key: cord-0843144-z4llz9h0
authors: Brunet-Ratnasingham, E.; Anand, S. P.; Gantner, P.; Moquin-Beaudry, G.; Dyachenko, A.; Brassard, N.; Beaudoin-Bussieres, G.; Pagliuzza, A.; Gasser, R.; Benlarbi, M.; Point, F.; Prevost, J.; Laumaea, A.; Niessl, J.; Nayrac, M.; Sannier, G.; Boutin, M.; Descoteaux-Dinelle, J.; Gendron, G.; Orban, C.; Butler-Laporte, G.; Morrison, D.; Zhou, S.; Nakanishi, T.; Laurent, L.; Richard, J.; Dube, M.; Fromentin, R.; Rebillard, R.-M.; Arbour, N.; Prat, A.; Larochelle, C.; Durand, M.; Richards, B.; Chasse, M.; Tetreault, M.; Chomont, N.; Finzi, A.; Kaufmann, D. E.
title: Integrated immunovirological profiling validates plasma SARS-CoV-2 RNA as an early predictor of COVID-19 mortality
date: 2021-03-20
journal: nan
DOI: 10.1101/2021.03.18.21253907
sha: 66b2eeb9d2374e8066c4eacfe1da20f4040b673d
doc_id: 843144
cord_uid: z4llz9h0

Despite advances in COVID-19 management, it is unclear how to recognize patients who evolve towards death. This would allow for better risk stratification and targeting for early interventions. However, the explosive increase in correlates of COVID-19 severity complicates biomarker prioritisation. To identify early biological predictors of mortality, we performed an immunovirological assessment (SARS-CoV-2 viral RNA, cytokines and tissue injury markers, antibody responses) on plasma samples collected from 144 hospitalised COVID-19 patients 11 days after symptom onset and used to test models predicting mortality within 60 days of symptom onset. In the discovery cohort (n=61, 13 fatalities), high SARS-CoV-2 vRNA, low RBD-specific IgG levels, low SARS-CoV-2-specific antibody-dependent cellular cytotoxicity, and elevated levels of several cytokines and lung injury markers were strongly associated with increased mortality in the entire cohort and the subgroup on mechanical ventilation. Model selection revealed that a three-variable model of vRNA, age and sex was very robust at identifying patients who will succumb to COVID-19 (AUC=0.86, adjusted HR for log-transformed vRNA=3.5; 95% CI: 2.0-6.0). This model remained robust in an independent validation cohort (n=83, AUC=0.85). Quantification of plasma SARS-CoV-2 RNA can help understand the heterogeneity of disease trajectories and identify patients who may benefit from new therapies.

Since the beginning of the pandemic, intense efforts have been deployed to define correlates of disease severity and to develop therapies targeting the virus or the pathogenesis of COVID-19. However, to date, only dexamethasone (1-3) and IL-6 blockers (tocilizumab (4), sarilumab (5) ) have convincingly shown to provide a survival benefit in randomized controlled trials. While other immune interventions may benefit some subgroups (6) , there is currently no consensus on how to predict which critical cases are likely to resolve their infection and which are at a greater risk of fatality, in part due to the high heterogeneity of patients and the very dynamic changes in biological features (2) .

Recent reports have identified features linked to severe COVID-19. One is high amounts of viral RNA (vRNA) in plasma, which has been associated with greater severity and worst outcome for other respiratory pathogens, such as SARS-CoV-1 (7, 8) , RSV (9, 10) , MERS (11) , and pandemic-causing strains of influenza A (H5N1 (12) , H1N1 (13) ). Plasma SARS-CoV-2 vRNA has also been linked with increased risk of severe COVID-19 and mortality (14) (15) (16) (17) .

Dysregulated immune responses are at least in part responsible for the exacerbated pathogenesis occurring in a minority of individuals with SARS-CoV-2 infection. Elevated cytokine levels were among the first reported markers associated to severe COVID-19 disease (18) , although inconsistent sampling times sometimes led to weak associations with mortality (19) .

Narrowing the window of sampling early after symptom clarifies plasma cytokine patterns (20), reminiscent of the Cytokine Release Syndrome (21) . Plasma profile around 10 days after symptom onset was highly differential for plasma cytokine profiles of critical versus moderate COVID-19 disease (21) and a number of cytokines have already been associated with increased mortality (22) , Multiple studies support a central role for antibody responses in protective anti-SARS-CoV-2 immunity. The main viral target of antibody immunity is the trimeric Spike glycoprotein, which facilitates SAR-CoV-2 entry into host cells via interaction of its receptor-binding domain 5

We investigated prospectively enrolled hospitalized COVID-19 individuals (n=144) with symptomatic infection and a positive SARS-CoV-2 nasopharyngeal swab PCR. These patients were infected during the first wave, and were not enrolled in immunotherapy trials. Our study population was split into a discovery cohort (n=61) in a first hospital and a fully independent validation cohort (n=83) in a second hospital (see Study Design, Figure S1A , and participant characteristics, Table 1 ). To allow for cross-sectional analysis of early plasma markers, we investigated patients for whom research blood samples were available at 11 (± 4) days after symptom onset (DSO11). Based on disease severity at DSO11, patients were grouped as critical (requiring mechanical ventilation) versus non-critical. The discovery cohort included 29 critical and 32 non-critical patients. Plasma profiles were compared to 50 asymptomatic uninfected donors as a control group (uninfected controls -UC) of non-diseased state.

