Main

Most patients with COVID-19 have only a few mild symptoms, but about 15% of patients progress to severe pneumonia, and about 5% develop acute respiratory distress syndrome (ARDS), for which effective therapeutic strategies are urgently required2. The immune system has a dual role in the pathology of COVID-19, contributing to both virus elimination and the development of ARDS2. A detailed characterization of the immune responses that occur during disease progression from mild to severe forms is therefore important for understanding the ways in which we can manipulate immunity to propose new therapies. In particular, given the urgent need for effective treatments for pneumonia in patients with COVID-19, the elucidation of the immune responses that occur during the course of COVID-19 could lead to the repurposing of approved immunomodulatory drugs and candidate drugs that have already been tested in clinical trials. We therefore monitored immune parameters in a cohort of 82 individuals: 10 healthy control individuals, 10 patients with COVID-19 who were paucisymptomatic, 34 patients with pneumonia and 28 patients with ARDS due to SARS-CoV-2 infection (Supplementary Table 1). We focused on molecular pathways that could block the overt inflammation associated with ARDS.

Disease severity was associated with an increase in the amounts of plasma C-reactive protein (CRP) and inflammatory cytokines—such as interleukin-6 (IL-6), and the chemokines CCL4 (macrophage inflammatory protein-1β), CCL2 (monocyte chemoattractant protein 1) and CXCL9 (monokine induced by interferon-γ)—that are produced by and act on myeloid cells (Fig. 1a). These results confirmed previous observations of the ‘cytokine storm’ that develops in patients with severe COVID-193. The ability of plasma from patients to neutralize the SARS-CoV-2 virus was also correlated with disease severity (Extended Data Fig. 1a), consistent with previous studies, which show higher titres of anti-SARS-CoV-2 antibodies in patients with severe COVID-194.

Fig. 1: Inflammation is associated with a cytokine storm and C5a production in patients with COVID-19.
figure 1

a, Concentrations of CRP, IL-6, CCL4, CCL2 and CXCL9 in plasma from healthy control individuals (HC) and patients with COVID-19. b, Concentration of C5a desArg in plasma of HC and patients with COVID-19. Healthy control individuals, n = 10; patients who were paucisymptomatic (pauci), n = 10; patients with pneumonia (pneumo), n = 31–34; patients with ARDS, n = 26–28. For CRP (a) and b, P values were computed using two-tailed Wilcoxon rank-sum tests. For IL-6, CCL4, CCL2 and CXCL9 (a), a global comparison was first performed in which all values were classified into two categories: above or below the limit of quantification (P values for two-sided Fisher’s exact tests show that there is an increase in the number of values above the limit of quantification with increasing severity); Fisher tests: P = 5 × 10−7 for IL-6, P = 0.02 for CCL4, P = 0.001 for CCL2, P = 9 × 10−6 for CXCL9. A comparison was then performed between patients with pneumonia and with ARDS, only for values above the limit of quantification (P values were computed using two-tailed Wilcoxon rank-sum tests). Each symbol represents a single donor. Box plots show the median and the 25th to 75th percentiles, and the whiskers denote the maximum and minimum values. The dotted lines for IL-6, CCL2 and CXCL9 (a) indicate the limit of quantification.

We decided to focus on the complement factor C5a, which mediates strong chemoattraction and activation of myeloid cells5, and has well-documented roles specifically in lung inflammation and injury6. The complement cascade is important for the sensing and clearance of pathogens and inflammation7, and involves several components, including cell surface receptors and soluble regulators. In the final phase of the response, the membrane attack complex (MAC: C5b9) and the potent chemoattractants and inflammatory mediators C3a and C5a are generated. The MAC forms transmembrane channels on the surface of pathogen cells, disrupting the cell membrane and leading to cell death. The C3a and C5a proteins regulate inflammation by binding to their respective receptors, C3aR and C5aR18. Exaggerated complement activation contributes to the pathogenesis of many inflammatory and immune diseases5. Numerous studies of the lung epithelium have reported depositions of complement components during inflammation and suggested that the systemic activation of complement leads, through C5a, to the recruitment, activation and adhesion of neutrophils to the pulmonary endothelium, which results in cell damage, and subsequent acute lung injury (ALI) and ARDS, which can be fatal1,6,9. We observed an increase in plasma C5a levels that was proportional to COVID-19 severity (Fig. 1b). C5a levels increased in a few patients in the paucisymptomatic group, and were significantly higher in individuals with lung damage in the pneumonia and ARDS groups than in healthy controls (Fig. 1b). The longitudinal follow-up of patients with COVID-19 revealed that the upregulation of circulating C5a levels was maintained for at least 10 days after the inclusion of the patients in our cohort (Extended Data Fig. 1b). The higher level of C5a in the patients with the most-severe symptoms suggests a role for this anaphylatoxin in the inflammation that occurs in patients who develop ARDS. Increased systemic and local complement pathway activity was confirmed by transcriptomic analysis of peripheral blood from patients with COVID-19, which showed an upregulation of C1Q and C2 expression (Extended Data Fig. 1c) and by the presence of C5b9, as shown by immunostaining, in lung sections from patients with COVID-19 (Extended Data Fig. 1d). Consistent with these results, high levels of C5a in patients with COVID-19 have recently been reported to be a consequence of overt activation of the lectin pathway of the complement cascade by the N protein of SARS-CoV-210. Furthermore, anti-SARS-CoV-2 antibodies4 and CRP may also contribute to the activation of the classical and alternative pathways of complement during COVID-19. Thus, factors that trigger the activation of the complement pathways are upregulated in COVID-19 and may sustain the high levels of C5a that are detected in patients with severe COVID-19.

