Early treatment with inhaled GM-CSF improves oxygenation and anti-viral immunity in COVID-19 induced lung injury – a randomized clinical trial


 Granulocyte-macrophage colony-stimulating factor (GM-CSF) instructs monocytes to differentiate into alveolar macrophages (AM) that preserve lung homeostasis. By comparing AM development in mouse and human, we discovered that COVID-19 patients showed marked defects in GM-CSF-dependent AM instruction. The multi-center, open-label, randomized, controlled SARPAC-trial evaluated the efficacy and safety of 5 days of inhalation of rhu-GM-CSF (sargramostim, Leukine®) in 81 non-ventilated patients with COVID-19 and hypoxemic respiratory failure identified by PaO2/FiO2 ratio < 350mmHg. At day 6, more patients in the sargramostim group experienced at least 25% improvement in oxygenation compared with the standard of care group. Higher numbers of circulating class-switched B cells and effector virus-specific CD8 lymphocytes were found in the sargramostim group. Treatment adverse events, including signs of cytokine storm, were not different between active and control group. This proof-of-concept study demonstrates the feasibility and safety of inhaled GM-CSF in restoring alveolar gas exchange, while simultaneously boosting anti-COVID-19 immunity. ClinicalTrials.gov (NCT04326920).


BACKGROUND
Gas exchange in the lungs occurs in the alveoli, over a thin alveolocapillary membrane that allows the rapid diffusion of oxygen and carbon dioxide between the alveoli and the red blood cells that circulate in the lung capillaries. When pathogens or dust particles enter the lung, inflammation and edema in the alveolar wall is therefore always kept to a minimum, to protect the delicate gas exchange apparatus. The alveolocapillary membrane is however also an easy portal of pathogen entry into the bloodstream and needs to be defended. One way by which evolution has solved the problem of lung defense in the absence of inflammation is via the unique placement of alveolar macrophages (AM) as first line of defense. Tissue resident AM adhere to and crawl on alveolar epithelial cells and in this exposed position continuously capture, phagocytose, conceal and neutralize a large cargo of pathogens and particles from inhaled air, without causing inflammation [1][2][3] . At the same time, AMs handle and recycle surfactant, a detergent that keeps alveoli in an open inflated state by lowering the surface tension of the alveolar lining fluid 4 . Tissue resident AM develop early in life, and can occupy the alveolar niche through local low-grade proliferation without the need for replenishment by circulating monocytes 5 . Within the first week of life, alveolar epithelial-cell derived granulocyte-macrophage colony stimulating factor (GM-CSF) provides the instructive signal that programs fetal monocytes to become homeostatic tissue-resident AMs in the alveolar niche [5][6][7][8][9] . These resident AMs then self-maintain with only minimal input from circulating monocytes in both mouse 5,10,11 and human 12-14 . Severe inflammatory insults can however temporarily deplete tissue resident AM and lead to the recruitment of monocytes that can later develop into long-lived AMs when tissue homeostasis is restored 10,15,16 .
In patients with severe Coronavirus disease 2019 (COVID- 19), infection with SARS-CoV-2 virus has profound effects on alveolar homeostasis, , resulting in hypoxemia through impaired alveolo-capillary gas exchange and disbalances in the ratio of lung ventilation and perfusion, potentially culminating in the acute respiratory distress syndrome (ARDS) 17,18 . The initial viral replication phase that occurs in lung alveolar epithelial cells (AECs), AMs and capillary endothelial cells is followed by a second hyperinflammatory phase in which alveolar homeostasis is severely disturbed. During this phase of hyperinflammation, there is activation of the complement and coagulation cascade leading to inflammatory cell recruitment, and to microthrombi in the alveolar capillaries 19,20 . Simultaneously, there is release of cytokines that act systemically to induce profound fever, overproduction of ferritin and acute phase proteins by the liver, progressive dysfunction of the liver, kidney and heart, and depletion of circulating lymphocytes and eosinophils through hemophagocytosis or suppression of hematopoiesis 18,[20][21][22][23][24][25][26] . Currently, oxygen supplementation and ventilatory support, anti-coagulation, systemic steroids, and cytokine blockade are the mainstay of therapy for severe COVID-19, but therapies aimed at restoring alveolar homeostasis are lacking.
Most reports that have studied inflammation in the lung alveoli in severe COVID-19 have found an accumulation of dysregulated myeloid cells including neutrophils and macrophages in the bronchoalveolar space, and these can lead to a state of immune silence hampering T cell activation [27][28][29][30][31][32][33][34] . Although macrophages and monocytes recovered by bronchoalveolar lavage are often termed AM, careful single cell mRNA analysis of BAL fluid cells has shown that the accumulation of heterogeneous pro-inflammatory and immunosuppressive monocytederived macrophages is at the same time accompanied by the loss of homeostatic tissue 6 resident AMs, that can be reliably identified using a set of gene transcripts that include FABP4 and PPARG 27, 30,35 . The precise reason for the loss of tissue resident AM is currently unknown, but of great interest, since restoring AM function might herald return of alveolar homeostasis, relief from immune suppression and could form the basis of innovative therapy.
