The dysregulated innate immune response in severe COVID-19 pneumonia that could drive poorer outcome

Background Although immune modulation is a promising therapeutic avenue in coronavirus disease 2019 (COVID-19), the most relevant targets remain to be found. COVID-19 has peculiar characteristics and outcomes, suggesting a unique immunopathogenesis. Methods Thirty-six immunocompetent non-COVID-19 and 27 COVID-19 patients with severe pneumonia were prospectively enrolled in a single center, most requiring intensive care. Clinical and biological characteristics (including T cell phenotype and function and plasma concentrations of 30 cytokines) and outcomes were compared. Results At similar baseline respiratory severity, COVID-19 patients required mechanical ventilation for significantly longer than non-COVID-19 patients (15 [7–22] vs. 4 (0–15) days; p = 0.0049). COVID-19 patients had lower levels of most classical inflammatory cytokines (G-CSF, CCL20, IL-1β, IL-2, IL-6, IL-8, IL-15, TNF-α, TGF-β), but higher plasma concentrations of CXCL10, GM-CSF and CCL5, compared to non-COVID-19 patients. COVID-19 patients displayed similar T-cell exhaustion to non-COVID-19 patients, but with a more unbalanced inflammatory/anti-inflammatory cytokine response (IL-6/IL-10 and TNF-α/IL-10 ratios). Principal component analysis identified two main patterns, with a clear distinction between non-COVID-19 and COVID-19 patients. Multivariate regression analysis confirmed that GM-CSF, CXCL10 and IL-10 levels were independently associated with the duration of mechanical ventilation. Conclusion We identified a unique cytokine response, with higher plasma GM-CSF and CXCL10 in COVID-19 patients that were independently associated with the longer duration of mechanical ventilation. These cytokines could represent the dysregulated immune response in severe COVID-19, as well as promising therapeutic targets. ClinicalTrials.gov: NCT03505281.

launched to find effective therapies likely to improve outcome in coronavirus disease 2019 .
SARS-CoV-2 infects type I and type II alveolar epithelial cells as well as alveolar macrophages, through binding to angiotensin-converting enzyme 2 (ACE2), triggering a type I interferon (IFN) response, and the release of a myriad of pro-inflammatory cytokines (i.e. interferon (IFN)-γ, interleukin (IL)-1RA, IL-6, IL-8, IL-10, IL-19, monocyte chemoattractant protein (MCP)-1, MCP-2, MCP-3, C-X-C motif chemokine ligand (CXCL)9, CXCL10, CXCL5, tumor-necrosis factor (TNF)-α). Due to the massive T cell stimulation, lower levels of T lymphocytes are observed and all of these abnormalities being associated with disease severity [2][3][4][5][6]. Complement activation, and especially the C5a/C5aR1 axis was also implicated in COVID-19 lung pathology [7]. Accordingly, the description of the so-called cytokine storm has been advocated as the cause of organ dysfunction and death during COVID-19. For now, dexamethasone is the only treatment that has proven to be effective in reducing 28-day mortality in severe COVID-19 patients receiving mechanical ventilation in a randomized clinical trial [8]. These data support the existence of a unique dysregulated immune response that could be one of the most promising therapeutic targets to date.
However, several caveats must be underlined. First, immune pathogenesis is poorly understood, and comparisons of the immune response between COVID-19 patients and patients with pneumonia of other origins are scarce [4,9]. Second, the relevance of the cytokine storm paradigm is being questioned [10,11]. We recently reported that COVID-19 patients had lower concentrations of interleukin (IL)-6 compared to non-COVID-19 patients with severe pneumonia [12]. Others showed that mean IL-6 concentrations were nearly 100 times higher in patients with cytokine release syndrome, 27 times higher in patients with sepsis and 12 times higher in patients with ARDS unrelated to COVID-19 [13]. Third, Remy et al. showed that COVID-19 patients display a severe immunosuppressive phenotype [14]. Finally, a delayed type I IFN response is associated with an impeded viral clearance and could promote the accumulation of pathogenic inflammatory monocyte-macrophages resulting in cytokine/chemokine release within the lung [15,16]. These results are of utmost importance, since we also recently reported that the alveolar viral load is tightly correlated with subsequent severity in COVID-19 acute respiratory distress syndrome (ARDS) [17]. Modulating immunity remains a challenge, and there is a compelling need to identify the dysregulated immune response driving the outcomes observed in COVID-19 patients, in order to find the most relevant therapeutic targets.
Thus, our study aimed to compare clinical and biological characteristics, immune response and outcomes between non-COVID-19 and COVID-19 patients with severe pneumonia.

