This non-interventional study was approved by the institutional review board of the ethical committee for research (CER) of the University of Paris Saclay (CER-Paris-Saclay-2020-050) and conformed to the principles outlined in the Declaration of Helsinki. Accordingly, all participants were informed in writing about the study and were given the option to not participate. We studied prospective samples from 48 COVID-19 and 21 non-COVID patients admitted at the Cochin (Paris, France), Ambroise Paré (Boulogne-Billancourt, France), and Raymond Poincaré (Garches, France) Hospitals between March and June 2020. All patients had a COVID-19 diagnosis confirmed by SARS-CoV-2 RNA RT-qPCR in nasopharyngeal swabs at the hospital. Clinical data submitted by the participating centers were anonymized and encrypted.

BAL samples were collected as described (24) and processed as indicated in our recent study (25). Briefly, a volume of 50 ml of isotonic saline was injected with a recovery of 6 to 18 ml, and the collected fluid was processed within 3 h. BAL was passed through a 70 μm strainer and collected in a 50 ml tube. After the centrifugation of 500 g for 10 min, fluid was collected, aliquoted at 1 ml, and stored until use at −80°C in a biosafety level 3 (BSL3) laboratory. The BAL cells were resuspended in 10% dimethyl sulfoxide (DMSO) in fetal calf serum (FCS) and stored at −80°C until use in flow cytometry analyses. BAL fluids without cells were aliquoted in 60 μl fractions, inactivated at 56°C for 30 min in the BSL3 facility, and stored at −80°C for subsequent use.

The concentration of total IgG and IgA in BAL secretions was measured by sandwich enzyme-linked immunosorbent assay (ELISA) as we described (26) using polyclonal goat antihuman IgG or IgA (Biosystems, Burlingame, CA, USA) for coating and polyclonal goat anti-human IgG (Nordic, Tilburg, The Netherlands) or polyclonal goat anti-human IgA (Nordic) for detection; standards were purified human serum IgG (Sigma, St. Louis, MI, USA; I2511) or purified human colostral IgA (Sigma, I2636). BAL IgG and IgA specific to SARS-CoV-2 S1, S2, RBD, and nucleocapsid protein (NP) were determined by an ELISA as described below. IgG and IgA anti-S1 detection was performed using the anti-SARS-CoV-2 ELISA Kit (EI 2606-9620 G and EI 2606-9620 A; Euroimmun, Mountain Lakes, NJ, USA) following the manufacturer’s instructions. IgG and IgA to NP were measured using the NOVATEC ELISA KIT according to the manufacturer’s instructions. For IgG and IgA anti-S2 and anti-RBD quantification, 96-well, flat-bottomed plates (Nunc-Immun Microwell, Thermo Fisher Scientific, Odense C, Denmark) were coated overnight at 4°C with 1 ng/well, or 100 ng/well, of SARS-CoV-2 spike S2 protein and SARS-CoV-2 spike RBD Wuhan protein (LifeTein, Somerset, NJ, USA) and recombinant human SARS-CoV-2 spike RBD variants Alpha (B.1.1.7), Beta (B.1.351), P.1 (Gamma), and Delta (B.1.617) (Diaclone, Besançon, France). After 24 h, a blocking solution (200 µl per well of bovine serum albumin (BSA) 2% diluted in phosphate-buffered saline (PBS) with 0.1% Tween 20 (PBST)) was added, and plates were incubated for 2 h at 37°C followed by five washes with PBST. BAL samples were diluted in PBST (1/50 and 1/100 dilution), and 100 µl of diluted samples were added to the plates for 2 h at 37°C. After several washes, goat anti-human IgG labeled with horseradish peroxidase (HRP) or goat anti-human IgA-HRP (Jackson Immunoresearch, Ely, UK) was added to each well for 1 h at room temperature. The reaction was developed with tetramethylbenzidine (TMB)-ELISA solution (Eurobio Scientific, Essonnes, Ile-de-France, France) for 15 min prior to stopping with H₃PO₄ (1 M). The absorption at 450 nM (OD450) was read on a Spectramax spectrophotometer (Molecular Devices, Wokingham, UK). Samples from COVID-19-negative subjects in the intensive care unit (ICU) were tested as negative controls. As for another mucosal sampling, the volume of mucosal BAL sampled in each individual varies from individual to individual for various reasons such as patient morphology and COVID-19 pathology specificities (27). Thus, to compare antigen-specific antibody isotype, IgG, and IgA, in the BAL, for each isotype, we normalized the OD450 values measuring specific binding to each antigen to the total antibody isotype concentration as we described earlier (26, 28, 29). Results are shown in arbitrary units (AU) calculated as follows: (OD450 measured in antigen-specific IgA or IgG ELISA/total IgA or IgG concentration (μg/ml)).

