delayed treatments with IL-6-targeting drugs can decrease their efficacy or be counterproductive (31,(37)(38)(39)(40)(41)(42) (Fig. 5B). The aEp9 Ab biomarker could identify patients most likely to benefit from anti-IL-6 therapeutics and avoid ineffective treatments.
This study demonstrates the usefulness of fine epitope mapping, but the following limitations should be noted. Short linear epitopes, unlike conformational epitopes in larger domains, might not resemble the tertiary structure of an antigen. Posttranslational modifications, such as glycosylation, were omitted for the phage-displayed S protein epitopes; the COVAM antigens, however, are produced in baculovirus or HEK-293 cells, which could glycosylate the antigens. Our analysis is largely based upon a population of 86 COVID-19 patients and 5 healthy individuals, with the majority being of Hispanic descent. The conclusions could be further strengthened with follow-up investigations in a larger population. Additionally, the population examined here included only three asymptomatic individuals, and additional testing is required to verify the absence of aEp9 Abs in such patients. The sample size of patients with multiple antibody targets was too limited to allow correlation analysis; future investigations could examine associations between aEp9 and other Abs. Abs recognizing other SARS-CoV-2 structural proteins could also exhibit characteristics similar to aEp9 Abs.
Existing diagnostic platforms could readily be adapted to test for aEp9 Abs (e.g., assay with eGFP-Ep9 fusion demonstrated here), and the DRFS calculation is quite simple to implement. As shown here, aEp9 Abs do not recognize orthologous sequences from closely related coronaviruses, providing good specificity for aEp9 as a prognostic. Previous studies have shown that the high homology of N protein among related coronaviruses can lead to high false-positive rates in serodiagnostics with full-length N antigen (43). Thus, the two-step prognostic reported here could mitigate the worst outcomes of COVID-19, particularly for patients at high risk.

MATERIALS AND METHODS
Cloning. For phage display of epitopes, the pm1165a phagemid vector as previously described (44) was engineered to encode an N-terminal FLAG tag and a C-terminal fusion to the P8 coat protein of M13 phage. This template, termed FlagTemplate, was used for subcloning of SARS-CoV-2, SARS, MERS, HKU-1, and NL63 epitopes. A vector map of the FlagTemplate (see Fig. S6A in the supplemental material), cloning procedures, and a list of oligonucleotides (Table S3) for Q5 site-directed mutagenesis and Gibson assembly are provided.
Short (approximately 30 amino acids) putative epitopes for phage display and Escherichia coli expression as eGFP fusion peptides in the pET28 vector were cloned via Q5 site-directed mutagenesis according to the manufacturer's instructions. A vector map of the peptide with Ep9 fused to eGFP, termed eGFP-Ep9, is shown in Fig. S6B. For large epitopes (.500 bp), such as Ep17, Gibson assembly (New England Biolabs) was conducted in two PCR steps with the FlagTemplate or pCAGGS containing the SARS-CoV-2 S protein gene (BEI Resources) to generate the vectors and inserts, respectively. The Gibson assembly (2 ml) or KLD (kinase, ligase, DpnI) mix (5 ml) was transformed into Nova Blue E. coli competent cells, and transformants were plated on a carbenicillin-supplemented (50 mg/ml) agar plate before incubation at 37°C overnight. Five single colonies were selected to inoculate 4 ml of super optimal broth (2% wt/vol tryptone, 0.5% yeast extract, 8.56 mM NaCl, 2.5 mM KCl, 10 mM MgCl 2 , 10 mM MgSO 4 ) in a 15-ml culture tube supplemented with carbenicillin (50 mg/ml). The seed cultures were incubated at 37°C with shaking at 225 rpm for 8 to 12 h. Phagemid DNA was isolated using the QIAprep spin miniprep kit according to the manufacturer's instructions. The successful subcloning of the open reading frame (ORF) encoding each epitope was verified via DNA sequencing (Genewiz). The full-length N protein in a pLVX-EF1a-IRES-Puro plasmid was a generous gift from Rachel Martin of University of California, Irvine (UCI).
