key: cord-0786096-d6knw9w5
authors: Saichi, Melissa; Ladjemi, Maha Zohra; Korniotis, Sarantis; Rousseau, Christophe; Ait-Hamou, Zakaria; Massenet, Lucile; Amblard, Elise; Noel, Floriane; Marie, Yannick; Bouteiller, Delphine; Medvedovic, Jasna; Pène, Frédéric; Soumelis, Vassili
title: Single cell RNA sequencing of blood antigen-presenting cells in severe Covid-19 reveals multi-process defects in antiviral immunity
date: 2020-07-21
journal: bioRxiv
DOI: 10.1101/2020.07.20.212837
sha: 956710bb8dbabf7f8d8c28e10efa6b733937875e
doc_id: 786096
cord_uid: d6knw9w5

COVID-19 can lead to life-threatening acute respiratory failure, characterized by simultaneous increase in inflammatory mediators and viral load. The underlying cellular and molecular mechanisms remain unclear. We performed single-cell RNA-sequencing to establish an exhaustive high-resolution map of blood antigen-presenting cells (APC) in 7 COVID-19 patients with moderate or severe pneumonia, at day-1 and day-4 post-admission, and two healthy donors. We generated a unique dataset of 31,513 high quality APC, including monocytes and rare dendritic cell (DC) subsets. We uncovered multiprocess and previously unrecognized defects in anti-viral immune defense in specific APC compartments from severe patients: i) increase of pro-apoptotic genes exclusively in pDC, which are key effectors of antiviral immunity, ii) sharp decrease of innate sensing receptors, TLR7 and DHX9, in pDC and cDC1, respectively, iii) down-regulation of antiviral effector molecules, including Interferon stimulated genes (ISG) in all monocyte subsets, and iv) decrease of MHC class II-related genes, and MHC class II transactivator (CIITA) activity in cDC2, suggesting a viral inhibition of antigen presentation. These novel mechanisms may explain patient aggravation and suggest strategies to restore defective immune defense.

Severe Acute Respiratory Syndrome-Coronavirus-2 (SARS-Cov-2) infection is at the origin of coronavirus disease 2019 , characterized by a first phase of benign flulike symptoms followed by an efficient control of the infection in most cases. In a second phase, disease aggravation may lead to acute respiratory failure, sepsis, and death 1-3 . This is due to a multiplicity of factors, which may act separately or in combination: i) exacerbated inflammatory reaction, with both systemic and organ-specific manifestations, ii) persistent viral load, and iii) defective anti-viral defense pathways [4] [5] [6] . Identifying the underlying cellular and molecular mechanisms is of paramount importance to understand COVID-19 physiopathology and guide the development of appropriate therapies.

As of today, studies have characterized the systemic inflammatory response, revealing an excess production of inflammatory cytokines such as IL-6 and IL-1 β 6-8 , while the role of TNF-α remains less clear 9 . The endothelium may also contribute to the overt inflammatory reaction through the production of soluble mediators [10] [11] . Insight also came from targeted immunotherapies, in particular anti-IL-6 compounds, which gave promising results in severe COVID-19 12, 13 .However, the cellular source and mechanisms underlying the excessive inflammatory response remain mostly unknown.

Another unresolved question relates to the inefficiency of the innate and adaptive immune system to control the virus in severe COVID-19 patients. It was suggested that production of interferon (IFN)-α , a major anti-viral cytokine, was decreased in the most severe patients, as compared to those with moderate disease 8 . However, a recent study argued that an increased IFN-α production might contribute to the pathogenic inflammatory response 14 . Other anti-viral mechanisms and their cellular source yet remain to be studied.

to the control of pathogens, and subsequent presentation of pathogen-specific Ag to T cells [15] [16] . Their study is challenging for three main reasons: i) they are found in very low numbers in the circulation and in tissue, ii) they lack specific lineage-defining markers, iii) they include an ever-increasing number of subsets [16] [17] .All DC subsets may potentially and variably contribute to modulating inflammatory response following viral sensing, producing anti-viral effector molecules, and priming Ag-specific adaptive immune response [17] [18] [19] . Plasmacytoid pre-DC (pDC) is a particular subset specialized in anti-viral immunity through the production of large amounts of type I IFN [20] [21] . Despite their central role in anti-viral defense, DC contribution to severe COVID-19 pathogenesis is not known.