As internal standard control, the WHO International Standard (WHO IS, National Institute for Biological Standards and Control, NIBSC, UK, cod. 20/136) and the WHO Reference Panel (WHO RP, NIBSC, cod. 20/268) for anti-SARS-CoV-2 antibody were tested at 1:100 dilution in the ELISA for S1 and S2 and the Novatec kit for NP to check the concordance with our results, as previously described (30).

For the detection of IgA-SARS-CoV-2 immune complexes, ELISA was performed as above, except for the coating conditions. Plates were coated with the polyclonal rabbit anti-SARS-CoV-2 spike (Genetex GTX135356) at 50 ng/well. Specificity was established using BAL from three different non-COVID individuals. These values were considered as background and subtracted from the presented data.

HEK 293T/17 cells were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA) and cultured in Dulbecco’s modified Eagle medium (Lonza, Basel, Switzerland) supplemented with 10% fetal bovine serum, 100 units/ml of penicillin, and 100 µg/ml of streptomycin (Euroclone, Pero, Italy). HEK 293T/17-ACE2/TMPRSS2 cells were generated by co-transfection of pCAGGS encoded human ACE2 and human TMPRSS2 using FuGENE^® HD Transfection Reagent (Promega, Madison, WI, USA) according to the manufacturer’s instruction. After 24 h, cells were detached and used for downstream assays.

A lentivirus-based SARS-CoV-2 pseudotyped virus (PSV) was generated, as previously described (31). Briefly, HEK 293T/17 cells were co-transfected with a SARS-CoV-2 spike encoding plasmid, a p8.91 HIV Gag-pol packaging construct, and a pCSFLW plasmid encoding a firefly luciferase reporter using Fugene^® HD transfection reagent, according to the manufacturer’s instruction. To generate PSVs of the SARS-CoV-2 variants of concern, spike plasmids encoding the mutations for Alpha (B.1.1.7), Beta (B.1.351), and Gamma (P.1) were commercially synthesized and used instead. Cells were incubated for 48 h prior to collecting and filtrating supernatant containing PSVs, using a 0.45 µm cellulose acetate filter. PSVs were then aliquoted and stored at −80°C.

Viral titers were determined by transducing 10⁴ HEK 293T/17-ACE2/TMPRSS2 cells with twofold serial dilutions of PSVs to each well of a 96-well titration plate, as previously described (31) After 48 h post-incubation at 37°C 5% CO₂, firefly luciferase expression was quantified by the Bright-Glo™ assay luciferase system (Promega) and the VICTOR X Light Luminescence Plate Reader (PerkinElmer, Waltham, MA, USA). Each relative luminescence unit (RLU) value obtained at different PSV dilution points was converted into RLU/ml, and the arithmetic mean of these concentrations was considered as the PSV production titer (expressed as RLU/ml).

Neutralization activity of previously heat-inactivated (56°C for 30 min) plasma from COVID-19 patients was measured using a single round PSV infection of HEK 293T/17-ACE2/TMPRSS2-transfected cells. Plasma collected prior to the emergence of SARS-CoV-2 was used as negative controls. Neutralization assays were performed by incubating 10⁶ RLU of SARS-CoV-2 Wuhan, Alpha, Beta, or Gamma pseudotyped viruses with endpoint twofold serial dilutions of BAL samples (starting from 1:5) at 37°C 5% CO₂ for 1 h before addition of 10⁴ HEK 293T/17-ACE2/TMPRSS2 cells per well. All samples were measured twice in duplicate. After 48 h at 37°C, the cells were lysed, and luciferase activity was measured as previously reported (32). Neutralization titers were converted into half-maximal inhibitory concentration (IC50) using Prism software, as previously described (32).