We clinically followed participants for at least 60 days after symptom onset (DSO60). The primary outcome, death by DSO60, occurred in 13 patients (21.3%), with close to half fatalities occurring between DSO30 and DSO60 ( Figure S1B ) and mostly in the critical group ( Figure S1C ).

We performed a more focused immunovirological assessment in the validation cohort, where 15 cases were critical, and 68 non-critical, and 12 deaths occurred before DSO60. Because of hospital referral coordination, the validation cohort was older than the discovery one, but with less severe respiratory compromise (Table 1) . Other basic demographics and prevalent risk factors were consistent with published studies (28) and overall showed minor differences between both cohorts. These features were also not different between the critical vs non-critical groups except for higher rates of admission to ICU and intubation, and duration of hospital stay in critical patients (Table 1) , in line with group definition.

Plasma viral load in early disease is strongly associated with COVID-19 mortality 6 As SARS-CoV-2 vRNA in plasma has been previously linked to mortality, we quantified it in the discovery cohort. We designed an ultrasensitive quantitative real time PCR (qRT-PCR) targeting the N sequence of its genome with a detection limit of 13 copies/mL. The assay was highly specific, with no vRNA detected in UC ( Figure 1A ). At DSO11, we detected plasma SARS-CoV-2 vRNA in a significantly greater fraction of critical than non-critical patients (76% vs 28%, Figure 1A ). These results suggest that systemic SARS-CoV-2 viremia is a signature of infection severity and/or itself plays a role in disease complications.

We next hypothesized that the amount of viral products, rather than their mere presence, was associated with severe pathogenesis. SARS-CoV2 vRNA levels were higher in critical than non-critical cases ( Figure 1B) . This difference held when the comparison was restricted to samples with detectable plasma vRNA (p=0.002 -Mann-Whitney test). Most patients who died had high vRNA compared to survivors ( Figure 1C ). In univariate Cox regression analysis ( Table   2 ) we found that an increase of 1 unit of log-transformed plasma vRNA led to a 3-fold increase in mortality risk [Hazard Ratio (HR)= 3.0 (95% Confidence Interval CI: 1. .0), p < 0.0001 for all COVID-19 ( Figure 1D ), and 2.4 (95% CI: 1.3 -4.4, p=0.006) for the critical group (Table 2) ]. The estimated survival proportions for undetectable (<13 copies/ml), low, or high plasma vRNA were extracted from Cox models (see methods for details) (29) . High plasma vRNA was associated with a greater risk of death, whereas there was substantial overlap between the subgroups with low or undetectable plasma vRNA ( Figure 1E) . A similar trend was observed in the critical group ( Figure 1F ). Therefore, plasma SARS-CoV-2 vRNA load is not only a correlate of contemporaneous respiratory compromise early in disease course, but is also associated with mortality, including in the critical group.

As early elevation of a number of cytokines and chemokines was also associated with adverse COVID-19 outcome (20, 21, 30) , we used multiplexed beads arrays to determine plasma 7 levels of 26 proteins associated with adaptive and/or innate immune responses, chemotaxis, or tissue insult related to severe acute respiratory distress syndrome (ARDS, See Table S1 for analyte list). Principal component analysis revealed that the plasma profile largely delineates UC from COVID-19 patients, and highlighted higher cytokine levels and greater heterogeneity in the critical group compared to the non-critical group (Figure 2A ) The outlier critical case at the upper left corner of the PCA was on extracorporeal membrane oxygenation (ECMO) at the time of sampling, a procedure known to affect plasma profile (31) . Unsupervised hierarchical clustering parsed apart 3 patient clusters: I) mostly critical; II) mixed; III) mostly non-critical cases ( Figure   2B ).

We next compared the levels of each analyte between groups ( Figure S2A -D). Several followed a stepwise increase, where non-critical cases had greater cytokine concentrations than UC, and critical cases had the greatest amounts ( Figure S2A ). These included pro-inflammatory cytokines such as IL-6, GM-CSF and TNFa and pro-inflammatory chemokines CCL2 and CXCL8.