Study design and participants
The present work is a prospective, exploratory substudy of the ongoing LYMPHONIE trial (ClinicalTrials.gov NCT03505281), initiated in November 2018 at the University Hospital of Dijon-Bourgogne (France). Patients were eligible if they had severe community-acquired pneumonia (CAP): 1) pneumonia (≥ 2 acute signs including cough, purulent sputum, dyspnea, chest pain, temperature < 35 °C or ≥ 38.5 °C, and new radiological pulmonary infiltrate); 2) at least two criteria of the quick-Sequential Organ Failure Assessment (SOFA) score (systolic blood pressure ≤ 100 mm Hg, respiratory rate ≥ 22, Glasgow score < 15) and/or the need for mechanical ventilation (MV) and/or vasopressors; and 3) diagnosed within 48 h following admission. Non-inclusion criteria were: < 18 years, pregnant women, persons under legal protection, decision to limit care, known immune deficiency, chronic disorder known to cause deep lymphopenia (i.e. cirrhosis, lympho-or myeloproliferative syndrome, solid cancer or active systemic lupus), hospitalization for sepsis within the previous 3 months. Non-COVID-19 CAP patients were included until February 20, 2020. COVID-19 patients were eligible if they were tested positive for SARS-CoV-2 by reverse transcriptasepolymerase chain reaction (RT-PCR) on one respiratory sample. Oral consent was obtained from the patient or their legal representative. Approval was obtained from the ethics committee (Comité de Protection des Personnes SUD MEDITERRANEE V; 2017-A03404-49).

Variables of interest, clinical outcomes, and data collection
Clinical and biological parameters, severity scores (SOFA [18], Simplified Acute Physiology Score (SAPS II) [19] and Pneumonia Severity Index (PSI) [20]) were calculated at the time of inclusion. ARDS was defined according to the Berlin definition [21], and septic shock was defined as persistent hypotension requiring vasopressors and a serum lactate level > 2 mmol/L despite adequate volume resuscitation. Clinical outcomes were recorded up to 30 days after admission, namely: 30-day mortality, hospital-and ICU-length of stay, duration of MV and the occurrence of ventilator-acquired pneumonia (VAP). Dedicated clinical research assistants collected all data using a standardized electronic case report form. Automatic checks were generated for missing or incoherent data.

Sample collection
Ethylenediamine tetraacetic acid blood (plasma biomarker) and heparin anticoagulated blood (cell stimulation) were obtained after inclusion of the patient (within 48 h of hospital admission, with a diagnosis of severe community acquired pneumonia and according to the inclusion and non-inclusion criteria). Within 4 h following sampling, plasma was collected after centrifugation at 2000 x g for 10 min at 4 °C and stored at −80 °C until use, without freeze-thaw cycle. All samples were collected and stored in the biological resource center of Dijon University Hospital (CRB Ferdinand Cabanne; http://www.crbfe rdina ndcab anne.fr/; NF S96-900 certification).

Lymphocyte phenotyping
Absolute counts for CD3 + , CD3 + CD4 + ,CD3 + CD8 + , CD3-CD19 + , CD3-CD56 + and/or CD16 + lymphocyte subsets were performed using an AQUIOS CL flow cytometer (Beckman Coulter, Hialeah, FL). The AQUIOS CL is a single platform, fully automated volumetric flow cytometry technology and uses a 488 nm solid state diode laser to measure light diffraction, fluorescence and electronic volume which estimates the relative size of the cells. Whole blood was incubated with the monoclonal antibody reagent followed by no-wash erythrocyte lysis.