BAL cells were thawed from frozen aliquots and fixed with 4% paraformaldehyde (PFA) for 30 min. After successive washes with PBS-BSA, cells were incubated for 30 min, at room temperature, with the following antibodies coupled to fluorophores diluted in permeabilization buffer (PBS 0.1% Saponin 2% FCS): CD3 Pacific Blue (BD ref: 558117, 1:20 v/v), CD19 APC-H7 (BD ref: 560177, 1:40 v/v), CD27 PE (BD ref: 566944, 1:20 v/v), CD21 PE-Cy7 (BD ref: 561374, 1:20 v/v), CD38 BV711 (BD ref: 563965, 1:40 v/v), and CD138 APC (BD ref: 347216, 1:20 v/v). Then, cells were labeled with either human fluorescein isothiocyanate (FITC)-conjugated IgA (Jackson ref: 309-095-011, 1:100 v/v) or human FITC-conjugated IgG (Jackson ref: 709-096-149, 1:50 v/v). Cells were then analyzed by flow cytometry (Guava easyCyte 12HT base system, Millipore, Billerica, MA, USA) using a gating strategy shown in Supplementary Figure 1 to evaluate various B-cell subset frequencies.

After frozen aliquots of BAL fluid were thawed, samples were directly processed for multiplex detection of the following cytokines, according to the distributor’s instructions: MIP-1α, G-CSF, M-CSF, IL-1α, IL-1β, IL-6, IL-8, TNF-α S100A8, S100B, and CXCL4 (R&D Luminex, R&D, Austin, TX, USA). Samples were analyzed in a Bio-Plex 200 system (Bio-Rad, Hercules, CA, USA) following the manufacturer’s instructions.

Analysis of data was performed using Microsoft^® Excel 2011 and GraphPad Prism^® version 9 (GraphPad software). Summary statistics, mean with standard error of the mean (SEM) and percentages, are shown. Statistical tests were performed considering non-normal distributions (non-parametric tests, unpaired Mann–Whitney test, or paired Wilcoxon test, as indicated). Correlations were assessed by two-tailed Spearman’s correlation coefficients. All tests were two-sided with p-values of 0.05 or less denoting statistical significance. The results are presented as box or violin plots with individual values represented as dots.

Univariate and multivariate logistic regression models were used to assess predictors of hospital mortality, with odds ratio (OR) and 95% confidence intervals [95% CIs] used as the measure of association with the outcome.

Bayesian logistic regression was applied to BAL cytokine measurements using the RStanArm package in the R language.

A total of 69 BAL samples were collected from SARS-CoV-2-infected (severe COVID-19, n = 34) and non-infected (non-COVID-19, n = 21) individuals in the intensive care unit between March and June 2020 whose clinical data are summarized in Table 1 . Samples were obtained at the enrolment, and 10 individuals provided longitudinal samples. Furthermore, BAL samples were stratified in two groups according to viral gene detection, referred to as SARS-CoV-2+ BAL (mean viral load in BAL: 3.27 × 10⁶ ORF1 copies/ml ± 2.11 × 10⁶), and RT-qPCR negative, referred to as SARS-CoV-2 neg BAL (no detectable viral load in BAL). The experimental design and the cross-sectional sampling during the disease course of the patients are shown in Figures 1A, B respectively.

When the presence of SARS-CoV-2 in BAL was correlated to patient clinical data, the mean time from the onset of symptoms to sampling was shorter for virus-containing BAL (22 ± 4.4 mean days) compared to virus-free BAL (38 ± 3 mean days, p = 0.003, unpaired Mann–Whitney test, Figure 1C ). Although they partially overlapped due to the heterogeneous dynamic of viral persistence (33, 34), the two groups appeared thus to be statistically significantly different when stratified by time from onset of symptoms. We therefore associated the presence of the virus in BAL with the phase of the disease (35). SARS-CoV-2+ BAL corresponded to an early phase of virus replication, whereas SARS-CoV-2 neg BAL corresponded to a late phase of the infection when after the virus has been cleared from the lung. Accordingly, SARS-CoV-2+ BAL and SARS-CoV-2 neg BAL are referred to as BAL from the early and late phases of COVID-19 disease, respectively.

To study the dynamics of the lung humoral anti-SARS-CoV-2 immune response in severe COVID-19, we first quantified the presence of total IgG and IgA in BAL samples by ELISA ( Figure 1D ). Although monomeric IgA can be present in BAL, the prevalent class of antibodies in BAL is mucosal secretory IgA (13). We thus refer in the following to IgA as mucosal IgA. Total IgG and IgA were detected in >90% of early COVID-19 samples, whereas the proportion decreased to >65% in late COVID-19 samples when the virus was undetectable. The concentration of total IgG was statistically higher than that of IgA in all BAL samples from COVID-19 patients (early COVID-19 sample mean 82 ± 22 μg/ml for IgG vs 27 ± 8.1 μg/ml for IgA, p < 0.0001, late COVID-19 sample mean: 27 ± 8.1 μg/ml for IgG vs 9 ± 3.8 μg/ml for IgA, p = 0.01).