In this study, we performed a high-resolution single cell RNA sequencing (scRNAseq) analysis of all APC subsets in the peripheral blood of COVID-19 patients. A pre-enrichment step enabled the characterization of even rare DC subsets, not captured in previous scRNAseq studies 14, 22 . This revealed previously unrecognized multi-process defects in anti-viral defense.

To characterize the state and molecular profile of circulating APC, we performed scRNAseq of APC-enriched PBMC from three moderate (non-mechanically ventilated, O2<10L/mn), and four severe (mechanically-ventilated or O2≥10L/mn) COVID-19 patients, at days 1 and 4 post-hospital admission, as well as two healthy controls (HC) ( Fig. 1a Uniform manifold approximation and projection (UMAP) and graph-based clustering of all severity group samples allowed us to manually annotate clusters with their cellular identities ( Fig. 1b ) . Using dimensionality reduction, we established a map of 21,416 APC (discovery set) identifying 6 subsets: 11,805 CD14 + and 1,354 CD16 + monocytes, 3,958 pDC, 301 CLEC9A + and 3,785 CD1c + DC (cDC1 and cDC2, respectively), and 213 Axl-Siglec 6 (AS)-DC ( Fig. 1b and Supplementary Fig. 1b ). These APC populations were captured across the whole dataset ( Supplementary Fig. 1a) . Their identity was confirmed by the expression of unique combinations of markers ( Fig. 1c ). CD14 + monocytes expressed exclusively their lineage-defining CD14 marker, whereas CD16 + monocytes expressed FCGR3A. As expected, all DC populations expressed higher levels of HLA-DR and CD86, as compared to monocytes ( Fig. 1c ) . AXL expression distinguished AS-DC from pDC, whereas CD1c and CLEC9A characterized the respective cDC subsets. The absence of CD3 ε confirmed the lack of T cell contamination ( Figure 1c ). Dimensionality reduction and graph-based sub-clustering allowed us to retrieve all APC subsets across patient groups.

Proportions, however, were different, with a relative decrease in pDC, and increase in CD14 + monocytes, in COVID-19 patients as compared to HC ( Fig. 1d and Supplementary Fig. 1c ).

Interestingly, we denoted an increased heterogeneity among pDC, CD14 + monocytes, and CD1C + DC in severe patients ( Fig. 1d ) . Overall, our enrichment strategy allowed the efficient identification of all APC populations including the rare pDC, AS-DC, and CLEC9A + DC, enabling further molecular and phenotypic characterization.

We used unsupervised pathway enrichment methods on total APC, in order to identify functional pathways associated with COVID-19-related inflammation. We detected 743 differentially expressed genes (DEG) between HC, moderate and severe COVID-19 patients, with an absolute fold change higher than 1.4 ( Fig. 2a ). Among DEG, 184 genes were up-regulated in HC, 158 in moderate, and 16 in severe, as compared to the two other groups, respectively ( Fig. 2b) . We performed pathway enrichment analysis on DEG using hallmark signatures compiled in MsigDB 23 , and focused on pathways enriched in each patient group, as compared to HC. "KRAS signaling", previously associated with down-regulation of IFN-induced anti-viral response 24 , and "cholesterol homeostasis" pathways, were enriched in moderate and severe patients ( Fig. 2c ). Both moderate and severe patients displayed enriched "Hypoxia", "Apoptosis", "Inflammatory response", and "TNF-α signaling" pathways ( Fig.   2c ). Differential expression of those pathways was retrieved in our validation set ( Supplementary Fig. 2 ). Pathway-level analyses were supported by the coordinated up-regulation of many genes involved in TNF-α and Hypoxia signaling ( Fig. 2d and e ) .