In an attempt to find new therapeutic options for severe COVID-19, we first studied the composition of BAL fluid immune cells in patients with confirmed COVID-19 pneumonia or patients with non-COVID lung infection or undergoing diagnostic bronchoscopy for other reasons. By comparing scRNAseq and CITEseq data from COVID-19 lung samples with transcriptomic data of murine fetal monocytes developing into AMs in wild type and Csf2 -/mice, we found that human COVID-19 lung macrophages are characterized by the loss of an evolutionary conserved GM-CSF-mediated instruction that drives AM development. Based on this translational finding, we initiated the SARPAC (SARgramostim in Patients with Acute COVID-19) randomized controlled proof-of-concept clinical trial that investigated the feasibility and safety of inhaled sargramostim (rhu-GM-CSF, Leukine®) treatment for hospitalized COVID-19 patients with signs of hypoxemia requiring oxygen supplementation 36 .
The primary objective was to study the impact and safety of 5 days of sargramostim inhalation treatment on parameters of alveolar gas exchange, while exploratory analysis included effects of sargramostim on development of cytokine storm and antiviral immunity.
The transitional monocyte cluster then further bifurcates into hemophagocytic macrophages 9 or tissue resident AMs (Fig. 1f). In healthy controls, the end state of most monocytes/macrophages was the tissue resident AM fate ( Fig. 1g left panel), whereas in COVID-19 patients most monocytes were predicted to differentiate into IFN-stimulated monocytes and hemophagocytic macrophages and not tissue resident AMs (Fig. 1g right   panel). The presence of CD163 + hemophagocytic macrophages in the alveolar lumen was also confirmed by immunohistochemistry in a fatal case of COVID-19 (Fig. 1d). So, in line with earlier reports 27, 30, 35 , we found that lung monocytes in COVID-19 patients fail to develop into AMs yet turn into highly pro-inflammatory monocytes and CD163 + hemophagocytic macrophages in the alveolar lumen.
We next aimed to unravel the mechanisms hampering monocyte to AM instruction during COVID-19. To more closely understand AM development, we turned to the mouse model, where the various stages of AM development from fetal monocytes and AM instruction by epithelial cell derived GM-CSF (encoded by the Csf2 gene) have been closely described by us and others [5][6][7]41 . We sorted lung monocyte and developing AM populations from lung tissue of wild-type mice at different time points from embryonic day 15 until adulthood and performed transcriptional micro-array analysis on these developmental stages (Fig. 1h). First, we focused on the genes that shared three characteristics: they were (i) AM specific and not found in other tissue resident macrophages 42 , (ii) lost in Csf2 -/mice and (iii) rescued in Csf2 -/mice upon treatment with inhaled GM-CSF. We regarded these genes reflective of the "murine GM-CSF-dependent lung macrophage signature" (Fig. 1i). Accordingly, neonatal and postnatal day 9 (PND9) lung macrophages already strongly resembled adult tissue resident AMs at a transcriptional level and acquired key adult tissue resident AM genes such as Ear1, Plet4 and Pparg (Fig. 1i). Macrophages sorted from the alveolar lumen of PND9 Csf2 -/mice lacked this GM-CSF-dependent AM signature, but early life inhaled treatment with recombinant GM-CSF (rGM-CSF) in Csf2 -/mice restored this gene signature (Fig. 1i right column). Next, we looked into genes that were (i) upregulated in macrophages in Csf2 -/mice as compared to wild type AMs, but (ii) were downregulated again upon treatment with inhaled GM-CSF (Fig 1j right column). These genes thus reflect the lack of GM-CSF instruction and therefore considered as the "murine lack-of-GM-CSF lung macrophage signature" (Fig. 1j). Some of these genes are clearly pro-inflammatory and driven by type I interferon signaling, such as Cxcl9, Cxcl10, Ifit1, Ifit2 and Rsad2 (Fig 1j). Altogether, this demonstrates that (i) GM-CSF is not only a critical instructive cytokine for murine tissue resident AM differentiation 5 but also that (ii) lung monocytes develop into pro-inflammatory cells in the absence of GM-CSF, a state that is however reversible by inhaled GM-CSF treatment.
Finally, we compared murine AM differentiation transcriptional states with human COVID-19 scRNAseq datasets. We projected the murine GM-CSF-dependent lung macrophage signature on the patients BALF clusters obtained from sequencing data. Genes associated with presence of GM-CSF (Fig. 1i), as for example PPARG 7 (Fig. 1l left UMAP), were highly expressed by human tissue resident AMs found in healthy controls ( Fig. 1k left UMAP). Vice versa, genes upregulated in the absence of GM-CSF (Fig. 1j), as CXCL10 ( Fig. 1l right UMAP), IL18BP, TNFSF13B and MMP14, were enriched in IFN-stimulated monocytes and hemophagocytic macrophages from COVID-19 infected patients (Fig. 1k right UMAP). In conclusion, these data 11 demonstrated that mononuclear cells in the lungs of COVID-19 patients lacked evolutionary conserved GM-CSF instruction, leading to a lack of tissue resident AM and accumulation of pro-inflammatory monocyte/macrophage populations typically seen in GM-CSF deficient states. This provided the rationale for a randomized clinical trial where inhaled GM-CSF was administered to hypoxemic COVID-19 patients, in an attempt to improve alveolar homeostasis, with the underlying hypothesis that such treatment would promote the differentiation of transitional monocytes into AMs, reduce hyperinflammation, and restore the gas exchange apparatus.