S1, which includes the ACE2 RBD, and S2 subunits, accessible at the virus surface, are likely targets for COVID-19 protective antibodies. The internal NP, the most abundant in infected cells, offers a sensitive marker of infection (36). IgG and IgA targeting these antigens were quantified in all BAL samples from severe COVID-19 patients.

We found that 27%, 33%, 47%, and 50% of SARS-CoV-2+ BAL had IgG to S1, S2, RBD, and NP, respectively, and 33%, 40%, 53%, and 57% had IgA specific to S1, S2, RBD, and NP, respectively ( Figure 2A ). The S1-specific IgA and RBD-specific IgA were slightly higher than IgG (no statistical differences due to low sample numbers). Conversely, IgA specific for S2 and NP was statistically significantly higher than IgG (32 ± 7.6 for S2-IgA vs 5.5 ± 2.3 for S2-IgG, p = 0.01; 62 ± 3.2 for NP-IgA vs 3.7 ± 1.4 for NP-IgG, p = 0.03).

Similarly, late in COVID-19 infection when virus replication was resolved, 38%, 59%, 50%, and 20% of SARS-CoV-2 neg BAL had IgG to S1, S2, RBD, and NP, respectively, and 65%, 65%, 55%, and 26% had IgA to S1 and S2, RBD, and NP, respectively ( Figure 2B ).

The same IgA to IgG ratio was observed in late SARS-CoV-2 neg BAL samples as shown in Figure 2B , with S1-specific IgA statistically significantly higher than IgG (112 ± 30 vs 19 ± 6.4 respectively, p = 0.02). S2-specific IgA was higher than IgG (69.5 ± 19 vs 26 ± 10 respectively, p = 0.02), whereas anti-NP of both isotypes was present in an equal proportion of patients (55% vs 50%, respectively). Finally, following the same pattern, N-specific IgA was higher compared to IgG (mean N-specific IgA and IgG: 44 ± 13 vs 15 ± 6.2 respectively, p = 0.007).

We next correlated the anti-S1, RBD, S2, NP IgG and IgA responses with the onset of symptoms to sampling to evaluate the antibody kinetic. As shown in Figure 2C , SARS-CoV-2-specific IgG was detected early during the infection in SARS-CoV-2+ BAL. S1 and NP-IgG persisted during the disease and after viral clearance from the lung, whereas S2-IgG slowly decreased in SARS-CoV-2 neg BAL. RBD-specific IgG declined once the virus was eliminated from the BAL ( Figure 2C , gray dots). Kinetics of the specific IgA responses mirrored that of the IgG: specific antibodies appeared at the initial phase of infection (SARS-CoV-2+ BAL, Figure 2D , red dots) and persist over time (SARS-CoV-2− BAL, Figure 2D , gray dots).

The specific level of BAL IgA anti-S1, anti-RBD, anti-S2, and anti-NP correlated with that of IgG in both the early ( Figure 2E ) and late phases ( Figure 2F ) of COVID-19. In SARS-CoV-2+ BAL, corelation between IgG and IgA was higher for S1 (0.8, p = 0.001) followed by RBD-specific response (r = 0.5, p = 0.001) ( Figure 2E ). The strongest correlation in SARS-CoV-2 neg BAL patients was observed for anti-S2 and anti-NP IgA and IgG (r = 0.6, p = 0.0002) followed by S1 (r = 0.4, p = 0.005, Figure 2F ). These results show that IgG and IgA simultaneously evolve during severe COVID-19 development, independently of viral replication.

In summary, severe COVID-19 patients are capable of mounting a virus-specific mucosal immune response, which persists after virus elimination with higher levels of IgA than IgG.

Unexpectedly, although in agreement with other studies on serum samples, a small but significant fraction (12%) of the BAL tested in our study had no detectable IgA against all antigens tested. In line with our findings, a recent study demonstrated that a high proportion of patients had neither detectable viral-specific IgG nor IgA in their nasopharyngeal compartments (37), although the reasons underlying these findings were not studied.