Whole blood leukocyte ex vivo stimulation (WBS)
The standardized functional immunoassay QuantiF-ERON Monitor ® (QFM, Qiagen) was used according to the manufacturer's instructions. Within 3 h after blood sampling, one milliliter of whole blood was incubated at 37 °C for 20 ± 1 h with a QFM LyoSphere containing anti-CD3 T-cell receptor ligand and R848 (TLR7/8 ligand), or without LyoSphere (non-stimulated blood). Plasma was harvested after centrifugation at 4000 rpm for 10 min and stored at −80 °C until use. Whole blood leukocyte production of IFN-γ (IU/ml) upon stimulation was measured using ELISA (Qiagen), and fifteen other analytes using the Human Th9/Th17/Th22 Discovery Luminex ® assay (R&D Systems, USA): CD40 ligand, GM-CSF, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-10, IL-12, IL-13, IL-15, IL-17A, IL-33, TNF-α, CCL20. All samples were measured the same day by the same person and using the same kit. Samples with values above the ranges were tested again with further dilution. The cytokine production after stimulation was expressed as the difference of concentrations between plasma from stimulated blood and those from non-stimulated blood.
As a reference, we used samples from 7 control patients included in the Pneumochondrie study (Clin-icalTrials.gov NCT03955887) and who underwent QuantiFERON Monitor ® assay in the same conditions. The control population consisted of outpatients without fever during the previous 15 days and who underwent bronchoalveolar lavage for a non-infectious condition [24]. Samples were conditioned and measured in the same way as for the LYMPHONIE study. Principal component analysis (PCA) was used to identify potentially significant patterns of 64 variables: clinical characteristics and outcomes (n = 6), biological findings (n = 13), plasma cytokines (n = 30), cytokine production upon ex vivo stimulation (n = 15). PCA identifies factors, called principal components, that induce the most variation in the overall data [25]. These factors can be expressed as a linear combination of the correlated original variables (OVs). By inversing these formulas, we can express each OV as a linear combination of the factors and coefficients defining these linear combinations are interpreted as correlation coefficients. Moreover, each factor describes a percent of variation in the OVs. The number of factor to retain was determined using the scree plot [26] and the clinical interpretability of factors [27]. Finally, patients OVs data can be projected on the plans defined by the retained factors, which allows observing patient's patterns in a two-dimensional plot.
Spearman correlations were computed between cytokines and the most pertinent clinical outcomes associated with Covid-19 status in univariate analyses and PCA patterns. To account for potential confounders, we constructed multivariable linear regressions, with the MV duration as an outcome, for each selected cytokine. adjusted for age, COVID-19 status and either SOFA score (model 1) or PaO 2 :FiO 2 (model 2). The interaction between COVID-19 status and cytokines was systematically tested. Absence of serial correlation and heteroscedasticity were assessed using the DW statistic [28] and the White test [29] respectively. The proportion of variance explained by the models was quantified using the R 2 coefficient. Measures of association are expressed as mean differences ± standard error (SE). A p-value < 0.05 was considered statistically significant. Analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).

Characteristics of the study population
Sixty-three patients were enrolled (36 in the non-COVID-19 group, and 27 in the COVID-19 group).

PCA identifies two patterns linking immune response and outcomes, with a clear distinction between non-COVID-19 and COVID-19 patients
In PCA, four factors were retained for interpretation, which together accounted for 53.81% of all the information obtained by using the whole 64 available variables (clinical characteristics and outcomes (n = 6), biological findings (n = 13), plasma cytokines (n = 30), cytokine production upon ex vivo stimulation (n = 15)). Two of the four factors in the final pattern were associated with outcomes ( Table 2, Additional file 1: Fig.  S1). Factor 1 associated [1] baseline severity and extrarespiratory organ dysfunction (SOFA and PSI scores, lactate, creatinine and NT-ProBNP levels which were higher in non-COVID-19, except for SOFA score), [2] "classical" inflammatory mediators (i.e. CXCL1, CXC2, IL-1β, IL-2, IL-6, IL-8, TNF-α) and (3) "T-cell exhaustion (inverse correlation with lymphocyte count, and production of 15 cytokines upon ex vivo stimulation of whole blood with anti-CD3 and TLR7/8 ligands). Factor 4 associated: [1] outcomes (duration of MV, ICU and hospital length of stay, which were higher in COVID-19 patients at 30 days) and (2) only GM-CSF, CXCL10 and INF-α levels (  (Fig. 3). Planes defined by factors 2 and 3 did not enable any such discrimination between patients according to COVID-19 status.
Plasma GM-CSF, CXCL10 and IL-10 were independently associated with the duration of mechanical ventilation As the particular severity of COVID-19 patients was represented by a significantly longer duration of MV, we investigated whether immune response could explain this poor outcome. We observed a significant correlation between the duration of MV and GM-CSF (p < 0.0001), IL-10 (p < 0.0001), CXCL10 (p < 0.0001), CCL-2 (p = 0.001), CX3CL1 (p = 0.0233), and Granzyme B (p = 0.0143) (Additional file 1: Table S4). Based on all these results, we performed a multivariate linear regression to identify factors associated with the duration of MV in the first 30 days, using two models (i.e. SOFA score (model 1) or PaO 2 :FiO 2 ratio (model 2) as the variable of adjustment to account for severity) ( Table 3). Only GM-CSF was independently associated with a longer duration of MV in both models. We estimated an excess of 22.11 ± 8.36 min of MV per increase of 1 pg/mL of GM-CSF (p = 0.0105, model 1; and p < 0.0001, model 2). Interleukin-10 and CXCL10 were independently associated with a longer duration of MV only when adjusted for respiratory severity (model 2; p = 0.0359 and p = 0.049 respectively) ( Table 3). No significant interaction was found between these cytokines and COVID-19 status. Autocorrelation and heteroscedasticity were non-significant in all models.