Next, we addressed the global contribution of APC to the expression of specific inflammatory cytokines and their receptors known to be involved in COVID-19 physiopathology. As compared to HC, the pro-inflammatory genes CXCL8 and IL1B increased in both severity groups, whereas CCL3 and CXCL2 were up-regulated in moderate patients ( Fig. 3a and   Supplementary Fig. 3a ). In parallel, we detected a decrease of TGFB1 and IL18, in both moderate and severe patients, together with a decrease in the expression of receptors for IL-6, IL-10 and TNF-α ( Fig. 3a and Supplementary Fig. 3a ). Next, we explored biological pathways associated to downstream cytokine signaling. In both moderate and severe patients, we detected higher scores for "p53", "inflammatory response", "TGF-β signaling", and "IL-6/JAK/STAT3 signaling" pathways, as compared to HC ( Fig. 3b ).

IFN family of cytokines is one of the most important mediators of anti-viral responses. Using previously shown UMAP coordinates, we represented the expression level of IFNL1, IFNL1R, IFNAR1, IFNAR2, IFNA1, IFNGR1 and IFNGR2, according to the severity and cell type ( Fig. 3c ). IFNL1 and IFNL1R were marginally detected in our discovery dataset ( Fig. 3c ), while absent from the validation dataset ( Supplementary Fig. 3b ), supporting that IFN-λ pathway was not significantly expressed. IFNAR1 and IFNAR2 were detected in almost all identified cell populations, with higher levels in pDC from HC, and CD14 + monocytes from both HC and moderate patients ( Fig. 3c and Supplementary Fig. 3b ).

Severe patients expressed much lower levels of both IFNAR1 and 2, suggesting a defect in IFN-α signaling ( Fig. 3c ). IFNGR1 and IFNGR2 were highly expressed by CD14 + monocytes and CD1c + DC, and lower in pDC from HC and moderate patients ( Fig. 3c ). Type I IFN, such as IFNA1, were not detected in any of the APC subsets and patient groups ( Fig.   3c and Supplementary Fig. 3b ).

To further investigate the involvement of IFN-stimulated genes (ISG), we quantified the expression level of six major ISG (MX2, ISG15, IRF7, BST2, IFITM2 and ADAR) 25 . Severe patients displayed low levels of all these genes, suggesting that the development of severe COVID-19 might be related to IFN dysregulation ( Fig. 3d and Supplementary Fig. 3c ). ISG have been separated into "anti-viral ISG" and "Regulators of IFN signaling" 25 . Both moderate and severe patients APC expressed lower levels of most genes in these two categories, suggesting a global perturbation of IFN downstream functions ( Fig. 3e ).

After having analyzed perturbations in COVID-19 APC at the global level, we sought to decipher alterations occurring in specific APC subsets. Sub-clustering of the pDC subset identified distinct pDC clusters for each severity group ( Fig. 4a ). DEG among pDC from the three patient groups revealed signatures associated to HC versus COVID-19 patients (Fig.   4b ) . Pathway enrichment analysis revealed an enrichment of "Hypoxia" and "TNF-α signaling via Nuclear Factor (NF) kB", in pDCs from moderate and severe, as compared to HC ( Fig. 4c ) . PDC from severe patients exhibited specific enrichment in a number of pathways, including "apoptosis", "p53 pathway", "IL6/JAK/STAT3 signaling", and "IL2/STAT5 signaling" ( Fig. 4c ). To accurately compare differentially enriched pathways among severity groups, we computed pathway-specific module scores using the gene lists corresponding to each hallmark signature. This confirmed that pDC from severe patients displayed significantly elevated scores for "Pro-apoptosis", "IFNG response", "Hypoxia", and "TNFa signaling" pathways, as compared to HC ( Fig. 4d ) . Overall, these analyses converge to establish a global deficiency in pDC through induced apoptosis, as well as a pro-inflammatory state related to cytokine-induced NFkB and JAK/STAT signaling.

We asked whether apoptosis and pro-inflammatory signaling signatures altered pDC effector functions. We defined four modules including immune effector molecules, using a literature-driven manual curation: immune cell attraction (hereafter "attraction") (18 genes), "innate sensing" (12 genes), "anti-viral effector molecules" (23 genes), and "cytotoxicity" (12 genes) ( Fig. 4e ). Each of these modules was crossed with the pDC expression matrix, and detected genes were depicted for each patient group based on average expression and percentage of positive cells ( Fig. 4e ). Only five genes were detected as part of the "attraction" module, with minor differences between groups, showing that this function is not significantly represented in HC, and COVID-19 pDC at this advanced-stage of the disease. On the contrary, many genes in the "innate sensing", "anti-viral effector molecules", and "cytotoxicity" modules were detected in the three groups, and followed the same pattern: baseline levels in HC, increased levels in moderate, and down-regulation in severe ( Fig. 4e and f ) ( Supplementary Fig. 4a ) . This was particularly striking for the viral sensors TLR7, DHX9, DHX36, and the cytotoxic molecule GZMB.

