Immune responses in COVID-19 respiratory tract and blood reveal mechanisms of disease severity

Although the respiratory tract is the primary site of SARS-CoV-2 infection and the ensuing immunopathology, respiratory immune responses are understudied and urgently needed to understand mechanisms underlying COVID-19 disease pathogenesis. We collected paired longitudinal blood and respiratory tract samples (endotracheal aspirate, sputum or pleural fluid) from hospitalized COVID-19 patients and non-COVID-19 controls. Cellular, humoral and cytokine responses were analysed and correlated with clinical data. SARS-CoV-2-specific IgM, IgG and IgA antibodies were detected using ELISA and multiplex assay in both the respiratory tract and blood of COVID-19 patients, although a higher receptor binding domain (RBD)-specific IgM and IgG seroconversion level was found in respiratory specimens. SARS-CoV-2 neutralization activity in respiratory samples was detected only when high levels of RBD-specific antibodies were present. Strikingly, cytokine/chemokine levels and profiles greatly differed between respiratory samples and plasma, indicating that inflammation needs to be assessed in respiratory specimens for the accurate assessment of SARS-CoV-2 immunopathology. Diverse immune cell subsets were detected in respiratory samples, albeit dominated by neutrophils. Importantly, we also showed that dexamethasone and/or remdesivir treatment did not affect humoral responses in blood of COVID-19 patients. Overall, our study unveils stark differences in innate and adaptive immune responses between respiratory samples and blood and provides important insights into effect of drug therapy on immune responses in COVID-19 patients.

and cytokine responses were analysed and correlated with clinical data. SARS-CoV-2specific IgM, IgG and IgA antibodies were detected using ELISA and multiplex assay in both the respiratory tract and blood of COVID-19 patients, although a higher receptor binding domain (RBD)-specific IgM and IgG seroconversion level was found in respiratory specimens. SARS-CoV-2 neutralization activity in respiratory samples was detected only when high levels of RBD-specific antibodies were present. Strikingly, cytokine/chemokine levels and profiles greatly differed between respiratory samples and plasma, indicating that inflammation needs to be assessed in respiratory specimens for the accurate assessment of SARS-CoV-2 immunopathology. Diverse immune cell subsets were detected in respiratory samples, albeit dominated by neutrophils. Importantly, we also showed that dexamethasone and/or remdesivir treatment did not affect humoral responses in blood of COVID-19 patients.
Overall, our study unveils stark differences in innate and adaptive immune responses between respiratory samples and blood and provides important insights into effect of drug therapy on immune responses in COVID-19 patients.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

INTRODUCTION
Symptoms of SARS-CoV-2 infection, known as coronavirus disease 2019 (COVID-19), vary from asymptomatic or mild disease to critical illness, including respiratory failure and death 1 .
Global efforts focused on developing new drugs and vaccines. While vaccines showed immunogenicity and safety towards SARS-CoV-2 2,3,4 , effects of drug treatments remain controversial. Dexamethasone, a synthetic glucocorticoid drug, can lower the 28-day mortality rate in COVID-19 patients receiving oxygen support, prolong ventilator-free days and improve oxygen partial pressure to fractional inspired oxygen (PaO 2 /FiO 2 ) ratio compared to placebo or standard care 5, 6, 7 . However, SARS-CoV-2 RNA can be detected for longer in patients receiving glucocorticoid treatment 8 . Treatment with remdesivir, a nucleoside analogue inhibiting RNA-dependent RNA polymerase (RdRp) in COVID-19 can shorten time to recovery and provide better clinical outcomes 9, 10, 11 . However, the effects of dexamethasone and/or remdesivir on humoral and cellular immune responses are unclear.
To dissect the breadth of immune responses during SARS-CoV-2 infection in the respiratory tract compared to those detected in blood, we collected paired longitudinal blood and respiratory samples from hospitalised COVID-19 patients and non-COVID-19 controls . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 5 to investigate innate, adaptive and humoral immunity. While discordant cytokine levels were detected in respiratory samples across COVID-19 patients, higher IgM and IgG seroconversion was found in respiratory samples compared to paired blood. While higher frequencies of neutrophils and effector memory (EM)-like CD4 + and CD8 + T-cells were found in COVID-19 respiratory samples, higher antibody levels also correlated with activated cellular immunity. Elevated soluble IL-6R alpha (sIL-6Rα) levels and more robust humoral responses were detected in blood of patients with severe disease, and importantly, dexamethasone (with/without) remdesivir therapy did not reduce immune responses in COVID-19 patients. Overall, our study unveils stark differences in innate and adaptive immune responses between respiratory and blood samples of COVID-19 patients and provides insights into potential biomarkers and immunotherapies for severe COVID-19.