Some of the markers of tissue insult (RAGE, Angiopoietin-2)(32) also increased with disease severity, likely reflecting the extent of lung and vascular damage. CXCL9, CD40L, IFNa and surfactant pulmonary protein D (SP-D) were significantly greater only in the critical cases of COVID-19 compared to UC ( Figure S2B ), while a few markers did not differ between all three groups ( Figure S2C ). Some analytes were significantly elevated in COVID-19 groups but did not differ between the critical and non-critical groups, such as the chemokines CXCL10 (IP10) and CXCL13, and D-dimer ( Figure S2D ). Taken together, the plasma profile reveals overall higher quantities of cytokines in the plasma of COVID-19 patients compared to UC, and select analytes are specifically associated with greater disease severity.

We reasoned these 26 analytes may be differentially linked to the amount of vRNA in plasma. We examined the correlations between individual plasma analytes ( Figure S2E ), as well as their association with vRNA ( Figure 2C ). Many analytes were co-upregulated, and several of 8 them also positively correlated with vRNA levels. These latter correlations were particularly robust for cytokines implicated in innate immune responses such as IL-6 ( Figure S2F ) and GM-CSF, the marker of endothelial damage RAGE ( Figure S2G ), and inflammatory chemokines CXCL8, CXCL10, and CCL2, suggesting a shared trigger or overlap in pathways.

To capture by a single parameter the overall magnitude of the difference in cytokine titers between COVID-19 patients and UC, we created a "CytoScore" from the linear combination of all 26 analytes (See methods for details). CytoScore followed a gradual difference, where the noncritical group had lower CytoScores than critical, and UC had the lowest scores ( Figure 2D ). The CytoScore correlated positively with vRNA ( Figure 2E ) and can have value as a way to reduce dimensionality of plasma analyte profiling.

As patients who died within DSO60 showed a greater CytoScore than survivors ( Figure   2F ), we applied Cox regressing analyses to examine the association between the cytokines and mortality over time. We focused on analytes whose concentrations are in the range of robust quantitation by the assay (19/26, see methods for details). For each, we calculated the HR associated with a 1-unit increase of log-transformed concentration ( Figure 2G ). Several individual analytes were significantly associated with increased fatality risk, with Angiopoientin-2, RAGE and CXCL13 showing the highest significance (p < 0.001). As the CytoScore was also highly significant, we compared the predicted survival probability of patients with low or high CytoScore at DSO11 ( Figure 2H ). The latter population showed a significantly lower rate of predicted survival at DSO60 than the low CytoScore population. This observation was maintained when we restricted our analysis to the critical group ( Figure 2I ). Therefore, overall cytokine levels as well as individual cytokines and markers of tissue damage measured at DSO11 are 1) in majority correlated with plasma vRNA and 2) associated with increased risk of mortality among COVID-19 patients.

Low SARS-CoV-2-specific IgG and limited ADCC responses are associated with COVID-19 mortality As SARS-CoV-2 antibody responses likely play a critical role in protective immunity against SARS-CoV-2 (27, 33) , we measured plasma SARS-CoV-2-specific antibody responses at DSO11. ELISA-based quantification using the SARS-CoV-2 RBD protein and isotype-specific secondary antibodies (25, 34) revealed a broad range in relative quantities of RBD-specific IgM, IgA or IgG in the non-critical and critical groups at DSO11. They did not differ between groups, and were not detected in UC ( Figure 3A ). These observations were corroborated by a flow cytometry-based assay measuring plasma binding to full-length Spike protein (Spike Ig) on cell surface ( Figure 3B ), which similarly showed no significant difference between the two COVID-19 groups.

We next assessed the SARS-CoV-2 Spike-specific antibody response for two key antiviral functions: neutralization ( Figure 3C ) and antibody-dependent cellular cytotoxicity (ADCC, Figure   3D ). Here again, the data showed high variability, and no significant differences between the critical and non-critical groups for both readouts. All serology measurements were interrelated ( Figure 3E ). In contrast, the serology measurements were inversely correlated with plasma vRNA and most cytokines ( Figure 3F ).

To assess potential consequences of defective antibody responses at this early time point, we compared SARS-CoV-2-specific antibody responses between survivors and non-survivors.

For RBD-specific isotypes ( Figure 3G ), only IgG amounts were significantly increased in survivors, although there was a similar trend for IgA as well. Spike Ig levels were also higher in survivors ( Figure 3H ). We observed contrasting patterns with regard to functional humoral responses: while neutralization capacity was similar for the two outcomes ( Figure 3I ), ADCC capacity was superior in survivors ( Figure 3J ). HR reflected the same observations, where higher ADCC, RBD-specific IgG and Spike Ig were associated with increased survival ( Figure 3K ). We further modeled this by comparing the survival curves at DSO60 of patients with low or high RBD-. 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 March 20, 2021. ; https://doi.org/10.1101/2021.03.18.21253907 doi: medRxiv preprint specific IgG amounts ( Figure 3L ), Spike Ig ( Figure 3M ) or ADCC ( Figure 3N ) at DSO11, and saw that participants with low responses for these three measurements showed an increased fatality risk. These observations were maintained when the analysis was restricted to the critical group ( Figure S3A-C) . Taken together, these results highlight that impairment of some SARS-CoV-2specific antibody responses may contribute to mortality.

