One hundred and forty patients from the CORIMUNO-19 cohort (25) were included and divided into two groups: group 1: patients with severe pneumonia requiring at least 3L/min of oxygen (n=81), and group 2: patients with critical pneumonia requiring non-invasive or mechanical ventilation in intensive care unit (ICU) (n=59). These patients were included in three randomized clinical trials, one comparing usual care (UC) to UC plus an anti-IL-6 receptor antagonist (sarilumab), the second one comparing usual care (UC) to UC plus another anti-IL-6 receptor antagonist (tocilizumab), the last one comparing UC to UC plus an IL-1 antagonist (anakinra). The blood level of IFN-α2 was monitored longitudinally in all patients, and plasmatic viral load in almost all. PBMC were collected at baseline in one site (Bicêtre hospital) in 14 patients included in the Anakinra trial (26).

In order to determine whether IFN-α2 variation during disease evolution may have a predictive value, we collected samples at entry in the hospital (D1), day 3 (D3), D6, and D14 in most of the patients to assess blood IFN-α2. At D6, 105 of the patients had a second measurement. Among the 27 patients without any measurement à D6, 4 were dead (three in group 1 and two in group 2), and 23 missed the second sample (all in group 1).

During the 90-day follow-up of the study, 29 patients died, including 6 between D1 and D6. Since none of these drugs directly interfered with type I IFN production, we chose in the main analysis to pool the overall population regardless of the treatment.

We set up a national cohort of COVID-19 patients with moderate, severe, or critical pneumonia (CORIMUNO-19 Cohort, NCT04324047). This cohort was used to perform a series of randomized controlled trials testing different therapeutic regimens in COVID-19 patients. Two separate populations were recruited: group 1: patients with moderate-to-severe pneumonia with WHO 10-points Clinical Progression Scale [WHO-CPS] score 5 receiving at least 3L/min O₂ but without high flow oxygen (HFO), non-invasive ventilation (NIV), or mechanical ventilation (MV) and group 2: patients with critical pneumonia defined as WHO-CPS score 6 or more (i.e., with HFO, NIV or MV) in intensive care unit (ICU).

All patients included in the ancillary study presented in this work were included in the CORIMUNO-19 cohort in three randomized trials evaluating the efficacy of tocilizumab, sarilumab (25) and anakinra (26) in participating hospitals (Hôpital Bicêtre, AP-HP, Université Paris-Saclay; Hôpital Européen Georges Pompidou, Hôpital Bichat, and Hôpital Lariboisière, AP-HP, Université de Paris; Hôpitaux Universitaires de Strasbourg, Université de Strasbourg).

Plasma and/or serum samples were collected from 84 COVID-19 patients, and 20 healthy donors from a single hospital (Hôpital Bicêtre, Assistance Publique-Hôpitaux de Paris), and 56 COVID-19 patients included in other hospitals participating in the CORIMUNO-19 trials. These samples are part of the French-COVID collection.

Plasma and sera were analyzed using the Simoa technology with the Quanterix SIMOA HD-X analyzer (Quanterix Corporation, Lexington, MA, USA). The levels of plasma IFN-α2 were measured using the commercial kit for IFN-α2 quantifications according to the manufacturer’s instructions (Simoa® IFN-α Advantage Kit, reference 100860; Quanterix). The plasma and serum IFN-α2 levels were expressed in femtogramme per milliliters (fg/mL).

We had both sera and plasma available from 12 additional patients (not included in the analysis since we do not yet have definitive data on the clinical evolution). We could confirm a very good concordance for IFN-α2 for dosages above 100 fg/ml in the plasma. However, the technique was much more sensible with serum for the low levels than with plasma. We could design an algorithm to convert all the plasma values as serum values and analyze the whole cohort of patients together ( Figure S1 ).