Purification and preparation of phage. Phage were propagated and purified using procedures previously described (45) with the following changes. A single colony was selected to inoculate 15 ml of yeast extract and tryptone media (2YT) (1.6% wt/vol tryptone, 1% wt/vol yeast extract, 0.5% wt/vol NaCl) and shaken at 37°C until the optical density at 600 nm (OD 600 ) reached 0.6. After incubation at 37°C for 45 min, 8 ml of the primary culture was used to inoculate 300 ml of 2YT supplemented with carbenicillin (50 mg/ml), kanamycin (20 mg/ml), and isopropyl-b-D thiogalactopyranoside (IPTG; 30 mM).
To precipitate the phage, the cultures were centrifuged at 10 krpm (15,300 Â g) for 10 min at 4°C. The supernatant was decanted into a centrifuge tube containing 60 ml polyethylene glycol (PEG) 8000 (20%, wt/vol) and NaCl (2.5 M). The tube was inverted 10 times and stored on ice for 30 min followed by an additional centrifugation at 10 krpm (15,300 Â g) for 20 min at 4°C. The supernatant was decanted, and tubes were centrifuged for an additional 4 min at 4 krpm (2,429 Â g) at 4°C. The pellets were resuspended in PBS (10 mM phosphate, 137 mM NaCl, pH 7.2) with Tween 20 (0.05%, vol/vol) and glycerol (10%, vol/vol), separated into 1-ml aliquots, flash frozen with liquid nitrogen, and stored at 280°C. For binding assays via ELISA, the purified phage was thawed on ice, precipitated a second time as before. The quality of each phage preparation was routinely checked by quality control ELISA, termed QC ELISA, to a FLAG peptide fused to the N terminus of each epitope (Fig. S1); additionally, PCR using Oligo69 and Oligo70 followed by DNA sequencing (Genewiz) was performed for every phage preparation. Such quality control allowed for identification of toxic clones; for example, C8 was apparently toxic to E. coli, and three protein epitopes failed to express in E. coli for unknown reasons. The phage concentration was determined by absorbance at 260 nm using a coefficient of molar absorptivity of 0.003 nM 21 cm 21 and diluted to 40 nM in PBS.
Expression and purification of eGFP-Ep9 and N protein. A pET28c plasmid containing Ep9 fused to an N-terminal eGFP (Fig. S6B) was transformed into BL21(DE3)* E. coli heat shock-competent cells. A single colony was transferred to LB medium (20 ml) supplemented with kanamycin (40 mg/ml) and incubated at 37°C for 18 h. An aliquot of the starter culture (2.5 ml) was transferred to LB medium with 1% glucose (250 ml LB in a 1-liter baffled flask). After reaching an OD 600 between 0.4 and 0.6, the culture was induced through addition of IPTG (0.5 mM) before incubation at 25°C for 18 h. The cells were centrifuged (15,300 Â g) for 20 min at 4°C, and the cell pellet was resuspended in lysis buffer (25 mM Tris-HCl and 200 mM NaCl, pH 8.0, and supplemented with protease inhibitor cocktail) followed by sonication.
The lysate was subjected to centrifugation (26,892 relative centrifugal force [rcf], 45 min, 4°C). The supernatant was incubated with charged nickel immobilized-metal affinity chromatography resin overnight on a rotary shaker (150 rpm at 4°C). The resin was equilibrated in a column and washed with wash buffer (20 mM imidazole in lysis buffer), and the purified protein was eluted using elution buffer (250 mM imidazole in lysis buffer). Elutions containing the purified protein were visualized using 10% or 12% SDS-PAGE (Bio-Rad Mini-Protean Tetra electrophoresis system) stained with Coomassie brilliant blue stain (Fig. S7). The eluted fractions containing the purified eGFP-Ep9 were pooled and buffer exchanged for 3 column volumes (20 ml) with lysis buffer without imidazole using a 10-kDa-cutoff microconcentrator (Vivaspin; Fisher Scientific). The protein concentration was determined by a bicinchoninic acid (BCA) assay or Bradford assay using the estimated molecular weight (MW) (http://www.expasy.org). Similar to eGFP-Ep9, the full-length N protein was expressed in 250 ml LB with 1% glucose and induced with 0.25 mM IPTG at an OD 600 of 0.8. Protein overexpression cultures were incubated at 16°C for 22 h. Lysis and purification were conducted as described above, using N protein lysis buffer (20 mM Tris-HCl, 300 mM NaCl, 5 mM MgCl 2 , 5 mM b-mercaptoethanol [BME], 10% glycerol, pH 8.0). The purified fulllength N protein was analyzed using 10% SDS-PAGE (Fig. S7).