Patients enrolled in SARPAC randomized clinical trial of inhaled GM-CSF
From March 25 through September 28, 2020, 81 hospitalized COVID-19 patients were included in the SARPAC trial at 5 participating sites in Belgium. Enrolled patients had acute hypoxemic respiratory failure (oxygen saturation below 93% on ≥ 2 liters oxygen per minute or a ratio of the partial pressure of oxygen (PaO2) to the fraction of inspired oxygen (FiO2; P/F ratio) below 350 mmHg). Patients with a serum ferritin > 2000 µg/L or already on mechanical ventilation were excluded from participation. All participants provided oral and written consent (full clinical study protocol available in Extended Data). Eighty-one patients were randomly assigned in a 1:1 ratio to receive 5 days of twice daily 125 µg of inhaled sargramostim (rhu-GM-CSF, Leukine®) on top of standard of care (SOC) or to SOC alone (Fig.   2). Two patients in the sargramostim group showed progression of disease within the first 5 days requiring initiation of mechanical ventilatory support. In these patients inhaled sargramostim was replaced per protocol by intravenous sargramostim 125 µg/m 2 body surface area, to minimize risk of aerosol spread in the ICU. Four additional patients in the sargramostim group and two patients in the SOC group received i.v. sargramostim from day 6 onwards (after the primary endpoint was measured), based on clinician's decision when deterioration occurred.
Seventy-three patients reached the evaluable primary endpoint (oxygenation parameters at day 6). Two patients discontinued treatment prematurely, three patients refused arterial puncture at day 6, and another three patients were excluded from analysis because they had a negative P(A-a)O2 gradient at randomization or day 6, signifying an error in FiO2 recording.
All primary efficacy evaluable patients (N=73) were included in a modified intention-to-treat