Last, we sought to identify potential mechanisms underlying the defective pDC functions in severe patients. Transcription factors controlling IFN-α production may be involved, among which IRF-7 20 . Although we found that IRF7 in moderate patients pDC was increased as compared to HC, expression was back to baseline in severe patients pDC ( Fig. 4g ) . We turned to IFNAR expression as an alternative hypothesis, since IFNAR signaling is key to drive downstream anti-viral IFN-induced molecules. We found that IFNAR1 and 2 were significantly decreased in pDC from severe patients, as compared to those from moderate patients and HC ( Fig. 4g ) ( Supplementary Fig. 4b ) .

Monocytes are important inflammatory cells and have been implicated in the physiopathology of severe sepsis and COVID-19. In order to get insight into monocyte diversity, we used UMAP to visualize total monocytes according to disease severity ( Fig. 5a, left plot ) , and subset diversity ( Fig. 5a, right plot ) . We observed that CD14 + and CD16 + monocytes were clearly distinct in HC, but more similar in patients ( Fig. 5a) . This suggested that COVID-19 might imprint common transcriptional programs to both subsets. To investigate alterations specifically encountered in CD14 + monocytes, we performed dimensionality reduction through Independent Component Analysis (ICA), and highlighted cells according to their severity group. We observed that IC1 clearly separated HC from COVID-19 monocytes, whereas IC3 distinguished severe CD14 + monocytes from the remaining cells ( Fig. 5b ).

Pathway enrichment analysis on the top50 genes contributing positively and negatively to IC1, followed by estimation of a score for each functional module, identified key pathways that segregated COVID-19 CD14 + monocytes from HC ( Fig. 5c ). Among them, we found decreased "apoptosis" and "p53 signaling", decreased "hypoxia", "TNFA signaling", and "IL6-JAK-STAT3 signaling », whereas « complement » and « coagulation » pathways were increased, suggesting a possible contribution to vascular events in the physiopathology of the disease ( Fig. 5c ). Similar analyses were done on IC3 top50 genes, and revealed lower "IFNA response" and "inflammatory response" in severe patients ( Fig. 5d ). Last, we asked whether decreased apoptosis in severe CD14 + monocytes would be associated to sustained or increased antiviral effector functions. We used our manually curated gene functional module, and estimated expression levels across patient groups ( Fig. 5e ). We observed a global decrease of almost all anti-viral effector molecules in severe patients as compared to HC ( Fig. 5e ) or moderate patients ( Supplementary Fig. 5a ). Overall, CD14 + monocytes showed defective anti-viral functions, while possibly contributing to vascular inflammation by triggering coagulation events.

In parallel, we sub-clustered CD16 + monocytes and reduced the data dimension using UMAP projection to depict the corresponding clusters for each severity group ( Fig. 5f ). Pathway enrichment analysis on top50 contributing genes to Umap2 dimension highlighted significantly lower enrichment scores corresponding to "TNFA signaling", "Hypoxia" and "Apoptosis" pathways in moderate and severe patients, versus HC ( Fig. 5g ). As a complementary approach, we performed differential expression analysis, and identified signatures specific to each of the CD16 + monocyte cluster ( Fig. 5h , top 10 DEG for each cluster ). These disease-associated changes in CD16 + monocytes paralleled the ones observed in CD14 + , suggesting common adaptation mechanisms.

Thanks to our APC-enrichment protocol, we could efficiently recover rare CLEC9A + DC and AS-DC subsets ( Fig. 6 ). Sub-clustering of CLEC9A + DC and AS-DC identified clusters specific to HC, moderate, and severe patients ( Fig. 6a and e ), although less clearly defined than pDC, and monocyte sub-clusters ( Fig. 4, 5 ). In addition, we identified a strong decrease in CLEC9A + DC proportions in moderate and severe patients ( Supplementary Fig. 1b ).