Fresh blood samples were used to analyze the phenotype of circulating myeloid cells from COVID-19 patients (n=14) and healthy donors (n=7). Briefly, 200 µL of blood sample was first stained with the Fixable Viability Dye eFluorTM 780 (eBioscience) and then with fluorochrome-conjugated antibodies, including anti-CD45 (BV785; Biolegend, clone 2D1), anti-HLA-DR (APC-R700; BD Biosciences, clone G46-6), anti-CD66b (FITC; Biolegend, clone G10F5), anti-CD3 (FITC; Biolegend, clone SK7), anti-CD56 (FITC; Biolegend, clone NCAM16.2), and anti-CD19 (FITC; Biolegend, clone HIB19), anti-CD14 (BV650; Biolegend, clone M5E2), anti-CD16 (PerCP-Cy5.5; Biolegend, clone 3G8), anti-CD123 (PE; BD Biosciences, clone 7G3), anti-CD11c (BV421; Biolegend, clone 3.9), anti-CD141 (BV510; BD Biosciences, clone 1A4), anti-CD1c (BV711; Biolegend, clone L161). Following incubation with the antibodies, the lysis of red blood cells (RBC) was performed using 1X FACS Lysing Solution (BD Biosciences), and the samples were fixed with 1% PFA. Flow cytometry was performed on a BD LSRFortessa instrument using FACSDiva software (BD Biosciences), and the data were analyzed using FlowJo software (Treestar, San Carlos, CA). DC populations were defined as follows: pDC (CD14 ^- CD16 ^- CD123 ⁺ CD11c ^- ), cDC1 (CD14 ^- CD16 ^- CD11c^low, CD141 ⁺ CD1c ⁺ ), and cDC2 (CD14 ^- CD16 ^- CD11c ⁺ , CD141 ^- CD1c ⁺ ). The flow cytometry strategy is indicated in Figure S2 .

Luciferase reporter assays. The blocking activity of anti-IFN-α2 and anti-IFN-ω auto-Abs was determined with a reporter luciferase activity. Briefly, HEK293T cells were transfected with a plasmid containing the Firefly luciferase gene under the control of the human ISRE promoter in the pGL4.45 backbone and a plasmid constitutively expressing Renilla luciferase for normalization (pRL-SV40). Cells were transfected in the presence of the X-tremeGene9 transfection reagent (Sigma-Aldrich, ref. number 6365779001) for 24 hours. Cells in Dulbecco’s modified Eagle medium (DMEM, Thermo Fisher Scientific) supplemented with 2% fetal calf serum (FCS) and 10% healthy control or patient serum/plasma (after inactivation at 56°C for 20 minutes) were either left unstimulated or were stimulated with IFN-α2 (Milteny Biotec, ref. number 130-108-984), IFN-ω (Merck, ref. number SRP3061), at 10ng/mL or 100pg/mL, or IFN-β (Milteny Biotech, ref. number: 130-107-888) at 10ng/mL, for 16 hours at 37°C. Each sample was tested once for each cytokine and dose. Finally, cells were lysed for 20 minutes at room temperature, and luciferase levels were measured with the Dual-Luciferase® Reporter 1000 assay system (Promega, ref. number E1980), according to the manufacturer’s protocol. Luminescence intensity was measured with a VICTOR-X Multilabel Plate Reader (PerkinElmer Life Sciences, USA). Firefly luciferase activity values were normalized against Renilla luciferase activity values. These values were then normalized against the median induction level for non-neutralizing samples and expressed as a percentage. Samples were considered neutralizing if luciferase induction, normalized against Renilla luciferase activity, was below 15% of the median values for controls tested the same day.