Effect of sargramostim inhalation on primary endpoint
The proportion of patients with at least 25% improvement in lung oxygenation parameters after 5 days of treatment compared to baseline values was higher in the sargramostim group than in the SOC group (Table 2 and Extended Data Fig. 3a-b) when assessed by measuring the gradient between partial pressure of oxygen between the alveolar air and the arterial blood   43 . However, no evidence for an enhanced treatment effect of sargramostim could be found in post-hoc specified subgroups (concomitant glucocorticoid use, P(A-a)O2 gradient above or below the median value at randomization, CRP and ferritin level at randomization above or below the median value), although patients numbers were small to draw definitive conclusions (Extended data Fig. 4).

Effect of sargramostim inhalation on secondary clinical and safety end points
In this small proof-of-concept study, no evidence for a treatment effect of sargramostim could be found for any of the supportive endpoints listed in Table 2, including duration of hospital stay, progression to mechanical ventilation or ARDS, and all-cause mortality rate at 4 weeks post-randomization. Initially, we had planned to look at incidence of secondary hemophagocytic lymphohistiocytosis (HLH), but incidence was so low that scoring for this parameter was discontinued. Patients were also seen at follow up 10-20 weeks following hospital discharge, to study development of secondary lung fibrosis by high resolution computed tomography (HRCT) fibrosis score analysis. However, the incidence of fibrotic changes was so low in both groups we could not make an assessment of this secondary endpoint.
Adverse events were generally balanced between both groups, except for mild epistaxis which was more reported in the sargramostim group (20.0% versus 4.9% in SOC). For the serious adverse events (SAEs) not leading to mortality, we did not observe differences between both groups (Table 3). Overall, twelve patients died during the study, of which one patient during 15 the first 6 days and 4 patients during the first 28 days after randomization. We did not find evidence for differences in mortality between study arms and if any, mortality was higher in the control group (Table 3).

Effect of sargramostim inhalation on innate immune landscape
Inappropriate and uncontrolled release of pro-inflammatory cytokines such as TNFα, IL-1, IL-6, IL-8 and IL-18 contributes to disease progression and is associated with worse prognosis in patients with severe COVID-19 25,43,44 . Since increased numbers of GM-CSF producing T cells 31 and higher serum concentrations 45 of GM-CSF were reported in some patients with COVID-19, and since GM-CSF can boost the production of pro-inflammatory cytokines and is even blocked in several ongoing COVID-19 trials 46 , an a priori defined pharmacodynamic endpoint was to study signs of enhancement of cytokine release syndrome by inhaled sargramostim.
For patients in 3 selected study sites, we quantified serum concentrations of pro-inflammatory cytokines at randomization and day 6, and compared these with a cohort of age-matched healthy controls (HC). Although pro-inflammatory cytokines were higher in COVID-19 patients at randomization, they declined over the next days and were not increased by 5 days of sargramostim ( Fig. 3a). At randomization, serum concentration of GM-CSF was low in most patients and comparable to HC samples (Extended data Fig. 5a). Cytokine release promotes complement activation and thrombosis, which are ominous drivers of severe COVID-19 immunopathology 20, 47 . The anaphylatoxin C5a promotes alveolar inflammation by recruitment of C5aR + neutrophils and monocyte-derived cells 48 , and C5a concentration was higher in COVID-19 patients in our cohort, compared with HC at randomization. At day 6, C5a concentration dropped independently of sargramostim treatment (Fig. 3b). High serum concentrations of ferritin and CRP, and low circulating numbers of lymphocytes and eosinophils can also be a sign of cytokine release syndrome or secondary HLH in COVID-19, and these were measured as part of the secondary and safety assessment. Between randomization and day 6, virtually all patients demonstrated amelioration of these key laboratory parameters, irrespective of their treatment arm ( Table 2). Systemic GM-CSF has been shown to promote the priming of neutrophils in other forms of ARDS, and low density CD24 + activated neutrophils 49 were seen in higher numbers in our COVID-19 cohort, compared with HC. Sargramostim treatment did however not promote this neutrophil activation state (Fig 3b). GM-CSF is the prototypical growth factor for DCs, and circulating cDC2s, cDC3s and pDCs were depleted in COVID-19 patients at randomization and after 6 days, irrespective of sargramostim treatment, and similar findings were seen for circulating basophils (Extended data Fig. 5b). Systemic GM-CSF promotes emergency hematopoiesis, and could be involved in expansion of myelomonocytic cells 21 . However, after 6 days of follow up, there was no increase in the percentage of CD14 + or CD16 + monocytes, in patients receiving sargramostim, compared with those in the SOC group (Fig. 3b). Downregulation of HLA-DR on monocytes is frequently found in ARDS associated immunosuppression 50 , and we did find evidence of this in our COVID-19 patients at randomization, compared with HC (Fig 3c). Six days later however, HLA-DR expression on monocytes was restored, irrespective of the treatment arm.