We found that COVID-19 was associated with peripheral blood neutrophilia (Fig. 2a), as reported in other cohorts11. No major changes were observed in the total peripheral blood monocyte population (Fig. 2b), but the proportion of conventional CD14+CD16 monocytes was increased, whereas the proportion of inflammatory CD14lowCD16+ monocytes was decreased in peripheral blood (Fig. 2c), consistent with the possibility that inflammatory monocytes leave the bloodstream and home to tissues. This hypothesis was supported by transcriptomic analyses of patients with COVID-19 who were symptomatic, which revealed not only an increase in the transcript levels of genes including IRAK3, MS4A6A, CD33, CD300C, VCAN, CD1D, CCR1, OAS1, CD163 and C3AR1 in peripheral blood mononuclear cells, but also an upregulation of macrophage and monocyte transcriptomic signatures in inflamed lungs (Extended Data Fig. 1c).

Fig. 2: C5aR1 is highly expressed on myeloid cells and promotes inflammation.
figure 2

a, Absolute numbers of circulating neutrophils per microlitre of peripheral blood from healthy control individuals and patients with COVID-19 at t0. b, Absolute numbers of circulating total monocytes per microlitre of peripheral blood from healthy control individuals and patients with COVID-19 at t0. c, Percentages of CD14+CD16 conventional monocytes and CD14lowCD16+ inflammatory monocytes in the peripheral blood of healthy control individuals and patients with COVID-19 at t0. d, Percentage of C5aR1+ neutrophils and monocytes in peripheral blood from healthy control individuals and patients with COVID-19 at t0. ad, Healthy control individuals, n = 10; patients who were paucisymptomatic, n = 10; patients with pneumonia, n = 31; patients with ARDS, n = 26. Each symbol represents a single donor. e, IL-6, TNF and CCL2 production by monocytes purified from peripheral blood mononuclear cells from patients with COVID-19. Peripheral blood mononuclear cells were activated overnight with LPS (0.5 ng ml−1) or R848 (50 ng ml−1) and recombinant human C5a (HsC5a) (1 μg ml−1), as indicated. Each dot represents the mean value obtained from duplicate or triplicate analyses for a single patient (n = 12 patients). Box plots show the median and the 25th to 75th percentiles, and the whiskers denote the maximum and minimum values. The P values were obtained using two-tailed Wilcoxon rank-sum tests (ad) and two-tailed Wilcoxon signed-rank tests (e). NS, no significant differences.

Both circulating neutrophils and monocytes displayed strong C5aR1 expression in healthy individuals that was also observed in the various groups of patients with COVID-19 (Fig. 2d). Longitudinal immune-monitoring follow-up of patients with pneumonia and ARDS showed that the levels of C5aR1 molecules remained stable on circulating neutrophils and monocytes, or even increased during the course of the disease (Extended Data Fig. 2a). Consistent with the inflammatory function of C5a12, and the expression of C5aR1 on monocytes, C5a increased the production of the inflammatory cytokines IL-6, TNF and CCL2 that is induced by lipopolysaccharide (LPS) on purified blood monocytes isolated from patients with COVID-19 (Fig. 2e). C5a also increased the production of cytokines by monocytes from patients with COVID-19 following stimulation with R848, which activates the TLR7 and/or TLR8 MyD88-dependent signalling pathway, mimicking the single-stranded RNA of SARS-CoV-2 (Fig. 2e).