DEG among CLEC9A + DC sub-clusters included specific transcriptional signatures segregating moderate and severe patients from HC ( Fig. 6b ). Up-regulated genes in moderate and severe patients compared to HC showed enrichment in "IFNG" and "Oxidative phosphorylation" pathways ( Fig. 6c ) . Specifically, important genes in the IFNG pathway (HLA-DRB1, B2M, PSMB9) were significantly increased in moderate and severe CLEC9A + DC, versus HC ( Fig. 6d ).

In order to characterize molecular pathways in AS-DC sub-clusters, we used two complementary approaches. First, we inferred transcription factor (TF) activity using Dorothea algorithm 26 and scored the activity of each regulon using Viper inference tool 27 .

This identified a large number of highly variant TF activity scores, including a group with increased activity in severe patients ( Fig. 6f ). Interestingly, these included STAT4 and RUNX1, which are usually associated to Th1 responses, and may control pro-inflammatory genes in AS-DC. Second, we used our manually curated functional modules to infer functional defects in COVID-19 AS-DC ( Fig. 6g ). As for pDC, the "Attraction" module was not well represented, although two of its members (IL18, TNFSF13B) were decreased in severe patients ( Fig. 6g ). The most important changes were observed for "anti-viral effector molecules", which increased in moderate patients as compared to HC, and further decreased in severe ( Fig. 6g ), compatible with a compromised anti-viral immune control.

We next focused on disease-induced alterations in CD1c + DC. Sub-clustering of CD1c + DC identified well-defined clusters specific to HC, moderate and severe patients ( Fig. 7a ).

Computation of DEG across the three clusters revealed specific transcriptional signatures ( Fig. 7b ). This included a significant down-regulation of the MHC-II genes HLA-DPB1, HLA-DRB5, and HLA-DQA2, in CD1c + DCs from moderate and severe COVID-19 patients, as compared to HC ( Fig. 7b and c ) . HLA-DQA1 and DQB1 were not differentially expressed (not shown). Interestingly, HLA-DRB1 was decreased in severe patients as compared to moderate ( Fig. 7c and Supplementary Fig. 5b ).

In order to identify upstream mechanisms, we studied the differential estimated activity of TF ( Fig. 7d ), similar to the strategy used for AS-DC ( Fig. 6g ). Three different estimated activity patterns were obtained according to disease severity. First, we found TF activities that were down-regulated in CD1c + DC from moderate and severe patients including REL, recently shown to be a myeloid checkpoint 28 , FOS and JUN. Second, we found TF activities down-regulated in severe patients as compared to HC and moderate patients, including STAT2, 3 and 6, involved in immune responses by controlling transcription of cytokine-induced genes. Third, we identified TF activities up-regulated in severe patients as compared to HC and moderate patients, including E2F2 and TFDP1 that regulate cell cycle and proliferation ( Fig. 7d ) . Overall, CD1c + DC from severe COVID-19 patients were characterized by a defect in TF activity involved in cytokine-induced gene expression, and an increased expression of cell cycle regulators.

In order to explain the down-regulation of MHC class II genes, we focused on their upstream TF regulators. We show a significant down-regulation of RFX5 in CD1c + DC from severe patients ( Fig. 7e and Supplementary Fig. 5c ). We also observed a down-regulation of CIITA, and most pronounced in severe patients ( Fig. 7e and Supplementary Fig. 5c ), which may be due to the down-regulation of USF1 ( Fig. 7d) , interacting with STAT1 to activate IFN-γ -induced transcription of the CIITA gene 29 Fig. 6 ) . Using our "reference partner cell" methodology, we inferred potential communication between each of the APC subsets and CD4 T cells, in each of the disease groups ( Supplementary Fig. 6a ). Cell connectivity networks revealed a global decrease in APC-T cell communication in severe, as compared to moderate and healthy, predominantly in CD1c + DC and CD14 + monocytes ( Supplementary Fig. 6a and b ) . We then explored the various communication molecular families that may explain this decrease.