As all categories of immunovirological parameters showed some perturbations that predicted fatality, we examined whether these alterations provided redundant information in terms of mortality risk, or if their combined analysis would improve associations with fatal outcome.

Within immunovirological categories, we retained only variables significant in univariate Cox analysis (p < 0.05; see Table 2 ), and among those, a global multivariate model was used to select top variables (See methods for details). To evaluate predictive accuracy of the resulting variables and multivariate models, time-dependent receiver operator characteristic (ROC) curves were calculated at DSO60 (principles illustrated in Figure 4A , see methods for details). The area under the curve (AUC), a measure of prediction accuracy, was examined at all distinct event times by plotting the AUC curve over time (principles illustrated in Figure S5A , see methods for details). All final Cox models were reassessed in the validation cohort, and their time dependent ROC curves were evaluated to validate the accuracy of our findings.

In the discovery cohort, time-dependent ROC for plasma vRNA showed a strong predictive capacity at DSO60 (AUC=0.83, 95%CI: 0.72-0.94), and a slight benefit when adjusting for age and sex (AUC=0.86, 95%CI:0.74-0.98) ( Figure 4B ). The AUC for the adjusted model reached a high of 0.95 ( Figure 4G ) at DSO33 but mostly hovered around 0.90 ( Figure S5B ). When applied to the validation cohort at DSO60, vRNA again had a good predictive capacity (AUC=0.76;

. 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 March 20, 2021. 95%CI:0.58-0.95), and a benefit when adjusting for age and sex (AUC=0.85; 95%CI: 0.71-1.00) ( Figure S4A ). The highest AUC of the adjusted model (0.92; 95%CI:0.85-1.00) was reached at DSO22 ( Figure 4G ), but was stable over time ( Figure S5C ). Therefore, vRNA is a strong predictor of fatality, and adjusting for age and sex improves its predictive power.

Next, we compared the time-dependent ROC curves for inflammatory and tissue damage markers of the discovery cohort ( Figure 4C ). Multivariate model selection retained only 1 analyte:

Angiopoientin-2. To compare predictive accuracies at DSO60, we selected 3 additional analytes highly significant (p < 0.001) in univariate Cox ( Figure Figure S5F ). We then applied the analysis to the validation cohort.

The cell-based ADCC assay requires significant infrastructure and technical expertise that may not be available in all clinical settings. We therefore removed ADCC from the validation list of . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 20, 2021. ; 12 variables, leading to its substitution by the technically simple RBD-specific IgG, in line with their strong correlation ( Figure 3C ). The time-dependent ROC curves in the validation cohort for Spike Ig (AUC=0.62; 95%CI: 0.15-1.00) and RBD-specific IgG titers (AUC=0.61, 95%CI: 0.17-1.00) were non-significant, and lower than in the discovery cohort. However, they displayed good predictive accuracy of mortality with DSO60 when adjusted for age and sex (Spike Ig: 0.78, 95%CI: 0.54-1.00; RBD-specific IgG: 0.78, 95%CI: 0.53-1.00) ( Figure S4C ). Taken together, these data reveal that the anti-SARS-CoV-2 antibody response is highly associated with mortality within 30 days of symptom onset, but less so afterwards.

After examining each variable in the setting of their category, we sought to identify which single parameter, or combination thereof, is the most robust. All variables selected by multivariate model within each category were considered for a global multivariate model, and age and sex covariates were forced regardless of their significance. In the discovery cohort, the variables 0.71-1.00) and stable over time ( Figure S5I ).

Taken together, these data indicate that, at DSO11, measuring plasma SARS-CoV-2 vRNA in hospitalised COVID-19 patients can be a powerful tool to predict mortality.