Given that severe COVID-19 is associated with lung disease, we then focused on this organ. C5a was detected in the bronchoalveolar lavage fluid (BALF) of patients with COVID-19 who had ARDS (Fig. 3a). Inflammatory cytokines, such as CXCL8, CXCL9, CCL2 and—to a lesser extent—CCL4, IL-6, TNF and IL-1β, were also detected in these samples (Fig. 3a). Large numbers of neutrophils and monocytes were found in the BALF of patients with COVID-19 who had ARDS and these cells expressed C5aR1 (Fig. 3b). In addition, the analysis of single-cell RNA-sequencing data from healthy control individuals and patients with COVID-19 who had ARDS13 revealed major changes in the myeloid cell population that infiltrated the lungs during the course of the SARS-CoV-2 infection. Indeed, the major myeloid cell subset (subset A) in healthy control individuals and the major myeloid cell subset (subset B) in patients with COVID-19 who had ARDS were clearly different (Fig. 3c and Extended Data Fig. 3a, b). The cells of subset B were characterized by higher transcript levels of inflammatory cytokine genes, such as CXCL8, CCL2, CCL4, CXCL9, TNF and IL-6 (Fig. 3d). Both myeloid cell subsets expressed C5AR1, although a slight upregulation was observed in subset B (Fig. 3d). A multiplex immunohistochemistry analysis of lungs from patients with ARDS with COVID-19 who were deceased confirmed pulmonary infiltration of CD68+CD163+ macrophages, a substantial proportion of which expressed C5aR1 (Extended Data Fig. 3c (right)), relative to lung tissue from a control individual without COVID-19 (Extended Data Fig. 3c (left)). It is becoming increasingly clear that severe COVID-19 is associated with the spread of the virus through the epithelial barrier and with endothelialitis14,15,16. We observed obliterating endarteritis that was associated with an accumulation of C5aR1+ macrophages around the arteries and in the thrombus (Extended Data Fig. 3d). Together with the high levels of C5a in patients with COVID-19 who were symptomatic, these data support the hypothesis that C5a production leads to the chemo-attraction and activation of myeloid cells in the lungs, and contributes to the overt release of inflammatory cytokines. As C5a can also promote the secretion of CCL2, a strong chemoattractant for monocytes, C5a could also indirectly promote the recruitment of inflammatory cells indirectly, through the induction of other chemokines. It is also possible that the vasculitis associated with severe COVID-19 is linked to the production of C5a, as other types of vasculitis, such as anti-neutrophil cytoplasmic antibody (ANCA)-associated vasculitis, are mediated by C5a17.

Fig. 3: C5aR1+ cells and C5a are detected in lung samples from patients with COVID-19.
figure 3

a, Concentration of C5a desArg and IL-6, CXCL9, CCL2, CCL4, CXCL8, TNF and IL-1β in the BALF of patients with ARDS (n = 4). #Values were above the detection limit of 52,200 pg ml−1; $values were below the detection limit of 29 pg ml−1. Box plots show the median and the 25th to 75th percentiles, and the whiskers denote the maximum and minimum values. b, Left, CD45+ immune cell infiltration in BALF from patients with ARDS using flow cytometry. Right, C5aR1 expression (red) compared with negative control (fluorescence without antibody; black) staining on CD45+CD14CD15+CD16+ neutrophils and CD45+CD33+HLA-DR+CD14+ monocytes in BALF from patients with ARDS. The plots shown are representative of analyses performed on samples from three patients with ARDS. c, Uniform manifold approximation and projection (UMAP) analysis of myeloid cell clusters by patient group. Healthy control individuals, n = 3; patients with severe COVID-19, n = 6. d, Dot plot of pro-inflammatory cytokine expression in each of the subsets from c.

On the basis of this hypothesis, we reasoned that blockade of the C5a–C5aR1 axis could be used as a potential therapeutic strategy for the treatment of severe COVID-19. Several molecules could be repurposed to this end, including anti-C5 monoclonal antibodies, anti-C5a monoclonal antibodies and C5aR1 antagonists. We focused on avdoralimab, a fully human Fc-silent monoclonal antibody against C5aR1 that prevents its binding to C5a. In vitro, C5aR1 blockade with avdoralimab inhibited C5a-induced activation of neutrophils, as shown by evaluation of the induction of CD11b expression at the cell surface (Fig. 4a). The C5a-mediated upregulation of CD11b was also inhibited by other C5aR1 antagonists or anti-C5a monoclonal antibodies (Extended Data Fig. 4a). Avdoralimab blocked the activation of neutrophils induced by very high concentrations of C5a (Fig. 4b). With regard to the infiltration of C5aR1+ myeloid cells in the inflamed lungs of patients with severe COVID-19, avdoralimab also inhibited the C5a-induced migration of neutrophils in vitro (Fig. 4c). We next investigated whether avdoralimab could block the development of ALI in a mouse model. As avdoralimab targets human C5aR1, we used mice with a knock-in gene of human C5aR1 (HsC5AR1 knock-in mice)18 that showed HsC5aR1 expression exclusively on CD11b+ myeloid cells (Extended Data Fig. 4b). HsC5AR1 knock-in mice that received an intranasal instillation of recombinant human C5a developed ALI and injury 18 h after injection, as shown by the infiltration of CD45+ immune cells including Ly6G+Ly6C+ neutrophils and Ly6GLy6C+ monocytes in the lung (Fig. 4d) and the release of albumin in BALF (Fig. 4e), a marker of alveolar-capillary permeability and lung injury. ALI was confirmed by histopathological analysis of lung sections that revealed massive inflammatory cell infiltration, alveolar haemorrhage and thickening of alveolar walls in lungs of C5a-treated mice (Extended Data Fig. 4c), as previously described9. Avdoralimab blocked the infiltration of both cell types (Fig. 4d), prevented albumin release in BALF (Fig. 4e) and limited C5a-induced ALI histopathological features (Extended Data Fig. 4c). Finally, avdoralimab also inhibited the increase in IL-6, TNF and CCL2 secretion induced in vitro by C5a in monocytes purified from patients with COVID-19 and activated with a single-strand RNA-virus-like stimulus (Fig. 4f). Therefore, avdoralimab seems to be suitable for blocking the C5a–C5aR1 axis, which is active during COVID-19. The high levels of C5a observed in patients with COVID-19 do not appear to be a passenger phenomenon, as preliminary efficacy data reported that two patients with COVID-19 recovered from ARDS after treatment with an anti-C5a blocking monoclonal antibody (IFX-1)10. Furthermore, four patients with severe COVID-19 treated with an anti-C5 monoclonal antibody (eculizumab) exhibited a drop in circulating inflammatory markers19. There are several advantages to blocking C5aR1 rather than other components of the complement cascade. First, blocking C5a or C5aR1 leaves C5b intact and preserves the MAC, which has a key role in controlling several infections. A loss of the MAC would raise safety concerns for the treatment of COVID-19, as patients who are symptomatic often develop comorbid conditions, such as bacterial infections, for which the MAC is required. Second, C5aR1 blockade has the advantage over C5a blockade of having no effect on the second C5a receptor, C5L2. The function of C5L2 remains unclear, but it has been suggested that it can act as a decoy receptor, with anti-inflammatory roles20.