We hypothesized that the presence of IgG or IgA immune complexes (ICs) could prevent the detection of anti-SARS-CoV-2 antibodies in these samples, in line with the recent report that ICs are potential determinants of the cytokine storm in severe COVID-19 (38). We thus established an ELISA to measure IgG and IgA complexed with spike antigens ( Figure 3 ).

IgG-spike ICs were detected in 32 of the 48 samples analyzed (66%, Figure 3A ). Interestingly, five out of 11 individuals negative for specific IgG against S1, RBD, S2, and NP (CA4, CC31, CA32, CC44, CA53, CA59, CA98, and CA30) had detectable levels of IgG-spike ICs (CC44, CA53, CA98, CA30, and CA58; Figure 3A , red asterisks), indicating that spike-specific IgG is present in these BAL but remains associated with the virus or free spike. Regarding IgA, spike ICs were present in 25 of the 48 samples analyzed (52%, Figure 3B ). Remarkably, four out of six individuals negative for specific IgA against S1, RBD, S2, and NP (CA4, CC51, CC89, CA30, CA58, and CA77) had detectable levels of IgA-spike ICs (CA4, CC51, CC89, and CA30; Figure 3A , red asterisks), mirroring what we measured in the case of IgG. IgG-spike ICs were detected in the same proportions in either the early or late phase of the infection ( Figure 3C ). On the contrary, IgA-spike ICs were predominant in individuals with no virus in BAL, in the late phase of infection (mean SARS-CoV-2+ BAL vs SARS-CoV-2 neg BAL 0.12 ± 0.009 vs 0.15 ± 0.007, p = 0.03, Figure 3D ).

The presence of both IgG and IgA ICs was not associated with survival ( Figures 3E, F respectively). However, we revealed that three out of the four individuals with detectable levels of IgA-spike ICs but undetectable anti-spike IgA underwent a fatal issue ( Figure 3F , red dots).

Finally, anti-S2 IgG statistically correlated with IgG ICs (p = 0.008, r = 0.4, Spearman’s correlation, Figure 3G ) and anti-S1 IgA with IgA ICs (p = 0.008, r = 0.4, Spearman’s correlation, Figure 3H ).

Altogether, these results demonstrate that BAL from SARS-CoV-2-infected individuals contain IgA-spike ICs, which were more predominant in the late phase of the infection. Furthermore, IgA ICs might impair direct detection of spike-specific IgA by direct ELISA, and more importantly, IgA ICs in BAL might be adverse for patient disease development, most likely by stimulating myeloid cells via Fc-alpha receptors, as shown recently for IgG via Fc-gamma receptors (39, 40).

The emergence of SARS-CoV-2 variants roused the question of whether the humoral response developed against the ancestral virus could offer cross-protection against the genetic variants. The spike protein is the main viral protein subjected to mutations, especially in the RBD, the principal spike subunit targeted by neutralizing antibodies. The N501Y mutation is the main mutation detected in the Alpha (B.1.1.7) variant that appeared in the United Kingdom (7, 41). The B.1.1.7 variant emerged independently in South Africa (42), whereas the Gamma variant (P.1) appeared in Brazil due to travelers from Japan and the B.1.617.2 Delta variant in India (43). A significant cross-protection of vaccinated individuals is detected against these new variants (44–46). The cross-reactivity of mucosal antibodies elicited from patients during natural SARS-CoV-2 infection remains unknown.

Therefore, we evaluated whether mucosal anti-RBD IgA and IgG elicited toward the Wuhan virus in the BAL we collected would cross-react with the RBD from the Alpha (B.1.1.7), Beta (B.1.351), Gamma (P.1), and Delta (B.1.617) variants. When tested in ELISA, >50% of the BAL samples had IgG cross-reacting with the Alpha, Beta, and Gamma RBD variants (52% to Alpha, 61% to Beta, and 50% to Gamma; Figure 4A ), whereas 43% cross-reacted with the Delta RBD variant, all compared with 52% of the Wuhan ancestral RBD ( Figure 4A ). Moreover, quantitatively, the level of IgG specific to the Delta RBD was lower than that of the Alpha (mean, 10 ± 3.1 vs 37 ± 14 respectively, p = 0.005) and the Beta ones (mean, 10 ± 3.1 vs 22 ± 5.9 respectively, p = 0.0005). In contrast, IgA targeting the Alpha, Beta, and Gamma RBD variants was only detected in 25% to 47% of the study population ( Figure 4B ). However, 50% of the samples had IgA specific to Delta RBD ( Figure 4B ), compared with the 43% IgG response. Of note, 18% and 11% of the individuals failed to raise IgG and IgA, respectively, against any of the variants or to the RBD Wuhan variant. In contrast, 15% and 5% of the individuals had IgG and IgA against all variants (including the ancestral Wuhan one), respectively. Finally, only 7% and 9% of the individuals developed IgG and IgA, respectively, to the sole Wuhan RBD ( Figure 4C ).