DRASTIC patient cohort
We recruited 66 patients hospitalized at Austin Hospital (Victoria, Australia) with suspected COVID-19 into the preDictoRs of diseAse Severity in criTically Ill  cohort prior to their PCR result (Fig. 1a, Supplementary Table 1). Sixty patients had PCRconfirmed SARS-CoV-2 infection, including 43 ward patients (34.9% requiring non-invasive oxygen support) and 17 patients requiring admission to the ICU (35.3% requiring mechanical ventilation; 58.8% requiring non-invasive oxygen support), while 6 patients were SARS-CoV-2-negative and respiratory IgG RBD ELISA-negative (1 ward; 5 ICU patients; Supplementary Table 2). 12 ward patients and six ICU patients were on dexamethasone treatment, while 8 ward patients and 10 ICU patients were on dexamethasone (with/without remdesivir) treatment. The median age of COVID-19 patients was 58 years (range 22-90) and 46.7% were females. The median age was 58 years within both ward and ICU patients (Supplementary Table 2).
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
Cytokine levels varied across both COVID-19 patients in respiratory samples and between paired respiratory and plasma samples. Five  had no detectable cytokines across 13 cytokines/chemokines, while high IL-18 levels were detected in plasma of DRASTIC-002, -003, -004 and -013 patients (Fig. 2a).
This disparity was also reflected when cytokine and sIL-6R levels were standardized separately for respiratory and plasma samples. With the red color indicating higher cytokine levels, donors with high cytokine concentration in the respiratory samples did not necessarily display high levels of the same cytokine in their plasma (Fig. 2b). For instance, patient DRASTIC-049 displayed higher IFN-2, IL-10, IL-12p70 and IL-17A in ETA (d8) than other COVID-19 respiratory samples, while the plasma (d7) level of IFN-2 was also higher, DRASTIC-049 had higher plasma level of IL-18 but not IL-10, IL-12p70 or IL-17A (Fig 2b).
Overall, while the inflammatory cytokine/chemokine levels were excessively higher in respiratory fluid compared to plasma in some COVID-19 patients, they were highly variable across COVID-19 patients, indicating that the plasma inflammatory milieu does not reflect the airway inflammation and that hospitalized/ICU COVID-19 patients should be monitored for inflammation in airways to understand disease severity and potential benefits of immunomodulatory treatments.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

High RBD-specific IgM and IgG seroconversion in COVID-19 respiratory samples
SARS-CoV-2-specific antibodies in respiratory samples are relatively unexplored. We measured SARS-CoV-2 RBD-specific IgM, IgG and IgA antibodies in paired respiratory and blood samples using RBD-ELISA and surrogate virus neutralisation test (sVNT) (Fig. 3).
Using sVNT, two ETA samples from DRASTIC-003 and -004 (with low cytokine levels) had detectable neutralizing activity, associated with high levels of RBD-specific IgM, IgG, and IgA antibodies (Fig. 3d, e). Neutralizing activity was not detected in the remaining respiratory samples at the acute time-points. Plasma samples with high neutralizing activity had high levels of all three IgM, IgG and IgA isotypes of RBD-specific antibodies, and anti-RBD IgG and IgA positively correlated with the neutralizing activity (p=0.0002; p=0.0004 respectively, Fig. 3d, e). Seroconverted levels of RBD-specific IgM and IgG antibodies were detected in the majority of COVID-19 respiratory samples (10/13) and patients (6/9) at 77% and 67% (Fig. 3f), suggesting prominence of RBD-specific IgM and IgG in respiratory samples during acute COVID-19.