Fig. 4: Targeting C5aR1 blocks the C5a-mediated activation and migration of myeloid cells.
figure 4

a, Representative flow cytometry histogram of the C5a-mediated upregulation of CD11b on whole-blood neutrophils, inhibited by avdoralimab. Grey, non-activated neutrophils; black, recombinant human C5a (HsC5a)-activated neutrophils; purple, HsC5a-activated neutrophils treated with avdoralimab (30 μg ml−1). b, CD11b induction through a dose-dependent response to HsC5a on untreated (grey) or 10 μg ml−1 avdoralimab-treated whole-blood neutrophils (purple). MFI, median fluorescence intensity. Data were obtained from two independent donors. c, Migrating neutrophils attracted by HsC5a. Neutrophils were treated with 10 μg ml−1 isotype control (IC) or avdoralimab (avdo) before the induction of migration by HsC5a. Data for neutrophils purified from four healthy control individuals are shown. d, e, BALF was collected from HsC5AR1 knock-in mice 18 h after intranasal instillation of HsC5a. Mice were pretreated with avdoralimab or isotype control, as indicated. Each symbol represents data from a single mouse; n = 5 mice (PBS) and n = 7 mice (all other groups). Data are representative of two experiments. d, BALF cell analysis by flow cytometry of CD45+ immune cells, Ly6C+Ly6G+ neutrophils and Ly6C+Ly6G monocytes. e, Albumin concentration in BALF. f, Production of IL-6, TNF and CCL2 by monocytes purified from the peripheral blood mononuclear cells of patients with COVID-19 activated by overnight incubation with R848 (50 ng ml−1) and HsC5a (1 μg ml−1). Before activation, monocytes were incubated with 20 μg ml−1 avdoralimab or isotype control. Each dot represents the mean value obtained from duplicate or triplicate analyses of a single donor; n = 10 (IL-6) and n = 12 (TNF and CCL2). Box plots show the median and the 25th to 75th percentiles, and the whiskers denote the maximum and minimum values. The P values were computed using a paired one-way analysis of variance (ANOVA) (c), two-tailed Wilcoxon rank-sum tests (d, e) and two-tailed Wilcoxon signed-rank tests (f).

Our data, which highlight the role of the C5a–C5aR1 axis in the pathogenesis of severe COVID-19, are consistent with several previous observations. Indeed, high C5a levels have also been described in various preclinical models of acute lung disease due to pathogenic viruses, such as SARS-CoV, H1N1, H5N1 and H7N91. High levels of C5a have also been found in the upper respiratory tract and in serum samples from patients infected with H1N121. Furthermore, BALF from patients with ARDS was found to display robust C5a-dependent chemotactic activity22. In a mouse model of Middle East respiratory syndrome coronavirus (MERS-CoV) infection, C5a concentrations were high in serum samples and lung tissues23, and anti-C5aR1 antibody treatment decreased viral replication in lung tissue23. In a green monkey model of H7N9 infection, treatment with an anti-C5a antibody significantly decreased the levels of IL-6, interferon-γ, TNF and IL-1β and neutrophil infiltration in the lungs24. Overall, C5a inhibition markedly decreased the ALI and systemic inflammation induced by viral infection25. Further evidence for the involvement of the complement system in the pathogenesis of severe COVID-19 is provided by the existence of genetic variants of the complement cascade, which are associated with clinical outcome of SARS-CoV-2 infection, such as CD5526.

In addition, the complement system links innate immunity to coagulation27,28, and its overactivation could promote thrombotic events in patients with severe COVID-1929. Complement blockade may, therefore, prevent thrombosis in affected individuals.