Effect of sargramostim inhalation on adaptive antiviral immune response
Both humoral and cellular (cytotoxic) immune responses are important for elimination of viral particles, and the accumulation of monocyte-derived macrophages seen in COVID-19 patients has been proposed to contribute to an immune silenced state in which it is hard to activate adaptive immune cells 29 . GM-CSF has demonstrated significant immune stimulating effects in models of bacterial and viral lung infection [51][52][53]  Together, these data demonstrate that inhalation of sargramostim in COVID-19 patients is feasible and safe, leading to improved gas exchange in the lung, while simultaneously boosting the immune response against the virus.

Several papers have shown that COVID-19 is accompanied by influx of mononuclear cells in
the alveolar space, at the expense of tissue resident AM which normally stably occupy this niche 27-30, 32 . The precise reasons for deficiency in tissue resident AMs in COVID-19 have remained elusive, and one possible explanation was that AMs are directly infected by SARS-CoV2 35 . This leaves the question why recruited newcomer monocytes fail to differentiate into tissue resident AMs when they encounter an empty alveolar niche. By comparing scRNAseq and CITEseq data from COVID-19 lung samples with mini-bulk transcriptomic data of the various stages of fetal and postnatal AM development in wild type mice and GM-CSF-deficient mice, we found that recruited lung macrophages in COVID-19 lungs lack GM-CSF instruction.
GM-CSF is the prime cytokine of the alveolar niche, produced by type II alveolar epithelial cells (AEC) 6 . This cytokine induces the master lipid-handling transcription factor PPARγ, and causes fetal and adult monocytes to differentiate into tissue resident AMs 5,7,41 . In adult mice, the return of tissue resident AMs following their depletion by infectious insults depends on GM-CSF production by type II AECs 6, 56 . We have tried to measure GM-CSF levels in the BAL fluid of COVID-19 patients and other pulmonary infections but failed to detect it (data not shown), and serum levels of GM-CSF were very low in our cohort, despite an earlier report 45 . Several explanations are possible for the lack of GM-CSF instruction on recruited monocytes in COVID-19 lungs. First, through their expression of ACE2 receptor, type II AECs are prime targets of SARS-CoV2 infection 28,35 , so demise of these cells in COVID-19 pneumonia would lead to loss of a major source of GM-CSF. Such a scenario has indeed been described in other forms of ARDS and acute lung injury, where maintained GM-CSF and thus reduced AECII injury or better AECII regeneration was associated with better survival 57,58 . Secondly, subversion of GM-CSF 20 production might be unique to the beta-coronavirus family. The SARS-CoV1 virus 3C-like proteinase, which is conserved in SARS-CoV2, specifically subverts the production of GM-CSF but not other cytokines when overexpressed in lung epithelial cells 59 . Finally, the inflammatory milieu of the COVID-19 lung with high numbers of neutrophils and pro-inflammatory cytokines could inhibit the differentiation of AMs by competing for available GM-CSF 60 or by antagonizing downstream signaling induced by GM-CSF instruction. Indeed, hyperinflammation in COVID-19 is accompanied by oxidative stress, a known suppressor of GM-CSF production by type II AECs 61 .
Previous work in patients with pulmonary alveolar proteinosis (PAP), a disease caused by autoantibodies to GM-CSF or by genetic defects in CSF2RA have also shown that human AMs fail to differentiate into the anti-inflammatory surfactant-handling cells that AMs normally are.
Not surprisingly, inhalation of GM-CSF has been proposed as a treatment for this rare disorder, with no noticeable side effects yet durable and favorable outcome on lung function 9, 62-64 .