High prevalence of SARS-CoV-2-NP-specific IgM antibodies in respiratory samples
While anti-RBD antibodies are essential for the neutralization of SARS-CoV-2 24 , nonneutralizing antibodies are important role in antiviral immunity 25 . To understand in-depth antibody profiles and cross-reactivity in respiratory samples, we adapted a multiplex bead array assay 25 (Supplementary Table 5). Antibodies targeting RBD, S proteins, NP of SARS-CoV-2, SARS-CoV-1 and human coronaviruses (229E, NL63, OC43, HKU1) along with isotypes/subclasses (IgM,IgG,binding with FcγR (FcγR2aH,FcγR2aR,FcγR2b,FcγR3aV,FcγR3aF) and C1q, totaling 315 features, were assessed in 14 COVID-19 and 5 non-COVID-19 respiratory samples, and paired plasma. Intermediate to high antibody levels across different SARS-CoV-2 antigens and isotypes were detected in a subset of COVID-19 respiratory and plasma samples (Fig. 4a), especially in patients who lacked . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 7, 2021. ; https://doi.org/10.1101/2021.09.01.21262715 doi: medRxiv preprint 9 inflammatory cytokines in their respiratory samples .

Increasing cellular infiltrates in respiratory specimens during disease progression
To determine cellular immunity in respiratory specimens of COVID-19 patients, samples underwent multi-parameter flow cytometry and analysis using the Spectre R package 26 . Cells were clustered using Flow Self-Organizing Map (FlowSOM) 27 and plotted using Fast Interpolation-based t-distributed Stochastic Neighbour Embedding (FIt-SNE) 28 . Two flow cytometry panels were used to ensure accurate profiling of myeloid and lymphoid cell populations ( Supplementary Fig. 3a, 4a, 4b, Supplementary Table 6).
Clustering of respiratory samples in the myeloid panel revealed that CD66b + neutrophils dominated, with varying levels of CD16 expression (Fig. 5a).
CD14 + macrophages and CD4 + and CD8 + T-cells were also detected but at lower frequencies.
While the cellular component was variable across samples, CD16 hi and CD16 lo neutrophils . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 7, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 were present in all COVID-19 patients apart from DRASTIC-043 (BMT recipient; Fig. 5a, Supplementary Fig. 4c). Although there were only two COVID-19 patients (DRASTIC-026, -049) with multiple ETA samples, we still observed an increase in cellular infiltrates was detected over time, including CD16 lo neutrophils (Fig. 5b). In the respiratory specimens of 6 non-COVID-19 patients, lower levels of neutrophils and macrophages were detected ( Supplementary Fig. 4c). DRASTIC-059 had a large population of CD16neutrophils, while a high frequency of CD16 lo neutrophils was detected in blood, indicating a dominant immature neutrophil population in this patient ( Supplementary Fig. 4c, e).
After excluding neutrophils and monocytes/macrophages in respiratory samples, CD8 + T-cells were the major population of lymphocytes, with varying levels of CD4 + T-cells and natural killer (NK) cells (Fig. 5c, Supplementary Fig. 4d). Increasing infiltrates of Tcells over time were found in patients 026 and 049 (Fig. 5d), similar to neutrophils.
Interestingly, in patient 026, the lymphocyte population was dominated by NK cells early (V1 and V2) and T-cells gradually infiltrated and dominated over time (V6). Low lymphocyte levels were detected in fatal patient-021.
Overall, neutrophils (CD16 +/-) dominated in the respiratory samples of COVID-19 patients, with varying levels of monocytes/macrophages, T-cells (CD4 + and CD8 + ), NK cells, and B cells. T cells in the respiratory samples exhibited an activated and EM-like phenotype compared to paired blood samples, with lower CD4 + to CD8 + T-cell ratio.