Discussion
In this study, we identify a dysregulated cytokine production of GM-CSF and CXCL-10 in COVID-19 patients that was independently associated with the duration of MV which represents the distinctive poorer outcome observed in severe COVID-19 patients.
COVID-19 pneumonia is unique in comparison with pneumonia of other origins, with, in particular, sudden deterioration 7-9 days after onset of symptoms, severity of hypoxemia that contrasts with the relatively wellpreserved lung mechanics, and the protracted nature of ARDS [30,31], as observed in our study. The beneficial effect of dexamethasone in the most severe forms of COVID-19 argues for a dysregulated immune response that mediates lung injury and outcome [8]. To date, the characteristics of the immune response in COVID- 19 have not been completely elucidated. The terms "cytokine storm" and "macrophage activation syndrome" have been widely adopted to explain the immunopathogenesis, since the release of myriad inflammatory mediators is correlated with disease severity [4,32]. In our study, we first showed that despite similar respiratory severity, plasma concentrations of numerous cytokines characterizing the "cytokine storm" (i.e. IL-1β, IL-6, IL-8, IL-15, TNF-α, CCL2, CCL4, CCL19, CCL20, TGF-α) were lower in COVID-19 compared to non-COVID-19 patients. These findings contrast with the results of McElvaney et al., showing that plasma IL-6 levels were higher in COVID-19, compared to non-COVID-19 patients (4), but are in line with Sinha's retrospective observations (10). Conversely, based on a standardized functional immune-assay, we found that patients with severe pneumonia (whether COVID-19-related or not) displayed severe alterations of T-cell functionality on ex vivo CD3 and TLR7/8 stimulation. However, we observed a more unbalanced inflammatory/anti-inflammatory cytokine response in COVID-19 patients, as reflected by the IL-6:IL-10 and TNF-α:IL-10 ratios. Our results and those of Remy et al. [14] clearly challenge the classical paradigm of "cytokine storm"-mediated inflammation and show a markedly immunosuppressive phenotype in COVID-19 patients rather than hyperinflammation. However, we identified a dysregulated immune response that was independently associated with the peculiar longer duration of MV observed in severe COVID-19 patients. PCA analysis, including a comprehensive study of inflammatory and anti-inflammatory immune responses and outcomes, identified two interesting patterns that clearly distinguish non-COVID-19 from COVID-19 patients. COVID-19 patients presented a unique phenotype associating higher levels of GM-CSF and CXCL10 and a longer duration of MV. In addition, multivariate regression analysis confirmed that GM-CSF, CXCL10, and also IL-10 were all independently associated with the duration of MV after adjustment for potential confounders. Based on this comprehensive analysis, we hypothesize that these cytokines could represent part of the dysregulated immune response driving the prolonged need for MV in COVID-19 patients.
GM-CSF is secreted by epithelial cells from injured tissue or leukocytes, to induce survival, proliferation and/ or differentiation of myeloid cells [33], playing a critical role in regulating microbial defense [34]. However, as a consequence of aberrant Th-1 cell activation and inflammatory monocytes [35], aberrant production of GM-CSF may result in excessive inflammation and tissue damage, mainly by macrophage M1 polarization and overactivation [36]. CXCL10 is a pro-inflammatory Th1-chemokine driving migration to the site of infection of Th-1 T-cells, monocytes and neutrophils that express its receptor CXCR3 [37]. Production of CXCL10 has already been shown to be increased in SARS-CoV-1 [38]. Plasma concentrations of CXCL10 were recently reported to predict disease progression in COVID-19 [39,40]. Finally, Ichikawa et al. showed that blocking the CXCL10-CXCR3 signaling pathway in viral and non-viral ARDS preclinical models improved survival [41]. High levels of interleukin-10 were also observed in our study, whether COVID-19-related or not, and were independently associated with the