As previously published plasma samples were extracted using the Cellfree200 V7 DSP 200 protocol with the QIAsymphony® DSP virus/pathogen mini kit (QIAGEN, UK) (27). Samples loaded onto the QIAsymphony® SP as instructed by the manufacturer, with a 200 μl sample input volume (composed of 150µl of plasma and 50µl of NaCl) and 60 μl elution output volume of AVE buffer, unless stated (QIAGEN, UK). The samples were then stored at -80°C until the droplet digital PCR (ddPCR) step. SARS-CoV-2 RT-ddPCR assays were performed using the One-Step RT-ddPCR Advanced Kit for 90 Probes (Bio-Rad Laboratories, Hercules, CA, USA) and the QX200 ddPCR platform (Biorad). A 2-plex RT-ddPCR assay was developed, which targets the Nucleocapsid (N1) gene of the SARS-CoV-2 positive-strand RNA genome with a specific FAM- probe, and primers Cy5-labeled probe for the detection of a human housekeeping gene (RNAseP). RNAseP positivity was necessary to validate the RT-PCR assay prior to any further analysis. Briefly, 9.9 μL of extracted RNA was diluted in a 22 μL final reaction volume containing 5.5 μL of One Step SuperMix, 2.2 μL of Reverse Transcriptase, 1.1 μL of 300mM DTT (One-Step RT-ddPCR Advanced Kit for Probes, Bio-Rad), 1.1 μL of primers and probes mix (final probe concentration: 200 nM each, final primer concentration: 600 nM each) and 2.2 µL (QS) of nuclease-free water. Then, each sample was primarily partitioned into 13000 to 20,000 droplets using the QX200™ Droplet Generator™ (Bio-Rad). PCR amplification was then performed on a C1000 Touch™ thermal cycler (Bio-Rad). The droplet reading and quantification were performed using the QX200™ Droplet Reader™, and data analysis was performed using the QuantaSoft™ Analysis Pro software (Bio-Rad). Plasma samples were considered of positive SARS-CoV2 RNA if RNAseP concentration was detected above 3.4 log copies per mL and SARS-CoV-2 RNA concentration above the assay limit of detection of 70 copies per mL.

Kruskall Wallis test, followed by a Dunn’s test, was used to compare unpaired multiple groups and determine p values. For paired data comparison of several groups, one-way ANOVA was performed, followed by Dunnett’s post-test to determine the p values between groups. For paired data comparison of two groups (severe vs. critical patients), divided into subgroups (D1, D3, D6, D14), two-way ANOVA was used, followed by Sidak’s post-test to determine p values in between each subgroup (comparison of the IFN-α level between the different time points). For paired data comparison of two groups (severe vs. critical) divided into subgroups (D1, D3, D6, D14), a two-way ANOVA test was used, followed by Tukey’s post-test to determine p values between the different time points compared to D1). Because the longitudinal serum IFN-α2 levels and the process of death could be interrelated, we used a joint model for the longitudinal process of IFN-α2 (after log transform) and survival between D1 and D6 was used to estimate individual slopes of variation of IFN-α2 levels between D1 and D6 while accounting for the informative censoring due to death and the association of IFN-α2 level with the death process. In this model, given the small number of longitudinal measurements between D1 and 6, a linear mixed effects model was used for the longitudinal process, with random intercept and slope. A proportional hazards model was assumed for the survival process for the time-specific IFN-α2 values, and the baseline hazard was approximated with splines. The model was fitted under a Bayesian approach using the JMBayes R package (28), accounting for the quantification limit of IFN-α2 (5 fg/ml in the serum and 96.37 fg/ml in the plasma). Slopes were then dichotomized at the median value to separate the half patients with the steepest IFN-α2 relative decrease between D1 and D6 (slope < median) and the others (slope > median). A landmark survival analysis with a landmark at 6 days was then used to compare the survival of both groups. This entails analyzing individuals alive at D6 only but avoids time-dependent bias. Statistical analyses were performed with R 3.6.3 (The R Foundation for Statistical Computing, Vienna, Austria) and GraphPad Prism.

All methods were performed in accordance with the relevant guidelines and regulations. The trial was approved nationally by the ethics committee on March 23, 2020 (file #20.03.20.56342, CPP Île De France VI, EudraCT: 2020-001246-18), by the French Medical Products Agency and by the Commission Nationale Informatique et Liberté. Written informed consent was obtained from all patients or from the patient’s legal representative if the patient was too unwell to consent to enter the CORIMUNO Cohort.

To better understand the role of type I IFN in the evolution of COVID-19, we performed a longitudinal study of blood IFN-α2 levels in 140 patients from the CORIMUNO-19 cohort (NCT04324047) divided into two groups, group 1: patients with severe COVID-19 requiring at least 3L/min of oxygen (n=81) and group 2: patients with critical COVID-19, requiring high-flow, non-invasive (NIV) or mechanical ventilation (MV) in intensive care unit (ICU) (n=59) (25). Twenty healthy blood donors were used as controls. The patients’ and controls’ baseline characteristics are indicated in Table 1 .