COVID-19 patients with higher NIH scores had more robust humoral immune responses
More patients with higher 4-5 NIH scores required ICU during hospitalization (Fig. 1d), while NIH scores of 2-3 were in the mild/moderate group. Although there were no differences in the overall cytokines/chemokines levels between the two NIH severity groups, IL-8 levels in the severe/critical group increased at discharge (V7) compared to admission (V1) (p=0.0004; Fig. 7a), indicating delayed or prolonged innate immune activation. Levels of sIL-6R were significantly higher in the severe/critical group than the mild/moderate group at both V1 (p=0.027) and V7 (p=0.0302), with the severe/critical group having higher sIL-6R and lower IL-6:sIL-6R ratio at V7 than V1 (Fig. 7a).
Anti-RBD IgG titres increased in both severity groups at discharge (p=0.0268; p=0.0002; Fig. 7b). The severe/critical group also displayed substantially higher microneutralization (MN) activity at discharge compared to admission (p<0.0001). PLSDA revealed that at discharge the severe/critical group had higher IgM and IgG antibodies targeting SARS-CoV-2 proteins compared to the mild/moderate group (Fig. 7c).
COVID-19 patients in the severe/critical group had comparable frequencies of immune cells, while they had lower T-cell and eosinophil frequencies (p=0.0011; p=0.0473) than the mild/moderate group at admission (V1; Fig. 7d). Interestingly, frequencies of mucosal associated invariant T (MAIT) cells and δ T-cells negatively correlated with days stayed in hospital (p=0.0022 and p=0.0024 respectively; Fig. 7d).
Overall, while cytokine levels were similar between the two severity groups, patients with more severe symptoms had more robust antibody responses towards the SARS-CoV-2.

Dexamethasone did not alter immune responses in COVID-19 patients
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. ; https://doi.org/10. 1101 Effects of dexamethasone, corticosteroid anti-inflammatory drug, with/without remdesivir on immune responses are unclear. We found very few differences in immune profiles between patients with/without dexamethasone. IL-8 and sIL-6Rα levels at discharge were significantly higher than at admission in the dexamethasone (with/without remdesivir) group, but similar levels were observed without treatment (Fig. 8a). Patients on treatment had lower anti-inflammatory IL-10 levels at discharge (p=0.0281; Fig. 8a). Conversely, humoral responses of patients receiving drugs were not compromised. Patients receiving dexamethasone (with/without remdesivir) generated robust SARS-CoV-2-specific antibody responses (Fig. 8b). Given that 29/33 severe/critical patients were on treatment, compared to 7/27 mild/moderate patients, high antibody levels in the drug group were likely due to disease severity rather than drug treatment. PLSDA revealed that patients prior to drug therapy had higher antibodies against the NP of human coronavirus OC43 rather than SARS-CoV-2 ( Fig.   8c). No significant differences were found in cellular responses, apart from lower T-cell frequency in the drug group (Fig. 8d).