The magnitude of the IgA response toward RBD was higher than that of the IgG for the Beta, Gamma, and Delta variants (mean for anti-Beta RBD response: 38 ± 7 vs 21 ± 6, respectively, p = 0.02; mean for anti-Gamma RBD response: 29 ± 5 vs 17 ± 6 respectively, p = 0.0; mean for anti-Delta RBD response: 8 ± 3 vs 28 ± 10 respectively, p = 0.01, Figures 4F–H ), as observed for the Wuhan strain ( Figure 4D ). In contrast, no differences were observed for the Alpha variant ( Figure 4E ). Finally, when analyzed at the individual sample level, only the magnitude of IgA to RBD Alpha (p = 0.01, r = 0.34, Spearman’s correlation, Figure 4I ) and Gamma (p = 0.01, r = 0.3, Spearman’s correlation, Figure 4J ) correlated.

Altogether, our results showed that i) in the majority of BAL, IgG and IgA had cross-variant neutralization capacity; ii) Beta, Gamma, and Delta RBD-specific IgA are higher than IgG in line with what was observed for Wuhan RBD-specific antibodies; iii) 11% to 18% of the individuals developed neither IgG nor IgA to any variant RBD studied; iv) conversely, 5% to 15% of the individuals developed IgG and IgA to all RBDs studied.

To gain insight into the functions of mucosal SARS-CoV-2-specific IgA and IgG, we evaluated BAL neutralization activities. First, we compared the IC50 neutralization titers of SARS-CoV-2+, SARS-CoV-2 neg, and non-COVID-19 BAL. We found that 38% and 51% of virus containing and virus lacking BAL neutralized SARS-CoV-2 infection, respectively, whereas BAL from non-COVID-19 patients lacked neutralizing activity ( Figure 5A ). IC50 neutralization titers were statistically significantly higher in SARS-CoV-2+ compared with SARS-CoV-2 neg BAL (mean 324 ± 123 vs 89 ± 26 respectively, p = 0.01). This indicated that neutralizing antibodies developed early after infection when the virus replicates and decreased later after the virus has been cleared from the lung. Accordingly, the IC50 neutralization titers, plotted as a function of onset of symptoms to sampling time, first sharply rose during the first 3 weeks of the disease before slowly declining ( Figure 5B ). For ethical reasons, we could not collect enough samples from severe COVID-19 patients to purify each antibody isotype from these mucosal fluids. To attribute the neutralization observed to one isotype, we had to rely on correlations. Early in infection, IC50 neutralization titers of SARS-CoV-2+ BAL positively correlated with S1-specific IgG and IgA ( Figures 5C, D respectively).

The presence of neutralizing activity has been associated with a worse outcome in many studies (47–49). To address this issue, hospitalization duration for each individual was calculated ( Supplementary Figure 2A ) and correlated with the corresponding BAL neutralization titers. A positive correlation was only observed between IC50 and patient hospitalization duration for SARS-CoV-2 neg BAL (p = 0.04, r = 0.4, Figure 5E ), suggesting that long-lasting mucosal neutralizing antibodies may be disadvantageous for patient recovery. Accordingly, the hospitalization duration was shorter for individuals with SARS-CoV-2+ compared to SARS-CoV-2 neg BAL (mean duration: 35 ± 5.8 vs 49 ± 5.4 days). Indeed, the former showed sharp neutralizing antibodies rise cross-sectionally, whereas the latter had a stabilized neutralizing response.

Altogether, the early mucosal neutralizing response might exert a protective function; conversely, in the later phases associated with virus clearance from the pulmonary mucosa, additional non-neutralizing roles of neutralizing antibodies might be responsible for adverse effects (50). More analyses in larger cohorts of patients are required to confirm this conclusion.

Neutralization activities against RBD from the Wuhan (WT), Alpha (B.1.1.7), Beta (B.1.351), and Gamma (P.1) variants were evaluated in the BAL from 14 SARS-CoV-2+ and 10 SARS-CoV-2 neg patients; neutralization activities against all the viruses considered were detected in 66%, 46%, 25%, and 33% of the individuals, respectively ( Figure 5F ).