The data presented here support a role for the C5a–C5aR1 axis in the inflammatory mechanisms that underlie the development of ARDS in patients at early or late stages of infection with SARS-COV-2, which is consistent with other reports10,30. As described above, in addition to pneumonia and ARDS, there are data that suggest a role of C5a in other COVID-19-related symptoms, including heart, kidney or endothelial cell dysfunction14, providing support for the testing of the blockade of the C5a–C5aR1 axis in patients with COVID-19. We suggest that such a blockade may prevent the transition from a localized epithelial disease (non-severe COVID-19) to a diffuse endothelial disease (severe COVID-19) (Extended Data Fig. 4d).

Methods

Data reporting

No statistical methods were used to predetermine sample size. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment.

Study participants and clinical considerations

Over a period of one month (27 March 2020–24 April 2020), 82 participants were recruited from three hospitals (Timone and Nord University Hospitals, and Laveran Military Hospital, Marseille). Of the 82 participants, 28 were on mechanical ventilation for COVID-19-related-ARDS (P/F ratio < 300) (ARDS group), 34 patients required oxygen support at a rate of less than 5 l min−1 for COVID-19-related pneumonia (pneumonia group) and 10 patients had a paucisymptomatic form of COVID-19 that was compatible with outpatient care (paucisymptomatic group). COVID-19 was diagnosed on the basis of positive SARS-CoV-2 real-time reverse–transcription PCR (RT–PCR) on nasopharyngeal samples and/or typical CT-scan findings31. We also included 10 healthy volunteers (control group), with no fever or symptoms on the days before sampling and negative for SARS-CoV-2 RT–PCR. The characteristics of the patients are described in Supplementary Table 1. Biological samples were first collected within 3 days of diagnosis and the start of care (t0, <72 h, early time point). When possible, the next two time points for sample collection were located between days 5 and 10 (t1, intermediate time point) and after day 10 (t2, late time point). Flow cytometry analyses were performed on fresh blood samples (EDTA tubes) and BALFs, immediately after collection. Clinical progression was evaluated between the early and intermediate time points and between the intermediate and late time points. A favourable outcome was defined as weaning from mechanical ventilation (ARDS group) or oxygen support (pneumonia group). Death or multiple organ failure (ARDS group) and admission to the intensive care unit (pneumonia group) were considered unfavourable outcomes. In other cases, patients were considered to be stable.

Ethics approval statement

All of the patients (and/or initially their families) provided written informed consent before sampling and for the use of their clinical and biological data. The study protocol was approved on 27 March 2020 by the Committee for the Protection of Persons Ile-de-France III – France (2020-A00757-32). The pathological examination used in this study was performed secondary to a medical autopsy after COVID-19-related death, with agreement from the family and notification of the representative of the Commission on Data Processing and Freedom (MR003 research).

Animals

Female C57BL/6J mice were purchased at Janvier Labs and used between 8 and 12 weeks of age. HsC5AR1 knock-in mice were bred at Charles River Laboratories under specific-pathogen-free conditions. Female mice were used at 8–12 weeks of age and were allowed to acclimatize to the housing facility for at least one week. All animal experiments were performed in accordance with the rules of the Innate Pharma ethics committee and were approved by the Ministère de l’Enseignement Supérieur, de la Recherche et de l’Innovation – France (APAFIS#25418-2020051512242806 v2).