DISCUSSION
Immunity to SARS-CoV-2 in the respiratory tract, the primary site of infection, is incompletely understood. We found discordant inflammatory status in the respiratory tract of COVID-19 patients, whereas non-COVID-19 hospitalized patients had consistently high respiratory cytokine levels. While high SARS-CoV-2-specific IgG and IgM were detected in COVID-19 respiratory samples, IgG with FcγR binding profiles were more prominent in blood. We found higher frequencies of neutrophils, intermediate CD14 + CD16 + monocytes, activated HLA-DR + CD38 + , EM-like CD4 + and CD8 + T-cells in COVID-19 respiratory compared to blood samples, and similar immune responses with dexamethasone (with/without remdesivir) treatment.
High levels of cytokines are commonly found in blood of COVID-19 patients 29,30,31,32 . In respiratory samples, variable cytokine levels (IL-10, IL-17A, IL-18) were detected, while monocyte chemoattractants (MCP-1, MIP-1α, MIP-1β) and innate cytokines (IL-6, IL-10) were at high levels 18,19 . We found hypercytokinemia in respiratory samples compared to blood, but only in selected COVID-19 patients with high IL-6, IL-8, and MCP-1, indicating an inflammatory environment that attracts leukocytes, including neutrophils and monocytes 33,34 . Since most patients did not have similar cytokine profiles in blood and respiratory samples, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. ; https://doi.org/10.1101/2021.09.01.21262715 doi: medRxiv preprint 13 thus measuring both blood and respiratory inflammation might be needed to accurately determine the inflammatory status of the patients.
SARS-CoV-2-specific IgG and IgA were detected previously in BALF, sputum and saliva of COVID-19 patients 35,36,37 . We found detectable anti-RBD IgM, IgA and IgG in COVID-19 respiratory samples, with higher IgM and IgG than non-COVID-19 respiratory As an anti-inflammatory drug, dexamethasone can reduce proinflammatory cytokines including IL-6 and IL-8 41, 42 . We found no significant difference in cytokine/chemokine levels between COVID-19 patients receiving dexamethasone and untreated patients.
Although it has been speculated that dexamethasone can reduce the ability of antibody production in B cells 43 , we showed similar antibody levels in patients receiving dexamethasone. Therefore, severely-ill COVID-19 patients might benefit from dexamethasone treatment as reported 5, 6, 7 , and such treatment does not dampen humoral immunity.
There are limitations to the current study. Firstly, ETA samples were only collected from patients with severe disease requiring invasive oxygen support, therefore, it is unclear whether COVID-19 patients with milder symptoms had less robust immune responses in the respiratory tract. Additionally, most patients in the severe/critical group received dexamethasone, which could be an intercorrelating factor for the differences observed between severity groups. Moreover, while the non-COVID-19 controls provided insights . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. ; https://doi.org/10.1101/2021.09.01.21262715 doi: medRxiv preprint onto the immune status in hospitalized individuals, the comparisons will benefit more if there were larger numbers of non-COVID patients with more homogenous diseases.
Overall, innate and adaptive immune responses are generated in respiratory and blood samples of COVID-19 patients. While immunological features detected in the peripheral blood can be associated with robust immune responses and predict clinical outcomes, monitoring immune responses in the respiratory samples can be of a benefit prior to initiation of therapeutic interventions for COVID-19 patients.

ACKNOWLEDGMENTS
We acknowledge all DRASTIC (The use of cytokines as a preDictoR of disease Severity in criTically Ill COVID-19) investigators from Austin Health, and thank the participants . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021.

Declaration of Interests
The authors declare no competing interests.

DRASTIC study participants and specimens.
We enrolled 60 SARS-CoV-2 PCR-positive patients admitted to Austin Health (Victoria, Australia) and six PCR-negative patients as negative serological controls. Two COVID-19 patients and three SARS-CoV-2 PCR-negative patients died during the study. Peripheral blood was collected in heparinized or ethylenediaminetetraacetic acid (EDTA) tubes during hospitalization. Peripheral blood monocular cells (PBMCs) were isolated via Ficoll-Paque separation. Single cell suspensions were isolated from tissues as previously described 44,45 . ETA samples were obtained as part of routine suctioning of the endotracheal tube airway and involved the passage of a catheter for suctioning into a sterile sputum trap. Sputum samples were spontaneously collected into a sterile container. Pleural fluid was collected by thoracentesis as part of a routine procedure.
The thoracentesis involved percutaneous insertion of a catheter into the pleural space and collection of pleural fluid into a sterile container. Demographic, clinical and sampling information for COVID-19 patients are described in Supplementary Table 1.