duration of MV. However, no statistically significant association between the COVID-19 status and IL-10 was observed, which may be explained either by the fact that IL-10 is not the only or main driver of the length of MV in COVID-19 patients, either by a lack of statistical power, or both. As recently described, concurrent immune suppression and hyperinflammation are a hallmark of the pathogenesis in non-COVID-19 CAP, and argue against two distinct phases of host response [42]. Nevertheless, in COVID-19 patients, we observed higher IL-6:IL-10 and TNF-α:IL-10 ratios that could reflect an unbalanced overproduction of IL-10. IL-10 can be produced by most cells of innate and adaptative immune system, including lymphoid and myeloid cells, resulting in pleiotropic immunosuppressive functions (e.g. inhibition of the release of pro-inflammatory mediators, inhibitory effects on T cells and monocytes/macrophages…) [43]. As for CXCL10, it can be advocated that an initial more robust TH1 response with monocytes/macrophage over activation in COVID-19 patients, as compared to non-COVID-19 patients, could lead to a subsequent IL-10 overproduction in order to limit T cell responses. The role of IL-10, either beneficial or deleterious remains a difficult issue. However, IL-10-mediated immune suppression could drive the onset of secondary infectious complications and morbi-mortality, especially since 59% of COVID-19 patients presented at least 1 VAP event.
Our results are in line with those of Hue et al. that recently identified a plasma chemokine signature in COVID-19 ARDS patients (CXCL10, GMCSF and IL-10) which was associated with mortality (9). In addition, we also recently reported that both plasma and alveolar CXCL10 concentrations were independently associated with the duration of mechanical ventilation in COVID-19 ARDS patients [24].
Corticosteroids were shown to improve survival in severe COVID-19 patients in a recent therapeutic trial [8], and this may be linked to a decrease in CXCL10 levels [40], via the inhibition of the Th-1 pathway. Since corticosteroids have many side effects, targeted therapies likely to dampen the dysregulated immune response in COVID-19 are urgently needed.   Specifically, GM-CSF blockade (e.g. with lenzilumab) is increasingly being considered as a promising therapy in COVID-19 [33,36] and is under investigation in a phase III clinical trial (NCT04351152). In addition, CXCL10 blockade (e.g. Eldelumab/MDX-1100) may also represent an attractive therapy likely to dampen the dysregulated immune response that could be driving the duration of MV. Dampening inflammation in a context of high immune suppression is not always a hazardous route. During chimeric antigen receptor T (CAR-T) cell therapy, GM-CSF inhibition reduces cytokine release syndrome and neuro-inflammation, but enhances antitumoral CAR-T cell function [44]. In addition, an IL-6 blocker could partially rescue immune dysregulation caused by SARS-CoV-2 (2). These considerations are of utmost importance, since we reported that COVID-19 ARDS patients had a persistent alveolar SARS-CoV-2 viral load that correlates with severity [17]. Combined therapies associating immunomodulatory and antiviral agents are the most promising strategy likely to improve outcome in COVID-19 patients.
This study has several limitations. The statistical analysis suffers from a lack of power given the large number of variables studied and the small sample size. Then, we only used one approach (cytokine production after stimulation with anti-CD3 and TLR7/8 ligand) to measure T-cells exhaustion phenotype. However, it would have been also important to measure several phenotypic markers of exhaustion, namely PD-1, LAG-3, TIM-3 and CTLA-4. However, it was an exploratory study in the context of a pandemic emergency. Additionally, we used several statistical methods to assess the association between COVID-19 status and immune targets. Comparisons of immune response between non-COVID-19 and COVID-19 severe pneumonia are still scarce in the literature, even though they are mandatory to understand the distinctive pathogenesis of severe forms of COVID-19.