Reagent list

DPBS (1×) (14190-094, Gibco); RPMI medium 1640 (1×) (31870-025, Gibco); sodium pyruvate 100 mM (100×) (11360-039, Gibco); l-glutamine 200 mM (100×) (25030-024, Gibco); minimum essential medium non-essential amino acids solution (11140-035, Gibco); trypan blue stain (0.4%) (15250-061, Gibco); Ficoll–Paque PLUS (17-1440-03, GE Healthcare); fetal calf serum (FCS; F7524, Sigma); dimethyl sulfoxide (DMSO; D2650-100ML, Sigma); CD14 microbeads human (130-050-201, Miltenyi Biotec); EasySep direct human neutrophil isolation kit (19666, StemCell Technologies); bovine serum albumin (BSA; A9418-100G, Sigma); UltraPure 0.5 M EDTA, pH 8.0 (15575-038, Invitrogen); LPS EK ultrapure (tlrl-peklps, Invivogen); R848 (tlrl-r848, Invivogen); C5a (IPH 1D9 batch 1A, Innate Pharma); C5a (2037-C5-025, R&D Systems); avdoralimab (Innate Pharma), isotype control (Fc-silent hIgG1) (Innate Pharma); CD33–PECF594 clone WM53 (562492, BD Biosciences); CD19–PECy7 clone SJ25C1(557835, BD Biosciences); CD3–BUV496 clone UCHT1 (564809/612940, BD Biosciences); CD15–BV510 clone W6D3 (563141, BD Biosciences); CD45–BV711 clone HI30 (564357, BD Biosciences); CD16–BUV395 clone 3G8 (563785, BD Biosciences); CD14–BUV737 clone M5E2 (564444/612763, BD Biosciences); HLA-DR–AF700 clone L243 (307626, BioLegend); LIVE/DEAD Near-IR (L34976, ThermoFisher); mouse serum (015-000-120, Jackson ImmunoResearch); CD88 C5aR PE clone S5/1 (344304, BioLegend); anti-CD16–FITC (556616, BD Biosciences); anti-CD11b–PE-Cy5 (555389, BD Biosciences); U-PLEX kit (N05235A-1, MSD); OptEIA HsC5a ELISA (557965, BD Biosciences); mouse albumin ELISA kit (E99-134, Bethyl Laboratories); Ficoll (11778538, Invitrogen); dextran (31382, Sigma); calcein AM (C3100MP, Invitrogen); fibrinogen (F3879, Sigma); Transwell Fluoroblok 3 μm insert (351151, Corning); EDTA (15575-038, Invitrogen); sodium azide (71290-100g, Sigma); Optilyse C solution (A11895, Beckman Coulter); CytoFix (554655, BD Bioscience); avacopan (HY-17627, Clinisciences); anti-C5a (Innate Pharma); anti-mouse Ly6C–BV510 clone HK1.4 (128033, BioLegend); anti-mouse Ly6G–BV786 clone 1A8 (740953, BD Biosciences); anti-mouse CD45–BUV395 clone 30F11 (564279, BD Biosciences), anti-mouse CD11b–BUV737 clone M1/70 (564443/612800, BD Biosciences); anti-mouse C5aR1–APC clone 20/70 (130-106-124, Miltenyi Biotec); anti-human CD88 clone S5/1 (HM2094-100UG, Hycult Biotech); anti-human CD68 clone KP1 (M0814, Agilent); anti-human CD163 clone EDHu-1 (MCA1853, BioRad).

SARS-CoV-2 detection by PCR

SARS-CoV-2 RNA was detected by RT–PCR, as previously described32.

Seroneutralization assay

Experiments were performed in BSL3 facilities with a clinical isolate of SARS-CoV-2. Virus-neutralization tests were performed as previously described33. In brief, virus-neutralization tests were performed in a 96-well plate, with Vero E6 cells and a SARS-CoV-2 strain (Ref-SKU:026V-03883 isolated at Charité University, Berlin, Germany; EVA-GLOBAL H2020 project; grant agreement 871029). Twofold serial dilutions of serum samples (final serum dilutions of 1:20 to 1:160) were mixed with a median tissue culture infective dose of 100 of SARS-CoV2 and dispensed on the confluent cell monolayer. The plates were incubated for 4 days and examined for the presence (no neutralization) or absence (neutralization) of cytopathogenic effects under an inverted microscope.

Preparation of peripheral blood mononuclear cells and plasma

Whole blood collected in EDTA tubes was pooled and diluted 1:2 in PBS. Peripheral blood mononuclear cells (PBMCs) were isolated by centrifugation on a Ficoll gradient, and 107 PBMCs per vial were frozen in freezing medium (90% FCS and 10% DMSO). Plasma was collected from the upper phase of the Ficoll gradient, aliquoted and used for the quantification of soluble factors.

Assessment of soluble factors

Human IL-6, CXCL9, CCL2, CCL4, CXCL8, TNF and IL-1β levels were analysed with the U-PLEX kit supplied by MSD (U-PLEX 10-Assay, 96-Well SECTOR plate, N05235A-1), according to the manufacturer’s instructions. The U-PLEX plate was loaded into an MSD instrument to measure the intensity of emitted light, which is proportional to the amount of analyte present in the sample. Circulating C5a desArg levels were analysed using a BD OptEIA HsC5a ELISA. Mouse albumin in BALF was analysed by ELISA (Bethyl). HRP-conjugated secondary antibodies were detected by incubation with a peroxidase substrate solution (TMB) and the reaction was stopped by acidification. Plates were read at 450 nm.

Flow cytometry

Blood collected into EDTA tubes was washed in PBS before staining with LIVE/DEAD (Thermo Fisher) according to the manufacturer’s instructions. Cells were incubated with mouse serum to saturate the Fc receptors, and were then incubated with the appropriate antibody cocktail. Red blood cells were lysed in Optilyse C solution (Beckman Coulter), according to the manufacturer’s instructions. Cells were fixed in CytoFix solution (BD Bioscience, 554655), according to the manufacturer’s instructions. Data were acquired in an LSRFortessaX20 flow cytometer. The FCS3.0 files obtained were exported from BD FACSDiva software and imported into FlowJo v.10.5.2 (BD Biosciences). Automated compensation was calculated with FACSDiva software and single-antibody-stained compensation beads. This compensation matrix was analysed in detail in FlowJo, by investigating the N-by-N view feature and the pairwise expression of all proteins stained in this study. Fluorescence minus one experiments were run before this study, to facilitate optimization of the compensation matrix. We then adjusted the compensation matrix where necessary due to over- or under-compensation by the automatic algorithm.