As expected, neutralization titers against the Wuhan virus were higher compared to those against the Alpha (mean 215 ± 55 and 74 ± 27 respectively, p = 0.004), Beta (mean 215 ± 55 and 37 ± 3.2 respectively, p = 0.01), and Gamma (mean 215 ± 55 and 75 ± 24 respectively, p = 0.02) variants ( Figure 5F ).

Altogether, these data supported our previous findings showing that the majority of BAL contained mucosal antibodies to viral variants indicative of a potent cross-neutralization capacity.

To investigate whether mucosal SARS-CoV-2-specific antibodies could play a role in the patient’s survival, we stratified BAL samples according to the outcome of patients, referred to as survivors and non-survivors. In survivors, the S2- but not S1- or RBD-specific IgA response was statistically higher compared with corresponding IgG (mean 41 ± 7.9 and 26 ± 10.5, respectively, p = 0.02, Figure 6A ). Conversely, in non-survivors ( Figure 6B ), S1- and RBD-specific but not S2-specific IgA predominated over IgG (mean S1-IgA: 71 ± 26 vs S1-IgG: 13 ± 7.8, p = 0.007; mean S2-IgA: 83 ± 30 vs S2-IgG: 12 ± 7.7, p = 0.009). The dominant NP-specific IgA responses occurred independently of survival (in survivors: mean anti-NP-IgA: 56 ± 21 vs anti-NP IgG: 13 ± 3.2, p = 0.007, Figure 6A ; in non-survivors, mean anti-NP-IgA: 36 ± 12, vs anti-NP IgG 5.9 ± 2.7 p = 0.01, Figure 6B ).

However, analyses of the IgA or IgG S-specific responses varied with the patient outcome with S2-specific IgG being statistically significantly higher in survivors compared with non-survivors (mean S2-IgG equal to 26 ± 4.0 vs 12 ± 7.7, in survivors vs non-survivors, p = 0.03, Supplementary Figure 3A ). Conversely, the S2-specific IgA response was statistically significantly higher in non-survivors compared with survivors [mean S2-IgA survivors, 41 ± 10 vs IgG anti-S2 non-survivors: 83 ± 30, p = 0.01 ( Supplementary Figure 3B )].

In a cross-sectional analysis, we then investigated whether the kinetics of specific antibodies might vary with patient survival. We found that the levels of IgG and IgA toward S1, RBD, S2, and NP were similar, irrespective of survival (mean of 33 ± 3.3 vs 33 ± 4.5 days for survivors and non-survivors, respectively, Supplementary Figure 2B ). In non-survivors only, S1-, RBD-, S2-, and NP-specific IgG and S1-specific IgA increased over time ( Figures 6C, D ), Moreover, surviving individuals had higher but transient S2 specific IgG response in the lungs ( Figure 6C ). These results highlighted that antibody persistence, but not their amount, might play an adverse role in COVID-19 pathogenesis.

We next evaluated the involvement of neutralizing antibodies in patient outcomes. For individuals developing neutralizing activities in BAL, the neutralization titers in survivors and non-survivors were not statistically significantly different (mean equal to 184 ± 70 vs 94 ± 35, respectively, Figure 7A ); and no differences in the kinetics of the neutralizing response between survivors and non-survivors were detected ( Figure 7B ). However, in non-survivors ( Figure 7C ), the IC50 neutralizing titers started to rise in direct correlation with time (from the onset of the disease to sampling) in SARS-CoV-2+ BAL (r = 1, p = 0.3), whereas it decreased in SARS-CoV-2− BAL (r = −1, p = 0.08). Of note, in both cases, the correlation was not statistically significant due to the small sample size. These data suggest that, in severe COVID-19 patients, a strong neutralizing activity mounted at the early phase of the disease when the virus replicates, followed by its decrease when the virus disappears is not sufficient. As previously speculated, mucosal antibodies might have additional activities, likely contributing to the fatal outcome of these patients (51, 52), although more samples need to be analyzed to confirm this hypothesis. In this context, some BAL samples were tested for antibody-dependent enhancement (ADE) activity using the protocol established by Wu et al. (53), but no ADE activity was detected (data not shown).