Patient sample collection. The UCI Experimental Tissue Resource (ETR) operates under a blanket IRB protocol (UCI no. 2012-8716) that gives ETR personnel "Honest Broker" status and enables the collection of any fluid or tissue remnant in excess of that needed for clinical diagnosis and distribution to investigators under the conditions of their own IRB approval. Patients undergoing COVID testing in the Emergency Department or on the inpatient service with confirmed COVID 1 pharyngeal swabs were followed for their blood collections daily. Specimens collected originally for diagnostic purposes were processed and stored by the hospital laboratory in a manner compliant with College of American Pathologists (CAP) standards. EDTA-anticoagulated whole blood was stored for 2 days at 4°C after clinical diagnosis and released for research purposes. Plasma from heparin-anticoagulated blood was centrifuged immediately after collection and preserved at 4°C for 3 to 4 days before being released for research use. All COVID 1 specimens were handled under biosafety level 2 (BSL-2) conditions, aliquoted into screw-cap cryovials, and stored at 280°C long term with constant temperature monitoring. Specimens were coded by the ETR with unique deidentifiers, and accompanying clinical information was stripped of protected health information such that investigators could receive specimens under a Non-Human Subjects Determination exemption from the UCI IRB. All samples from SARS-CoV-2-infected patients were inactivated by incubation in a water bath at 56°C for 30 min (46), aliquoted (40 ml each), and stored at 280°C.
Phage ELISA with plasma. The phage-displayed SARS-CoV-2 epitopes were used in phage ELISAs with patient plasma samples diluted 100-fold in coating buffer (50 mM Na 2 CO 3 , pH 9.6). After incubation in a 96-well Nunc MaxiSorp flat-bottom microtiter plate with shaking at 150 rpm at 4°C for 12 to 18 h, plasma was aspirated by a plate washer (BioTek). Next, the plate was treated with 100 ml per well of ChonBlock blocking/sample dilution buffer (Chondrex, Inc.) for 1 h with shaking at 150 rpm at room temperature and washed three times with wash buffer (0.05% [vol/vol] Tween 20 in PBS). The epitope displaying phage and controls were diluted to 1 nM in ChonBlock blocking/sample dilution buffer, and 100 ml was added to each well before incubating for 2 h with shaking (150 rpm) at room temperature. The plate was then washed three times with wash buffer. The primary antibody, anti-M13-horseradish peroxidase (HRP) (Creative Diagnostics), was diluted 1:5,000 in ChonBlock secondary antibody buffer, and 100 ml was added per well; the plate was incubated for 1 h at 150 rpm and room temperature. Following three washes with wash buffer, 1-Step Ultra 3,39,5,59-tetramethylbenzidine (TMB)-ELISA substrate solution (100 ml per well; Thermo Scientific) was added. Absorbance of TMB substrate was measured twice at 652 nm by a UV-visible (UV-Vis) plate reader (BioTek) after 5 and 15 min of incubation.
ELISA of eGFP-Ep9 and full-length N protein with plasma. Various doses, with a maximum concentration of 1.7 mM, of eGFP-Ep9, eGFP-FLAG, or full-length N protein (fl-N) were diluted in PBS (pH 8.0) and then immobilized on a 96-well Nunc MaxiSorp flat-bottom microtiter plate before incubation on a shaker (150 rpm) at 4°C for 12 to 18 h. After incubation, unattached proteins were removed through aspiration using a plate washer (BioTek) and wells were blocked with 100 ml ChonBlock blocking/sample dilution buffer (Chondrex, Inc.) for 30 min with shaking (150 rpm) at room temperature. The plate was then washed three times with wash buffer (0.05% [vol/vol] Tween 20 in PBS). Pooled plasma from five patients within each experimental group was diluted 100-fold in ChonBlock blocking/sample dilution buffer, and 100 ml was added to each well before incubating for 1 h with shaking (150 rpm) at room temperature. The plate was then washed three times with wash buffer. The detection antibody, IgG Fc goat anti-human-HRP (Invitrogen), was diluted 1:5000 in ChonBlock secondary antibody buffer, and 100 ml was added per well; the plate was incubated for 30 min at 150 rpm and room temperature. Following six washes with wash buffer, 1-Step Ultra TMB-ELISA substrate solution (100 ml per well; Thermo Scientific) was added. Absorbance of TMB substrate was measured twice at 652 nm by UV-Vis plate reader (BioTek) after 5 and 15 min of incubation.