. 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) 

In the perspective of clinical translation, it is essential to rigorously select among the multitude of markers linked to COVID-19-related mortality. In patients with a spectrum of disease severity, we studied perturbations within three categories of plasma molecules associated with death: circulating SARS-CoV-2 vRNA (16), elevated immune and tissue injury markers (30) and

inadequate SARS-CoV-2-specific antibody responses (27) , all of which can be probed by quick and technically robust assays. Strong associations of early parameters with our primary outcome, fatality within 60 days of symptom onset, were observed, and largely maintained when the analyses were restricted to the critical group of patients on mechanical ventilation. Multivariate analyses demonstrated that, because of collinearity between several variables, a limited number of biological features was sufficient to build robust models predicting mortality. stand out as an early feature critically, and consistently, associated with higher mortality risk. Combined analysis of SARS-CoV-2 vRNA, Angiopoietin-2, age and sex had greatest predictive accuracy in a discovery cohort, although a simpler model with vRNA, age and sex was almost as robust. This three-parameter model maintained significant predictive accuracy in an independent validation cohort. In our study, plasma vRNA levels therefore stand out as an early feature critically, and consistently, associated with higher mortality risk.

The strength of the association between plasma vRNA levels and mortality risk was stronger than previously reported for nasopharyngeal swabs (NSW) (35) . In contrast to plasma, quantification of vRNA in NSW is hard to normalize, varies between types of tests, and depends on sample quality. Cox models showed a 3-fold increase of fatal outcome for every 1-unit increase in log-transformed plasma vRNA quantity. While this association is reminiscent of the remarkable predictive value of plasma viral load for disease progression in untreated HIV-1 infection (36) , no study has thus far convincingly demonstrated that therapeutic reduction of SARS-CoV-2 viral loads decreased mortality risk. For example, the antiviral remdesivir reduced viral loads in NSW, duration of symptoms, and hospitalization, but had no significant impact on survival (37, 38) .

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The source and precise nature of the plasma vRNA remains to be better determined. Viral nucleic acids in the plasma do not prove the presence of replication-competent viral particles, as they could be viral debris translocated from damaged lung tissue. This is supported by the correlation we saw between vRNA and RAGE. Whereas the increase in plasma of RAGE's ligand EN-RAGE was linked to its' increased mRNA expression in the PBMCs of severe COVID-19 patients, RAGE mRNA was not expressed in this compartment (30) , supporting that plasma RAGE originates directly from damaged tissue (32) . Besides the direct cytopathic effects of SARS-CoV-2 on lung epithelium, immunopathological mechanisms are thought to play a key role in severe COVID-19 pathogenesis (41). Systemic vRNA may trigger pathogen-recognition receptors such as TLRs, in line with strong co-upregulation of interferon-stimulated genes (ISGs) and other inflammatory pathways in vRNA-containing cells (42) . This could contribute to the strong correlation observed between the amount of vRNA and IL-6, a pathogen-associated molecular pattern (PAMP)-triggered inflammatory cytokine (43) .

Consistent with previous studies (20, 21), we found significant associations between levels of several immune and tissue damage markers with both disease severity and mortality. Despite strongly significant HR for fatality risk for some analytes, the small sample size of our study resulted in sizeable overlaps between confidence intervals and variable rankings of HR values between the discovery and validation cohorts. The use of an integrated CytoScore partially compensated for individual marker variability by giving an overall assessment of the magnitude of the cytokine storm. Notable individual markers were associated with fatal outcome, namely . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 20, 2021. ;

Angiopoietin-2, CXCL13 and RAGE. While Angiopoietin-2 was less strongly correlated with vRNA than RAGE, it appears of significant interest in severe COVID-19. This angiogenic factor has proinflammatory effects on the vascular endothelium, can disrupt vascular integrity and has been associated with ARDS (44) . Angiopoietin-2 levels have been associated with severe COVID-19 (45) , and might be a potential druggable target.

Antibody responses likely contribute to viral control in acute SARS-CoV-2 infection, and the negative associations we observed between plasma vRNA and SARS-CoV-2-specific antibody responses support this model. Although there were no global differences in antibody levels between the critical and non-critical groups, fatality was differentially association with humoral responses. Mortality was overrepresented among patients whom, at DSO11, had low RBD-specific IgG and low total Spike-binding Ig, although not in those with low RBD-specific IgM response. As only the IgG isoform among RBD-specific antibodies is lower in deceased patients, there may be a disruption in B cell functions requiring T-cell help, like class-switching to IgG, possibly linked to inadequate T follicular helper (TFH) and/or germinal centers (GC) disruption (46) . CXCL13 is a key chemokine for recruitment to the GC of TFH and B cells (47) , and plasma CXCL13 is a marker of GC activity (48) . As such, the positive associations of CXCL13 levels with vRNA loads and fatality risk, and the inverse correlation of CXCL13 levels with antibody responses, may seem paradoxical, but high amounts of circulating CXCL13 might disrupt the dynamics of B cell recruitment to the GC. In addition, heightened systemic inflammation can impaired development of adaptive immunity (49, 50) . These mechanisms may explain the reduced RBD-specific IgG in patients who succumb to their infection.