This revealed a dominant contribution of checkpoint molecules and cytokines for CD1c + DC-T cell communication, while cytokines were mostly underlying the decrease in CD14 + monocytes-T cell communication ( Supplementary Fig. 6b ). We could further identify ligand-receptor pairs within each of these molecular families, to explain the decreased CD1c + DC-T cell communication: MHC II-LAG3, CD80-CD28, and CD48-CD2 interactions ( Supplementary Fig. 6c ). Among cytokines, IL-10 and TGFB-mediated interactions were predominantly impacted in severe patients, which may contribute to immunopathology through excessive Th1 responses ( Supplementary Fig. 6c ).

Last, we asked whether functional pathway alterations observed across APC subtypes were sustained over time during the course of SARS-CoV-2 infection. We compared scRNAseq datasets generated at day 1 versus day 4 post-hospital admission ( Fig. 8 ). In a first approach, unsupervised differential expression of severe APC mean expression genes, between Day 1 and Day 4, revealed a low number of DEG, most of them being mitochondrial genes, not related to any specific functional module ( Fig. 8a ). We hence used an alternative hypothesis-driven approach, by selecting genes involved in previously identified altered functions, in severe COVID-19 APC, and systematically compared Day 1 and Day 4 expression levels, and percent cell expression ( Fig. 8b ). Most Day 1 defects were sustained at Day 4, in particular the low expression of anti-viral effector molecules in pDC and monocytes, as well as the increased apoptosis pathway in pDC ( Fig. 8b ). Interestingly, we noted an increase in two MHC-II molecules, HLA-DQA2 and DRB5, in CD1c + and CLEC9A + DC, between Days 1 and 4 ( Fig. 8b ) . This suggests a possible recovery of Ag presentation with severe COVID-19 evolution. 34 . Through dedicated enrichment steps performed immediately after blood sampling, we were able to capture sufficient cell numbers to define molecular profiles and identify specific defects in all known DC subsets.

As most immune cells, DC are not limited to a single function 16 . They play a key role in the first line of immune defense by sensing microbial pathogens, and also contribute to direct pathogen control, through the production of anti-microbial peptides 35 and anti-viral effector molecules 36 . Other effector functions include the secretion of pro-and anti-inflammatory cytokines, and cytotoxic molecules 16 . Last, they function as APC to T cells, with which they communicate through a large number of secreted and surface molecules expressed within the immune synapse 37 . By using scRNAseq, and a combination of supervised and unsupervised bioinformatics methods, we were able to uncover defects in almost all of these processes, in specific APC subsets, associated to COVID-19 severity. This provides the first detailed molecular map of DC subsets and underlying molecular pathways in COVID-19.

Several studies have shown an increase of inflammatory cytokines in severe COVID-19, which may contribute to the severity of the disease 31 . Increased circulating levels of IL-1 β and IL-6 were detected in severe patients 8, 31, 38 . However, the cellular source does not seem to be from circulating cells, but rather from inflammatory monocytes attracted to the lung 39 , as well as endothelial cells 10, 11 . Our study corroborates these findings for IL-6, with no significant expression detected across APC subsets. However, we did find increased 

This study was part of the Dendrisepsis project aimed at investigating the functional profiles of APC in patients with sepsis. The study was approved by the appropriate institutional Table 1 and   Supplementary Table 1 .

Blood samples (20ml) were collected from each patient at days 1 and 4 post-hospital admission, and from HC. PBMC were isolated by centrifugation on a density gradient 

Cell suspensions were subjected to in Gel Bead Emulsion using Chromium 10x Genomics controller according to manufacturer guidelines. To perform single cell RNA sequencing after cDNA amplification, the concentration of each sample was measured using Tapestation 2200 (Agilent). To prepare the cDNA libraries for 10x Genomics Chromium controller, we used the single-cell 3 ′ v3.1 kit. QC libraries were performed using Tapestation 2200 (Agilent).

Illumina Novaseq6000 (100cycles cartridge) with a sequencing depth of at least 50,000 reads per cell was used for sequencing. The input number of cells was estimated at 20,000 cells/ samples.

The raw scRNA-seq fastq files were processed using CellRanger 3.1.0 from 10X Genomics technology and aligned to Grch38 reference genome. Bam files and filtered expression matrices were generated using "cellranger_count". All expression matrices were loaded into R 4.0.0 using "Read10X" function from Seurat library ( https://github.com/satijalab/seurat ) version 3.1.5. The latter library was used to perform the analysis workflow.