Ethics statement. Experiments conformed to the Declaration of Helsinki Principles and the
Australian National Health and Medical Research Council Code of Practice. Written informed consent was obtained from all blood donors prior to the study. The study was . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. Genomic sequencing and bioinformatic analysis. Extracted RNA from RT-PCR positive samples underwent tiled amplicon PCR and Illumina short-read sequencing, quality control, consensus sequence generation and alignment as previously described 46 . A single sequence per patient was used for phylogenetic analysis 22 , with a maximum-likelihood phylogenetic tree generated using IQ-Tree 47 and visualized using the ggtree package in R 48 . Genomic clusters were defined using a hierarchical clustering algorithm; genomic transmission networks grouped multiple clusters supported by epidemiological and genomic data.
Phenotypic whole blood immune analyses. Fresh whole blood (200μl per stain) was used to measure CD4 + CXCR5 + ICOS + PD1 + follicular T cells (Tfh) and CD3 -CD19 + CD27 hi CD38 hi antibody-secreting B cell (ASC; plasmablast) populations as described 15, 49 as well as activated HLA-DR + CD38 + CD8 + and HLA-DR + CD38 + CD4 + T cells, intermediate CD14 + CD16 + and classical CD14 + monocytes, activated CD3 -CD56 + NK cells, MAIT cells, δ -T cells, as per the specific antibody panels (Supplementary Table 6; gating strategy is presented in Supplementary Fig. 4b, c). After whole blood was stained for 20 minutes at room temperature in the dark, samples were lysed with BD FACS Lysing solution, washed and fixed with 1% PFA. AccuCheck Counting Beads (Thermo Fisher Scientific) were added for calculating absolute numbers just prior to acquisition. All samples were acquired on a LSRII Fortessa (BD). Flow cytometry data were analyzed using FlowJo v10 software.  Table   6). After fixing with 1% PFA, the samples were acquired on a LSRII Fortessa (BD).

Phenotypic immune analyses in
AccuCheck Counting Beads were added for calculating absolute numbers just prior to acquisition. Flow cytometry data were analyzed using FlowJo v10 software.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
Inter-and intra-experimental measurements were normalised using a positive control plasma from a COVID-19 patient run on each plate. Endpoint titres were determined by interpolation from a sigmodial curve fit (all R-squared values >0.95; GraphPad Prism 9) as the reciprocal dilution of plasma that produced >15% (for IgA and IgG) or >30% (for IgM) absorbance of the positive control at a 1:31.6 (IgG and IgM) or 1:10 dilution (IgA). Seroconversion was defined when titres were above the mean titre (plus 2 standard deviations) of non-COVID-19 control respiratory or plasma samples.
Microneutralization assay. Microneutralization activity of serum samples was assessed as previously described 52 . SARS-CoV-2 isolate CoV/Australia/VIC01/2020 53 was propagated in Vero cells and stored at -80°C. Sera were heat-inactivated at 56°C for 30 min and serially diluted. Residual virus infectivity in the serum/virus mixtures was assessed in quadruplicate wells of Vero cells incubated in serum-free media containing 1μg/ml of TPCK trypsin at 37°C and 5% CO 2 . Viral cytopathic effect was read on day 5. The neutralizing antibody titer was calculated using the Reed-Muench method 52 .