Based on our translational findings in mice and humans, and the prior success of inhaled GM-CSF therapies in humans, we initiated a randomized controlled proof-of-concept clinical trial to study if inhaled sargramostim treatment would improve alveolar oxygenation, a primary readout of the function of the alveolus. After 5 days of inhaled sargramostim, we found more patients with at least 25% improved oxygenation, as measured by a decrease in the P(A-a)O2 gradient, a measure for the degree of ventilation/perfusion mismatch and shunting often seen in patients with ARDS. When measured using the PaO2/FiO2 index, part of the standard assessment of ARDS, the outcome was less clear however. Recently, it was proposed that the reported as hypofunctional 73 . GM-CSF is the prime cytokine boosting the numbers and the function of DCs that cross-present antigens derived from infected AECs to CD8 T cells 52,53 .
Despite the beneficial effects of lung GM-CSF on alveolar homeostasis, gas exchange and antimicrobial immunity, not only described in the literature, but also emanating from our clinical intervention, there is still a lot of controversy surrounding GM-CSF as a therapeutic target in this disease 46,74 . Indeed, given the known role of GM-CSF in emergency myelopoiesis and CAR-T cell associated cytokine release syndrome 75 , and the observation of increased numbers of GM-CSF producing T cells 31  In conclusion, this translational study from mice to humans with COVID-19 identified inhalation with GM-CSF as a potential therapy for COVID-19 pneumonia, improving blood oxygenation while at the same time boosting antiviral immunity with minimal side effects.

Mice and treatments
The following mice were used in this study; female C57BL/6 mice (aged 6-10 weeks) were purchased from Janvier (France); Csf2 −/− mice were bred at the animal facility of the VIB- were performed according to local guidelines and Belgian animal protection law.

Flow cytometry and cell sorting of murine samples
Cell sorting was performed on a FACSAria II cytometer. After cell sorting, purity was checked (always >95%). For flow cytometry, lungs were cut into small pieces, incubated in RPMI containing Liberase TM (Roche) and DNase (Roche), and then syringed through a 19-gauge needle to obtain a homogenous cell suspension. Red blood cells were lysed for 4 min at room temperature in 1 ml osmotic lysis buffer. Cells were sorted exactly as in 5 (macrophages, pre-AMs and AMs in WT see Fig.3, GM-CSF treated mice gating see Fig.6

27
The microarrays were analysed using the limma R package (v 3.42.2). The Robust Multi-array Average (RMA) procedure was used to normalize data within arrays (probeset summarization, background correction and log2-transformation) and between arrays (quantile normalization). Probesets were filtered and converted into gene symbols using the mogene10sttranscriptcluster.db R package (v 8.7.0).
To identify the AMF signature genes that are CSF2 dependent we first calculated the DE genes between the AMF group and the primitive MF groups combined with the BM_mono and
The demultiplexing of the raw data was performed using CellRanger software (10x -version 4.0; cellranger mkfastq which wraps Illumina's bcl2fastq). The reads obtained from the demultiplexing were used as the input for 'cellranger count' (CellRanger software), which aligned the reads to a merged human/SARS-CoV-2 genome using STAR and collapses to unique molecular identifier (UMI) counts. In order to maintain explicit control over all gene and cell quality control filters, we used the raw feature-barcode matrix instead of the the filtered feature-barcode matrix generated by CellRanger. As an initial filtering, we removed all cells with less than 200 genes and genes expressed in less than 3 cells. Key safety endpoints included all-cause mortality, serious adverse events, sepsis and septic shock during hospital stay. Adverse events were recorded according to the system organ class and preferred terms in the Common Terminology Criteria for Adverse Events, version 6.0.