COVAM. Serum coronavirus antigen microarray (COVAM) included 67 antigens across respiratory virus subtypes including 11 antigens from SARS-CoV-2 expressed in either baculovirus or HEK-293 cells as previously detailed (22). These antigens were provided by Sino Biological U.S. Inc. as either catalog products or custom synthesis service products. The antigens were printed onto microarrays, probed with human sera, and analyzed as previously described (47)(48)(49). Briefly, lyophilized antigens were reconstituted with sterile water to a concentration of 0.1 mg/ml protein in PBS, and printing buffer was added. Antigens were then printed onto Oncyte Avid nitrocellulose-coated slides (Grace Bio-Labs) using an OmniGrid 100 microarray printer (GeneMachines). The microarray slides were probed with human sera diluted 1:100 in 1Â protein array blocking buffer (GVS Life Sciences, Sanford, ME) overnight at 4°C and washed with TTBS buffer (20 mM Tris-HCl, 150 mM NaCl, 0.05% Tween 20 in double-distilled water (ddH 2 O) adjusted to pH 7.5 and filtered) three times for 5 min each. A mixture of human IgG and IgM secondary antibodies conjugated to quantum dot fluorophores Q800 and Q585, respectively, was applied to each of the microarray pads and incubated for 2 h at room temperature, and pads were then washed with TTBS three times for 5 min each and dried. The slides were imaged using an ArrayCam imager (Grace Bio-Labs) to measure background-subtracted median spot fluorescence. Nonspecific binding of secondary antibodies was subtracted using a saline control. The mean fluorescence of the 4 replicate spots for each antigen was used for analysis.
Statistical analysis. The ELISA data were analyzed in GraphPad Prism 8. Since the total antibody content differs from person to person, the raw absorbance values for every patient sample were normalized and represented as the ratio compared to a negative control. Analysis of variance (ANOVA) with Dunnett's multiple-comparison test was performed to determine if values were statistically significant. Correlations between COVAM IgG/IgM and ELISA were determined by plotting normalized values on an xy graph and performing a nonparametric correlation analysis using a Spearman rank correlation coefficient test.
For data visualization of clinical patient data, trends in data were evaluated using Knime Analytics Platform software. GraphPad Prism was used to calculate column statistics including mean, standard deviation, standard error of the mean (SEM), P values, odds ratios, and likelihood ratios defined as sensitivity/(1 2 specificity). ANOVA with Tukey's multiple-comparison test was used to evaluate antibody response and disease severity between patients with aEp9 Abs, non-Ep9 Abs, aN Abs, or non-aN Abs. Comparisons of patients with aEp9 Abs and non-aEp9 Abs were conducted using unpaired, two-tailed, parametric t tests. Contingency graphs were statistically evaluated using Fisher's exact test, for groups with binary categorization, and the chi-squared test for groups with multiple categories. Different data sets were fitted with linear or nonlinear regression methods; the fit with the higher R 2 value was chosen. Correlations between two clinical parameters (e.g., IL-6 and AST) were evaluated using the Pearson coefficient or Spearman coefficients (r) for linear or nonlinear regressions, respectively; r values between 1.0 and 0.7 were considered strong correlations, r values between 0.7 and 0.5 were considered moderate correlations, and values below 0.5 were considered weak correlations (50). The significance of the correlation was evaluated based on a P value of ,0.05.
Data availability. Microarray data have been deposited under BioProject accession no. GSE172471.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.