SARS-CoV-2 surrogate virus neutralisation test (sVNT)
. The plasma samples were tested in neat, and the respiratory samples were tested at 1:9 dilution or at their original dilutions for more diluted samples. The sVNT blocking ELISA assay (manufactured by GenScript, NJ, USA) was carried out essentially as described 51 , which detects circulating neutralizing SARS-CoV-2 antibodies that block the interaction between RBD and ACE2 on the cell surface receptor of the host. HRP-conjugated recombinant SARS-CoV-2 RBD fragment bound to any circulating neutralizing antibodies to RBD preventing capture by the human ACE2 protein in the well, which was subsequently removed in the following wash step.
Substrate reaction incubation time was 20 mins at room temperature and results were read spectrophotometrically. Colour intensity was inversely dependent on the titre of anti-SARS-CoV-2 neutralizing antibodies.
Coupling of carboxylated beads. As previously described 25 , a custom multiplex bead array was designed and coupled with SARS-CoV-2 spike 1 (Sino Biological), spike 2 (ACRO . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
Luminex bead-based multiplex assay. Using the coupled beads mentioned above, a custom CoV multiplex assay was formed to investigate the isotypes and subclasses of pathogenspecific antibodies present in collected plasma samples 25 . Briefly, 20µl of working bead mixture (1000 beads per bead region) and 20µl of diluted plasma (final dilution 1:200) or 20µl of diluted respiratory secretions (final dilution 1:800) were added per well and incubated overnight at 4°C on a shaker. Fourteen different detectors were used to assess pathogen-specific antibodies. Single-step detection was done using phycoerythrin (PE)conjugated mouse anti-human pan-IgG, IgG1-4 and IgA1-2 (Southern Biotech; 1.3µg/ml, 25µl/well). C1q protein (MP Biomedicals) was first biotinylated (Thermo Fisher Scientific), then tetramerized with Streptavidin R-PE (SAPE; Thermo Fisher Scientific) before dimers or tetrameric C1q-PE were used for single-step detection. For the detection of FcγR-binding, soluble recombinant FcγR dimers (higher affinity polymorphisms FcγRIIa-H131, lower affinity polymorphisms FcγRIIa-R131, FcγRIIb, higher affinity polymorphisms FcγRIIIa-V158, lower affinity polymorphisms FcγRIIIa-F158; 1.3µg/ml, 25µl/well; kind gifts from Bruce D. Wines and P. Mark Hogarth) were first added to the beads, washed, and followed by the addition of SAPE. For the detection of IgM, biotinylated mouse anti-human IgM (mab MT22; Mabtech; 1.3µg/ml, 25µl/well) was first added to beads, washed, followed by SAPE.
Assays were read on the Flexmap 3D and performed in duplicates.

Data normalization.
For all multivariate analysis, Tetanus, H1Cal2009, and BSA antigens (positive controls) were removed, as well as SIV (negative control). Low signal features were removed when the 75 th percentile response for the feature was lower than the 75 th percentile of the BSA positive control. Right shifting was performed on each feature (detector-antigen pair) individually if it contained any negative values, by adding the minimum value for that . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. ; https://doi.org/10.1101/2021.09.01.21262715 doi: medRxiv preprint feature back to all samples within that feature. Following this, all data were log-transformed using the following equation, where x is the right-shifted data and y is the right-shifted logtransformed data: y = log10(x + 1). This process transformed the majority of the features to having a normal distribution. In all the subsequent multivariate analyses, the data were furthered normalized by mean centering and variance scaling each feature using the z-score function in Matlab. Plasma and respiratory samples were analysed separately. When analysing samples at time of hospital discharge, to adjust for the confounder of time from symptom onset, each of the features were iteratively regressed with ordinary least squares regression, using the residuals as input for the analysis 54 .
Feature selection using Elastic Net/PLSDA. To determine the minimal set of features (signatures) needed to predict categorical outcomes (COVID-19 diagnosis, NIH scores, drug therapies), a three-step process was developed 55 . First, the data were randomly sampled without replacement to generate 2000 subsets. The resampled subsets spanned 80% of the original sample size, or sampled all classes at the size of the smallest class for categorical outcomes, which corrected for any potential effects of class size imbalances during regularization. Elastic-Net regularization was then applied to each of the 2000 resampled subsets to reduce and select features most associated with the outcome variables. The Elastic-Net hyperparameter, α , was set to have equal weights between the L1 norm and L2 norm associated with the penalty function for least absolute shrinkage and selection (LASSO) and ridge regression, respectively 56 . By using both penalties, Elastic-Net provides sparsity and promotes group selection. The frequency at which each feature was selected across the 2000 iterations was used to determine the signatures by using a sequential step-forward algorithm that iteratively added a single feature into the PLSDA model starting with the feature that had the highest frequency of selection, to the lowest frequency of selection. Model prediction performance was assessed at each step and evaluated by 10-fold cross-validation classification error for categorical outcomes. The model with the lowest classification error within a 0.01 difference between the minimum classification error was selected as the minimum signature. If multiple models fell within this range, the one with the least number of features was selected and if there was a large disparity between calibration and crossvalidation error (over-fitting), the model with the least disparity and best performance was selected.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. ; https://doi.org/10.1101/2021.09.01.21262715 doi: medRxiv preprint PLSDA. Partial least squares discriminant analysis (PLSDA), performed in Eigenvectors PLS toolbox in Matlab, was used in conjunction with Elastic-Net, described above, to identify and visualize signatures that distinguish categorical outcomes (COVID-19 diagnosis, NIH scores, drug therapies). This supervised method assigns a loading to each feature within a given signature and identifies the linear combination of loadings (a latent variable, LV) that best separates the categorical groups. A feature with a high loading magnitude indicates greater importance for separating the groups from one another. Each sample is then scored and plotted using their individual response measurements expressed through the LVs. The scores and loadings can then be cross-referenced to determine which features are loaded in association with which categorical groups (positively loaded features are higher in positively scoring groups, etc.). All models go through 10-fold cross-validation, where iteratively 10% of the data is left out as the test set, and the rest is used to train the model. Model performance is measured through calibration error (average error in the training set) as well as cross-validation error (average error in the test set), with values near 0 being best. All models were orthogonalized to enable clear visualization of results. Maxisorp ELISA plates (ThermoFisher, plasma) were coated with capture antibody overnight, followed by blocking with 1% w/v BSA for a minimum of 1 hour. Samples and standard proteins were added and incubated for 2 hours at room temperature, followed by detection . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
Subsets were evaluated for expression of CD38, HLA-DR, and PD-1 expression using manual gating in FlowJo.
Volcano plots and heatmaps were created using the Spectre R package 26  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021.    . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021. ; https://doi.org/10.1101/2021.09.01.21262715 doi: medRxiv preprint Statistical significance was determined with Mann-Whitney test. c ELISA titration curves against the SARS-CoV-2 RBD for 3 COVID-19 patients with serial respiratory samples. d