Defective early ADCC responses were also significantly associated with fatality, whereas we found only a non-significant trend for neutralization capacity. These observations support that Fc-mediated functions could be important in controlling SARS-CoV-2, in line with recent reports . 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) showing that compromised Fc receptor binding strongly correlated with COVID-19 mortality (27), and Spike-specific humoral responses, including higher Fc-effector functions, were enriched among survivors (51) . Furthermore, antibodies with intact Fc-effector functions were required for optimal protection against infection and correlated with decreased viral loads in animal models (52, 53) .

The significant interactions we observed between a number of the features measured are compatible with different, non-mutually exclusive mechanisms. Poor development of protective antibody responses may allow persistently high levels of viral replication, which in turn will lead to a cytokine storm. Conversely, high cytokine levels, perhaps driven by systemic vRNA, may disrupt adaptive immune responses. Although our observational study does not allow addressing the question of causation between the immunovirological alterations observed, these measurements can be useful tools to understand heterogeneity in disease trajectories and response to therapy, particularly in the context of large, well-controlled randomized controlled trials. High viral loads and low levels of SARS-CoV-2-specific IgG may be mitigated through antivirals, monoclonal antibodies or convalescent plasma therapy with high IgG content. People with high levels of selected cytokines may benefit the most from targeted immunotherapies. Recent trials have already resulted in marked improvement in clinical patient care. As our study has been conducted on patients hospitalized during the first COVID-19 wave in spring and early summer 2020, it will be important to assess how t interventions that only recently became part of standard clinical care, as well as future therapeutic strategies, affect the potential of such immunovirological monitoring not only to predict outcome, but potentially to individualize patient management.

. 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. Every author has read, edited and approved the final manuscript.

. 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. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 20, 2021. ; † Values displayed are medians, with IQR: interquartile range in parentheses for continuous variables, or percentages for categorical variables. ‡ "Non critical illness" includes hospitalized patients with no oxygen support (no O2) (moderate disease) and oxygen support on nasal cannula (NC) only (severe, but non-critical disease)."Critical illness" includes hospitalized patients on mechanical ventilation, either : positive pressure non-invasive ventilation (NIV), endotracheal intubation (ETI), extracorporeal membrane oxygenation (ECMO . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 20, 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. 

All COVID-19+ Critical . 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. . 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. Medical charts were reviewed by two physicians for data collection on demographics, comorbidities, risk factors, severity state, time of infection, etc (see Table 1 ). Median age and range for UC cohort was 37 (32) (33) (34) (35) (36) (37) (38) (39) (40) (41) (42) (43) (44) (45) (46) , and 30 individuals were males (60%).

293T human embryonic kidney cells (obtained from ATCC) were maintained at 37°C under 5% CO2 in Dulbecco's modified Eagle's medium (DMEM) (Wisent) containing 5% fetal bovine serum (VWR) and 100 µg/mL of penicillin-streptomycin (Wisent). The 293T-ACE2 were previously reported (25) .

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 20, 2021. ; https://doi.org/10.1101/2021.03. 18.21253907 doi: medRxiv preprint For the generation of CEM.NKR CCR5+ cells stably expressing the SARS-CoV-2 Spike protein, transgenic lentiviruses were produced in 293T using a third-generation lentiviral vector system, as previously reported (54) . Briefly, 293T cells were co-transfected with two packaging plasmids were quantified by Nanodrop and the RNA copy numbers were calculated using the ENDMEMO online tool (http://www.endmemo.com/bio/dnacopynum.php). Aliquots of 1011 copies/μL were stored at -80°C. For each qPCR batch, one aliquot was thawed and six serial dilutions were prepared to generate a standard curve (500,000 to 5 copies per PCR well).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 20, 2021. ; https://doi.org/10.1101/2021.03.18.21253907 doi: medRxiv preprint

Never-thawed plasma aliquots were thawed at RT and SARS-CoV-2 virus was inactivated using 1% Triton-X100 for 2 hrs at RT. After inactivation, measurements were performed in duplicates Some cytokines and tissue damage markers were at very low concentrations, and the quantification platform we used was not sensitive enough to reliably them in most samples. As such, analytes with extrapolated values >90% and negative values>15% were identified by AE in Figures 2, 3 and S2.