The baseline (at entry in the hospital [D1]) blood level of IFN-α2 was elevated in both groups compared to controls with higher levels in severe patients compared to critical patients ( Figure 1 ). The results remain the same if we consider separately dosages made on serum and plasma ( Figure S3 ). Baseline and day 6 blood IFN-α2 levels were not associated with survival ( Figure S4 ).

The anti-IFN-α2 auto-Abs detected were neutralizing against 10 ng/mL IFN-α2 in 3.7% and 5.2% of severe and critical patients, respectively, and against 100 pg/mL IFN-α2 in 6.2% and 5.2%, of severe and critical patients, respectively. The presence of neutralizing anti-IFN-α2 auto-Abs to both doses of IFN-α2 was associated with a lower level of blood IFN-α2 at baseline ( Figure S5A ). Interestingly, in these patients, the blood IFN-α2 levels were persistently low, although not associated with subsequent mortality either in severe or critical patients ( Figure S5B ). If we consider anti-IFN-α2 or anti-IFN-ω, neutralizing anti-IFN auto-Abs were present at high and low concentrations (10 ng/mL and 100 pg/ml) in 5% and 19.4% of patients, respectively, but with no correlation with the level of blood IFN-α2 and with subsequent mortality.

In the peripheral blood of a subset of patients (n=14), compared with healthy donors (n=7), the percentage of pDC was low. In contrast, the percentage of conventional dendritic cells (cDC1) and cDC2 was normal ( Figure 2 ). A low percentage of peripheral blood pDC was present in most of these patients with severe or critical COVID-19 pneumonia. This decrease in pDC was not correlated with the baseline, D6, or the D6/D1 ratio of plasma IFN-α2 level. The baseline percentage of pDC was not predictive of subsequent mortality.

In order to determine whether IFN-α2 variation during disease evolution may have a predictive value, we collected samples over time to assess blood IFN-α2. Blood IFN-α2 levels decreased in both groups of COVID-19 patients over time ( Figures 3A , S6 ) and did so regardless of the treatment group (usual care (UC), sarilumab, tocilizumab, or anakinra) ( Figure 3B ). Baseline (D1) IFN-α2 was not different between groups and treatment did not induce different changes in IFN-α2 between groups (data not shown). Thus, we merged the treatments group and analyzed the severe and critical groups. The magnitude of the decrease measured by the D6/D1 blood IFN-α2 ratio was significantly higher in severe patients who subsequently died than in patients who survived (p=0.026) ( Figure 4 ). Separated dosages on serum and plasma are shown in Figure S7 . Interestingly, the D6/D1 blood IFN-α2 ratio was predictive of subsequent mortality only in patients without neutralizing anti-IFN-α2 auto-Abs (data not shown), probably because IFN-α2 levels were already low at baseline in patients with neutralizing anti-IFN-α2 auto-Abs ( Figure S5 ).

Since computing the D6/D1 ratio is complicated by values under the limit of quantification (LOQ) and because the longitudinal serum IFN-α2 levels and the process of death could be interrelated, we used a joint model for the log IFN-α2 levels and death during the first 6 days in order to estimate a slope of log IFN-α2 for each individual, thereby providing similar information as the D6/D1 ratio ( Figure 5A ). Using a landmark analysis and dichotomizing the individual slopes at their median value, we then found that the half of patients with the steepest slope (lowest D6/D1, n=66) had higher mortality than the others (n=65): HR=3.15 (95% CI 1.14–8.66) in rapid decliners versus slow decliners (p=0.019) ( Figure 5B ), without interaction between severe and critical patients (p=0.21) ( Figures 5C, D ).

Viral load was detectable in plasma at baseline (mean 10 days from the first symptoms) in 70/140 (50%) patients and significantly decreased progressively over time in both groups ( Figures 6A, B ). Interestingly, severe patients who died after day 6 displayed a higher viral load at D1 and D6 than patients who survived ( Figures 6C, D ). SARS-CoV-2 viral load and serum IFN-α2 level correlated at admission (D1) ( Figure 6E ) but not at D6 ( Figure 6F ).