Heatmap of percentage (%) inhibition tested by surrogate virus neutralization test (sVNT)
and anti-RBD ELISA titres. e Correlation between anti-RBD antibody titres and (%) sVNT inhibition. Correlation was determined with Spearman's correlation. f Number of (i) samples and (ii) patients with seroconverted anti-RBD IgM, IgG, IgA and positive % sVNT inhibition.
Pink curved lines surrounding the donut graphs indicate the samples/patients with seroconverted IgM. Earliest samples were used for each patient when determining seroconversion which was defined as average titre +2×SD of non-COVID-19 samples.
Positive % sVNT inhibition was defined as % sVNT inhibition ≥ 20%.   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 September 7, 2021. ; https://doi.org/10.1101/2021.09.01.21262715 doi: medRxiv preprint 29 between multiplex and non-multiplex immune features. Correlation was determined with Spearman's correlation and p-values of the correlation matrix were adjusted with False Discovery Rate adjustment. c Volcano plot showing fold difference of 149 cellular features in respiratory samples between patients with "low cytokine" (004, 011, 013) and "high cytokine" (016, 021, 026, 043, 049). d Heatmaps with unsupervised clustering of serological and cellular features in COVID-19 respiratory and blood samples.  Least-Squares Discriminant Analysis was performed for antibodies measured with multiplex bead array assay. Volcano plots were created using a Wilcoxon rank-sum test and statistics were corrected with FDR adjustment. V1, hospital admission; V7, hospital discharge.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021.    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 September 7, 2021.

Total=13
IgA + IgM + IgG + sVNT IgM + IgG + IgA + IgM+IgG+ IgM + IgG +    . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021.    Normalized Responses   . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 7, 2021.   . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 7, 2021.  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 7, 2021.  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 7, 2021. ; https://doi.org/10.1101/2021.09.01.21262715 doi: medRxiv preprint