Sample collection and processing
Peripheral venous blood specimens were collected from healthy individuals and study patients Bovine Serum Albumin, 0,05% NaN3, 1 mM EDTA) and Brilliant Stain buffer (BD Biosciences).
Cells were fixed, permeabilized and intracellular stained with antibodies using FoxP3 staining buffer (Thermofisher; 00-5523-00) following manufacturer's protocol. Acquisition and analysis of labeled cell suspensions was performed with a FACSymphony flow cytometer (BD biosciences) and subsequent analysis of data with FlowJo10 software (BD biosciences).
Antibodies used to define PBMC populations can be found in extended data table 1.

T cell restimulation and FluoroSpot
To quantify SARS -CoV2 specific CD4 and CD8 T cells, peptide restimulation with Fluorospot for   IFNγ and IL-2

Sample calculation and power analysis
The target difference is the difference from baseline measured at the primary endpoint (at day 6) between the control and the treated group. Given a sample size of 40 patients on each treatment, a minimal improvement of 10% in the treated group relative to the control group will be detected as significant at a significance level of 0.01 with a power of 0.90. The error variance was set at 100 units, corresponding with a standard deviation of 10 units. Sample calculation and power analysis were performed using Genstat.

Statistical analysis
All efficacy endpoints were analyzed in the intention-to-treat analysis which included all patients who had undergone randomization. Patients with missing data (i.e. no arterial blood gas analysis at baseline and/or day 6) were excluded from the analyses for which the missing data are necessary. Patients with a negative P(A-a)O2 gradient were excluded for oxygenation analyses, given these values are biologically not possible.
The number of patients that experienced at least 25%, 33% and 50% improvement in oxygenation was compared between the sargramostim group and the standard of care group by a Chi-square test. The median change from baseline in oxygenation to day 6 was analyzed by a Brown-Mood test. P values were two sided, and any P value less than 0.05 was considered   Male sexn (%) 25         h, Schematic overview of mini-bulk micro-array setup used on monocytes and macrophages isolated from lungs of WT or Csf2 -/mice after PBS or rGM-CSF treatment.
i, Heatmap showing the relative expression of the top genes present in the murine GM-CSF-dependent lung macrophage signature. The relative expression of these genes by monocytes or macrophages sorted from the lung at different time points during embryonic development or post-natally are shown. The relative expression of these genes by macrophages sorted from lungs of PND9 Csf2 -/mice treated with PBS (left) or rGM-CSF (right) is shown in the last two columns. j, Heatmap showing the relative expression of the top genes present in the murine lack-of-GM-CSF lung macrophage signature. The relative expression of these genes by monocytes or macrophages sorted from the lung at different time points during embryonic development or post-natally are shown. The relative expression of these genes by macrophages sorted from lungs of PND9 Csf2-/-mice treated with rGM-CSF is shown in the last column. k, Projection of the murine GM-CSF lung macrophage signature on patient BAL CITE-Seq data. In the left UMAP, cells are highlighted that have a gene signature that corresponds to the murine GM-CSFdependent lung macrophage signature. In the right UMAP, cells are highlighted that have a gene signature that corresponds to the murine lack-of-GM-CSF lung macrophage signature. l, UMAP representing the expression two conserved genes between human and mouse that represent a GM-CSF gene signature (PPARg, left UMAP) or a lack-of-GM-CSF gene signature (CXCL10, right UMAP).