Immune cell counts

Absolute counts per microlitre of blood were determined with BD TBNK Trucount Tubes. Absolute counts for a particular cell population (A) were obtained by dividing the number of positive cell events (X) by the number of bead events (Y), and then multiplying by the BD Trucount bead concentration (N/V, where N is the number of beads per test and V is the test volume). A = X/Y × N/V. The number of positive counts for neutrophils and monocytes was established with the CD45+SSChigh and CD45+SSCint gating strategies, respectively.

C5a inhibitors

Avdoralimab is a fully human mutated Fc-silent IgG1 monoclonal antibody against C5aR1 (US 2020/0017599A1). Anti-C5a monoclonal antibody is a chimeric mutated Fc silent IgG1 isotype cloned from the sequences of mouse anti-HsC5a INab308 (WO2015/140304A1), and has the same variable sequences as IFX-1. The C5aR1 antagonist (avacopan) was purchased from Clinisciences (HY-17627).

Neutrophil migration

Neutrophils were isolated from fresh blood by sedimentation in 6% dextran to separate plasma and leukocytes, followed by centrifugation on a Ficoll density gradient. The pellet, containing neutrophils, was recovered, and the red blood cells were lysed by incubation in 0.2% NaCl. Osmotic balance was restored by adding an equal volume of 1.6% NaCl. Isolated neutrophils were loaded with 10 μM calcein AM. Cell density was adjusted before the addition of avdoralimab or its isotype control at a final concentration of 10 μg ml−1. Neutrophils were dispensed into the top chamber of a fibrinogen- and BSA-coated Transwell Fluoroblok 3 μm insert. The lower chamber was filled with RPMI 1640 with or without 3 nM HsC5a (R&D Systems,) and the same antibody was added to the top chamber (avdoralimab, isotype control or PBS). After 30 min of incubation at +37 ± 1 °C under an atmosphere containing 5 ± 1% CO2, images of the bottom side of the inserts were acquired on a Biotek Cytation 5 plate-reading microscope, and analysed with Halo software (Indica Labs), using the CytoNuclear FL module to count the cells that had crossed the membrane.

Neutrophil activation

Various concentrations of avdoralimab were added to the blood samples in culture-treated 96-well U-bottom plates, and incubated for 20 min at 37 °C under an atmosphere containing 5% CO2. We then added 18 nM HsC5a (R&D Systems) to the samples. Plates were incubated for 20 min at 37 °C under an atmosphere of 5% CO2. Samples were then stained for flow cytometry analysis with anti-CD16–FITC and anti-CD11b–PE-Cy5 antibodies. Erythrocytes were lysed with Optilyse C solution (Beckman Coulter, A11895), according to the manufacturer’s protocol, and resuspended in CytoFix (BD Bioscience 554655) for fixation. Cells were then analysed on a FACS Canto II flow cytometer (BD Biosciences) with FACS Diva software.

Monocyte activation

Monocytes were purified with the CD14+ selection kit (Miltenyi). We used 30,000 monocytes to seed 96-well U-bottom plates. Cells were activated by overnight incubation with R848 (50 ng ml−1), LPS (0.5 ng ml−1) and HsC5a (1 μg mL−1; IPH). In some conditions, monocytes were incubated with avdoralimab (20 μg ml−1; IPH) or its isotype control (20 μg ml−1; IPH) for 30 min before stimulation. IL-6, TNF and CCL2 levels were quantified in the supernatant.

Mouse model of lung inflammation

Isoflurane-anaesthetized HsC5AR1 knock-in mice received 3.1 μg of recombinant HsC5a (R&D) in 40 μl PBS, by intranasal instillation. After 18 h, mice were killed with a lethal dose of ketamine/xylazine cocktail (ketamine 300 mg kg−1; xylazine 30 mg kg−1). The lungs were flushed with 2 ml 2 mM EDTA in PBS to obtain BALF. After centrifugation (300g, 5 min, 4 °C), BALF cells were counted and stained for flow cytometry analysis with anti-CD45, anti-Ly6C, anti-Ly6G and anti-CD11b antibodies. For histology analysis, 18 h after intranasal instillation of HsC5a, lungs were fixed in formalin, dissected, embedded in paraffin and sectioned to 5 μm. Sections were dewaxed and stained with haematoxylin and eosin (H&E). Slides were scanned using a Nanozoomer S60 (Hamamatsu) and examined for evidence of lung damage.