. 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) For k analytes (n=26), the CytoScore for each sample was calculated as follows

where cn is the concentration for analyte n, 

The SARS-CoV-2 RBD assay used was recently described (25, 34) . Briefly, recombinant SARS-CoV-2 S RBD proteins (2.5 μg/mL), or bovine serum albumin (BSA) (2.5 μg/mL) as a negative . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

For evaluation of antibody-dependent cellular cytotoxicity (ADCC), parental CEM.NKr CCR5+ cells were mixed at a 1:1 ratio with CEM.NKr. Spike cells. These cells were stained for viability (AquaVivid; Thermo Fisher Scientific, Waltham, MA, USA) and cellular (cell proliferation dye eFluor670; Thermo Fisher Scientific) dyes to be used as target cells. Overnight rested PBMCs . 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 March 20, 2021. ;

were stained with another cellular marker (cell proliferation dye eFluor450; Thermo Fisher Scientific) and used as effector cells. Stained target and effector cells were mixed at a ratio of 1:10 in 96-well V-bottom plates. Plasma from COVID-19 or uninfected individuals (1/500 dilution) or monoclonal antibody CR3022 (1 µg/mL) were added to the appropriate wells. The plates were subsequently centrifuged for 1 min at 300xg, and incubated at 37°C, 5% CO2 for 5 to 6 hrs before being fixed in a 2% PBS-formaldehyde solution. ADCC was calculated by gating on Spikeexpressing live target cells and using the formula: patients, as well as critical patients' subgroup only. Analytes were log-transformed when they naturally followed exponential distribution, for example vRNA and cytokines. Next, the estimated survival proportions at any given point in time for a undetectable (when applicable), low lower . 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 March 20, 2021. ; https://doi.org/10.1101/2021.03.18.21253907 doi: medRxiv preprint interquartile range level of detectable) or high (upper interquartile range of detectable) were extracted from Cox models (29) and presented in the graphical form (29) .

Part 1. Multivariate Cox model. Potential risk factors were grouped in 3 categories: i) vRNA, ii) 26 cytokine variables and iii) 6 antibody variables. Model building was performed in three steps.

In the first step, univariate models for risk factor of death by DSO60 were developed, one for each of the covariates in the category; only risk factors p value <0.05 were retained. For the second category of 26 cytokines, an additional criterion of variable selection was applied to ensure the quality of the measurements: the cytokines with extrapolated values >90% and negative values>15% were excluded for future investigation. These exclusion criteria were added as the quantification platform we used was not sensitive enough to reliably quantify some lowconcentration analytes, and we wanted to rely on analytes which are well quantified for our multivariate model. 19 cytokines out of 26 were satisfied these criteria. In the second step, categories for which more than one variable had been retained in the first step were focused on;

then the stepwise Cox model selection based on the Bayesian Information Criterion (BIC) was used to obtain the most-parsimonious model (lowest value) for each of these categories. This penalized likelihood criterion selects the best variable at predicting data, then adds one additional variable at a time while accounting for potential overfitting, in the end only selecting the multivariate model with the lowest BIC value, i.e. the most parsimonious. In addition, to keep the risk of overfitting low, no more than six predictor parameters were entered in the multivariate model for our sample of 61 patients (57, 58) .

In the third step, all variables retained in the second step were considered; then the BIC was used to obtain a global parsimonious model. Based on the literature (59), age and sex are associated with the mortality for COVID-19 patients, however in the small homogeneous sample in might be hard to detect these relations. Thus, in each model Age and Sex covariates were forced in the multivariate model regardless their significance. Potential interactions between each covariate with age and sex were tested to verify if the effect is consistent across different age and . 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 March 20, 2021. ; https://doi.org/10.1101/2021.03.18.21253907 doi: medRxiv preprint 35 between sex. Potential presence of multicollinearity was assessed by calculating the variance inflation factor (VIF) for each variable. This allowed us to identify and treat in separate models subsets of covariates which were highly correlated.

To evaluate predictive accuracy of survival models the time-dependent receiver operator characteristic (ROC) curves for right-censored data (60) were calculated, compared across different Cox models and presented in the graphical form. The inverse probability of censoring weighting technique (IPCW) was used for estimating timedependent ROC curves (61) . The area under the curve (AUC) was examined at 60 days as well at all distinct event times by plotting the AUC curve and the 95% confidence limits over time. The day 48 corresponds the last event (fatality) day in the discovery cohort.

Part 3. Independent cohort validation. All final multivariate Cox models were reassessed in the validation cohort by executing independently the multivariate models with the same list of variables obtained, in the discovery cohort, in steps 2 and 3. Then, using the same approach described above, the time dependent ROC curves were evaluated in validation dataset to validate our finding.