Mucosal antibodies are raised locally after mucosa-specific homing of B cells (54). Therefore, we characterized B-cell populations in BAL from our cohort. Five B-cell populations were analyzed by flow cytometry: i) CD27- CD21+ naïve B, ii) CD27+ CD21- activated memory B cells, iii) CD38+ CD138+ plasma B cells, iv) CD27+ CD21+ resting memory B cells, and vi) CD27- CD21+ tissue memory B cells. In addition, B cells were stratified according to the presence of IgG and IgA (as shown in the gating strategy, Supplementary Figure 1 ).

IgG plasma B-cell proportion was higher in BAL from COVID-19 compared to non-COVID-19 subjects (mean 22 ± 33.8 vs 3.9 ± 1.7 p = 0.004, Figure 8A ). A trend toward a higher proportion of activated and resting memory B cells in infected compared with non-infected was also observed ( Figure 8A ). Similarly, IgA activated, plasma, and resting memory B-cell proportions were higher in COVID-19 than in non-COVID-19 subjects ( Figure 8B ). The percentages of IgA and IgG naïve, activated, plasma, resting, and tissue memory B-cell proportions remained similar irrespective of virus detection in BAL ( Figures 8C, D ) and did not have an impact on COVID-19 outcome ( Figures 8E, F ).

The following cytokines related to inflammation and B-cell response were quantified: MIP-1 alpha, G-CSF, IL-1β, IL-8, S100A8, TNF-alpha, MCP-1, CXCL10, IL-1α, IL-6, M-CSF, and S100B. BAL from SARS-CoV-2-infected individuals had higher levels of MIP-1 alpha, G-CSF, IL-1β, IL-8, S100A8, TNF-alpha, MCP-1, IL-1α, IL-6, M-CSF, and S100B, compared to those of uninfected ones ( Figure 9A ). The levels of all cytokines persisted after virus elimination from BAL ( Figure 9B ). Moreover, IL-1α, IL-1β, and IL-8 were statistically significantly higher in virus-free compared to virus-containing BAL, and their levels increased during disease progression (mean IL-1α SARS-CoV-2+ BAL vs SARS-CoV-2 neg BAL 15 ±7.7 vs 72 ± 21 respectively, p = 0.02; mean IL-1β: 1,815 ± 718 vs 71 ± 24, respectively, p = 0.01; mean IL-8: 14,255 ± 4,167 vs 2,540 ± 1,543 respectively, p = 0.04). These data indicated that the persistence of these cytokines is independent of the presence of the virus in the BAL. Conversely, an opposite scenario was observed for CXCL-10 ( Figure 9A ): high levels of CXCL-10 were detected in SARS-CoV-2+ BAL and then decreased with time (mean 591 ± 187 vs 52 ± 12, respectively, p = 0.001). The kinetics of the cytokine level revealed a positive correlation between the amounts of IL-1β and IL-1α and the time from symptom onset to sampling (IL-1β: p = 0.04, r = 0.3; IL-1α: p = 0.02, r = 0.3, Figure 9C ). Additionally, both IL-1β and IL-8 correlated with RBD-specific IgA (IL-1β p = 0.006 r = 0.4, IL-8 p = 0.0003 r = 0.5, Figure 9D ), whereas both S100A8 and IL-6 correlated with S2-specific IgA (S100A8 p = 0.04 r = 0.4, IL-6 p = 0.02 r = 0.37; Figure 9E ). Higher levels of IL-1β were found in BAL from survivors but only when the virus replicates (mean survivors vs non-survivors 102 ± 40 vs 15 ± 7.7, p = 0.05, Figures 9F, G ). Therefore, patients having low levels of IL-1β in SARS-CoV-2+ BAL might be prone to develop fatal COVID-19 as compared to those with high levels. In contrast, at a later phase of the disease, IL-1β levels might not affect survival.

These results were confirmed by applying Bayesian logistic regression on BAL cytokines measurements to model the association of the odds ratio of non-surviving against surviving probability (referred to as odds ratio). Cytokines were significantly associated with the odds ratio only when samples were further stratified according to the phase of the disease. In agreement, only IL-1β was statistically significantly associated with the odds ratio early in infection. Indeed, the estimated one-sided 97.5% credible interval of its coefficient was less than 1, indicating it was a negative risk factor. IL-1β loses its association with the odds ratio later in the disease. These data suggest that a higher level of lung IL-1β might be predictive of patient survival only during the early phase of infection. Additional studies using longitudinal sampling are needed to definitively establish whether IL-1β could be a marker of severity during the course of the diseases.