Multiplex immunohistochemistry staining protocol, image acquisition and data analysis

Multiplexed immunohistochemistry (IHC) was performed with a Leica Bond Rx on 5-μm-thick formalin-fixed paraffin-embedded lung tissue sections from individuals with and without COVID-19. Consecutive staining was performed by heat-induced antigen retrieval followed by incubation with primary antibody (anti-C5aR1 clone S5/1 at 1 μg ml−1). The signal was amplified and detected with Opal polymer horseradish peroxidase and Opal 520 (Akoya Biosciences). The sections were then subjected to heat-induced antibody stripping and incubated with the next antibody (anti-CD163 clone EDHu-1 at 1 μg ml−1, detected with Opal 620, and, finally, anti-CD68 clone KP1 at 0.1 μg ml−1, detected with Opal 690) and spectral DAPI. All Opal reagents were used at a dilution of 1:150. Slides were mounted in ProLong Diamond anti-fade mounting medium (Thermo Fisher) and scanned with a Vectra Polaris (Akoya Biosciences). H&E-stained slides were scanned with a Nanozoomer (Hamamatsu). After spectral deconvolution and whole-slide reconstruction of the multiplexed IHC stained sections, digital pathology methods were used to determine the density of positive cells. All analyses were performed with Halo (Indica Labs) and R.

Transcriptomic analysis

Transcriptomic analyses were performed on previously reported data13,34. The RNA-sequencing (RNA-seq) data for two BALF samples from patients (each in duplicate), three PBMC samples from healthy control individuals and three PBMC samples from patients were downloaded from the National Genomics Data Center (https://bigd.big.ac.cn/; accession number PRJCA002326). The RNA-seq data for three BALF samples from healthy control individuals were downloaded from the SRA database (accession numbers, SRR10571724, SRR10571730 and SRR10571732).

RNA-seq pipeline

The reads were mapped to the human genome (hg38) release 96 from Ensembl with STAR35. PCR replicates mapping to the human genome were removed with the Picard MarkDuplicates program (Broad Institute 2019, http://broadinstitute.github.io/picard/). Gene expression was calculated with featureCounts in the SubReads package (v.1.6.4)36. Transcripts per million values were calculated from the raw counts and log2-transformed. The depth of sequencing of the patient BALF samples was low (<1 million).

Batch effect correction

We corrected for the batch effect between the datasets for BALF samples from healthy control individuals and those from patients and PBMC samples with Combat37, using the model: ~Batch + Status (Patient or Healthy) + Sample Type (PMBC or BALF). An analysis of differential expression between PBMC samples from healthy control individuals and patients was performed on raw counts with DESeq238. Significance was defined as an adjusted P < 0.05. The metagene IRAK3 MS4A6A, CD33, CD300C, VCAN, CD1D, CCR1, OAS1, CD163, CD14, FCN1, AIF1 and PLA2G7 was used to calculate the macrophage and monocyte transcriptomic signature. The significance of the difference between healthy control individuals and patients was evaluated using a Wilcoxon rank-sum test. The single-cell RNA-seq data for 12 BALF samples from three healthy control individuals, three patients with mild COVID-19 and six patients with severe COVID-19 were downloaded from the Gene Expression Omnibus (GEO) database (accession number GSE145926). Quality controls were applied to each cell, for all samples, with the same criteria as for the initial analysis (gene number between 200 and 6,000, UMI count >1,000 and mitochondrial gene percentage <0.1) with the Seurat package (v.3.1.0). After filtering, 63,740 cells were validated. As previously described13 the filtered gene–barcode matrix was first normalized with ‘LogNormalize’ methods in Seurat v.3, with default parameters. The top 2,000 variable genes were then identified by the ‘vst’ method with the Seurat FindVariableFeatures function. The variables ‘nCount_RNA’ and ‘percent.mito’ were regressed out in the scaling step and a principal component analysis was performed on the top 2,000 variable genes. For the reanalysis presented here, the batch effects across different donors were removed by Harmony39 and UMAP was performed on the top 50 dimensions for visualizing the cells. Graph-based clustering was performed on the Harmony-corrected data, with a resolution of 1.2, and defined major clusters composed of epithelial cells, B and plasma cells, T and NK cells, dendritic cells, monocytes, macrophages and neutrophils. The 50,610 myeloid cells were reintegrated and reclustered.

Data analysis and statistics

All statistical analyses were performed with R (v.3.6.1). The ggpubr (v.0.2.5) and lmerTest (v.3.1.2) packages were used for statistical tests. The gtsummary package (v.1.3.0) was used for the table containing clinical information (Supplementary Table 1). The sva package (v.3.32.1) was used to correct the batch effect of RNA-seq. Packages ggplot2 (v.3.2.1) and pheatmap (v.1.0.12) were used for the graphical representations of RNA-seq analyses. The Seurat package (v.3.1.0) was used for all analyses of single-cell RNA-seq. For the comparison of groups at time point t0, P values were obtained for two-tailed Wilcoxon rank-sum tests. For longitudinal analysis in the pneumonia group, the P values for comparisons of t1 and t0 were obtained using two-tailed Wilcoxon signed-rank tests. No statistical tests were performed for t2 in this group. For the ARDS group, a mixed model was computed, with time point as a fixed effect (categorical variable) and patient as a random effect. Confidence intervals and P values were based on the t-distribution, with degrees of freedom according to the Kenward–Roger method, and the normality of residuals was verified. Plots were drawn with GraphPad Prism v.8.1.1. Box plots show the median (centre line) and 25th to 75th percentiles (box) and the whiskers denote the maximum and minimum values.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.