. 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 March 20, 2021. ;

Association Between Administration of Systemic Corticosteroids and Mortality Among Critically Ill Patients With COVID-19: A Metaanalysis

Effect of Dexamethasone on Days Alive and Ventilator-Free in Patients With Moderate or Severe Acute Respiratory Distress Syndrome and COVID-19: The CoDEX Randomized Clinical Trial

Dexamethasone in Hospitalized Patients with Covid-19 -Preliminary Report

Interleukin-6 Receptor Antagonists in Critically Ill Patients with Covid-19 -Preliminary report. medRxiv

Tocilizumab in patients admitted to hospital with COVID-19 (RECOVERY): preliminary results of a randomised, controlled, open-label, platform trial. medRxiv

Early High-Titer Plasma Therapy to Prevent Severe Covid-19 in Older Adults

Viral loads in clinical specimens and SARS manifestations

Detection of SARS coronavirus in plasma by real-time RT-PCR

Respiratory syncytial virus (RSV) RNA loads in peripheral blood correlates with disease severity in mice

Respiratory syncytial virus lower respiratory disease in hematopoietic cell transplant recipients: viral RNA detection in blood, antiviral treatment, and clinical outcomes

Viral RNA in Blood as Indicator of Severe Outcome in Middle East Respiratory Syndrome Coronavirus Infection

Fatal outcome of human influenza A (H5N1) is associated with high viral load and hypercytokinemia

Clinical and virological factors associated with viremia in pandemic influenza A/H1N1/2009 virus infection

SARS-CoV-2 RNA in plasma is associated with ICU admission and mortality in patients hospitalized with COVID-19

High Frequency of SARS-CoV-2 RNAemia and Association With Severe Disease

SARS-CoV-2 viral load is associated with increased disease severity and mortality

SARS-CoV-2 viral load predicts COVID-19 mortality

Clinical features of patients infected with 2019 novel coronavirus in Wuhan

Cytokine prediction of mortality in COVID19 patients

A dynamic COVID-19 immune signature includes associations with poor prognosis

Longitudinal analyses reveal immunological misfiring in severe COVID-19

An inflammatory cytokine signature predicts COVID-19 severity and survival

Structure, Function, and Antigenicity of the SARS-CoV-2 Spike Glycoprotein

SARS-CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven Protease Inhibitor

Cross-Sectional Evaluation of Humoral Responses against SARS-CoV-2 Spike

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

Compromised Humoral Functional Evolution Tracks with SARS-CoV-2 Mortality

Clinical Characteristics and Severity of COVID-19 Disease in Patients from Boston Area Hospitals. medRxiv

Survival Analysis Part II: Multivariate data analysis -an introduction to concepts and methods

Systems biological assessment of immunity to mild versus severe COVID-19 infection in humans

COVID-19 and ECMO: the interplay between coagulation and inflammation-a narrative review

Biomarkers for Acute Respiratory Distress syndrome and prospects for personalised medicine

COVID-19-neutralizing antibodies predict disease severity and survival

Decline of Humoral Responses against SARS-CoV-2 Spike in Convalescent Individuals

SARS-CoV-2 viral load in nasopharyngeal swabs in the emergency department does not predict COVID-19 severity and mortality

Changes in plasma HIV-1 RNA and CD4+ lymphocyte counts and the risk of progression to AIDS. Veterans Affairs Cooperative Study Group on AIDS

Remdesivir for the Treatment of Covid-19 -Final Report

Repurposed Antiviral Drugs for Covid-19 -Interim WHO Solidarity Trial Results

SARS-CoV-2 Neutralizing Antibody LY-CoV555 in Outpatients with Covid-19

REGN-COV2, a Neutralizing Antibody Cocktail, in Outpatients with Covid-19

Tissue-Specific Immunopathology in Fatal COVID-19

COVID-19 immune features revealed by a large-scale single-cell transcriptome atlas

The Role of Interleukin 6 During Viral Infections

Plasma angiopoietin-2 in clinical acute lung injury: prognostic and pathogenetic significance

Dynamic angiopoietin-2 assessment predicts survival and chronic course in hospitalized patients with COVID-19

Loss of Bcl-6-Expressing T Follicular Helper Cells and Germinal Centers in COVID-19

Follicular helper CD4 T cells (TFH)

CXCL13 is a plasma biomarker of germinal center activity

The pathophysiology and treatment of sepsis

Polymicrobial Sepsis Chronic Immunoparalysis Is Defined by Diminished Ag-Specific T Cell-Dependent B Cell Responses

Distinct Early Serological Signatures Track with SARS-CoV-2 Survival

Human neutralizing antibodies against SARS-CoV-2 require intact Fc effector functions and monocytes for optimal therapeutic protection. bioRxiv

Antibody potency, effector function, and combinations in protection and therapy for SARS-CoV-2 infection in vivo

Longitudinal analysis of humoral immunity against SARS-CoV-2 Spike in convalescent individuals up to 8 months post-symptom onset. bioRxiv

Self, nCal: an R package for non-linear calibration

Simultaneous inference in general parametric models

Regression modelling strategies for improved prognostic prediction

Relaxing the rule of ten events per variable in logistic and Cox regression

Sex Differences on Clinical Characteristics, Severity, and Mortality in Adult Patients With COVID-19: A Multicentre Retrospective Study

Lecture Notes in Statistics

Evaluating Prediction Rules for t-Year Survivors with Censored Regression Models