Pre-processing steps were applied to remove genes expressed in less than 20 cells, and remove cells with fewer than 50 genes or displaying more than 50% mitochondrial transcripts.

To minimize technical confounding factors related to the sequencing steps, we evaluated the violin plot distribution of the number of Unique Molecular Identifiers (nUMI) , along with the total number of detected genes (nFeatures) per cell for all samples. two upper cutoffs of 6,000 and 50,000 were manually set for the nUMi and nFeatures respectively for each sample.

These quality-control metrics filtered-out low quality cells. Normalization to 10,000 reads, centering, and scaling were sequentially applied on the expression matrices to correct for the sequencing depth variability. To reduce computational time of samples integration, we filtered-out cells from other cell types than APC. Cell type annotation is detailed in (Part3).

To decipher specific alterations occurring in each specific APC subset, we separately sub-clustered each cell subtype, scaled the data and applied graph-based clustering to obtain cell clusters. Genes encoding for immunoglobulins were removed prior to performing the sub-clustering step for each cell type to get rid of ambient RNA.

To allow comparison across severity states, we integrated the whole expression matrices from all the samples using SCTransform algorithm. Integration anchors, retrieved from the first 20 principal components using the "FindIntegrationAnchors" Seurat function, were then used to integrate the datasets using "IntegrateData()" function. This crucial step added an "integrated" assay to the Seurat Object. Scaling and Principal Component Analysis (PCA) dimension reduction were performed on the integrated assay with 50 principal components. High resolution (resolution=0.8) graph-based clustering, and Uniform Manifold Approximation and Projection (UMAP) dimension reduction were conducted to retrieve and visualize cell clusters. ICA dimension reduction was specifically performed for CD14 + monocytes, using 30

dimensions.

Cells were manually annotated based on their expressing levels of their respective set of cell type markers, defined as "cell-type signatures". For each cell-type signature, enrichment scores were computed using "AddModuleScore()" function per cell with 100 randomly selected control genes, split on 25 bins. Each cell cluster was annotated with a particular cell type if its signature-score median value > 0. Cell-type signatures included: pDC: expression of ("TCF4", "CLEC4C", "IRF7", "IRF8", "LILRA4", "IL3RA", "TLR9", "SPIB"), cDC ("ANPEP", "CD1C", "ITGAX", "CST3", "FCER1A"), Monocytes("CD14", "FCGR1A"," S100A12", "FCGR3A", "MS4A7", "LYZ", "CXCR3"), AS-DC ("AXL", "SIGLEC6","CD22"), NK cells ("NCAM1", "FCGR3A", "GNLY", "XCL1", "XCL2", "NCR1", "NKG7"), T cells ( "CD3D", "CD3E", "CD3G"), B cells (CD19", "MS4A1", "CD79A", "CD79B"), Plasma cells ("IGHG2", "IGHG1", "IGLC2", "IGHA1", "IGHA2", "IGHA3", "JCHAIN", "IGHM", "XBP1", "MZB1", "CD38", "IGLL5"), Erythrocytes c("HBB", "HBA1"), Platelets (PPBP). Cells that were annotated as non-APC were discarded for each sample, before integration, to avoid high computational load during integration step.

For Monocytes and cDCs, a subsequent classification of cells was performed according to their expression levels of monocytes and cDC subset markers (CD14, FCGR3A for monocytes, CD1C and CLEC9A for cDCs).

Differential expression analysis between severity groups was performed using the "FindAllMarkers()"Seurat function, using Wilcoxon-ranked test and a cutoff was set to 

Pathway enrichment analysis was performed to seek for the perturbed or enriched pathways in severity groups, as compared to the HC. Human MsigDB hallmark signatures ( https://www.gsea-msigdb.org/gsea/msigdb/index.jsp ) were loaded into R session using the "msigdbr" library version 7.0.1, the category was set to "H" for "Human". Enrichment test was performed using the "enricher" function, from the "gprofiler" version 3.16.0. Msig

Database hallmark signatures were given as input to the "enricher" function. P-values were corrected using Benferroni correction method. Encoding genes for each enriched pathway were extracted and used as module to construct a "pathway-score" signature using "AddModuleScore" from Seurat library.

We sought to decipher variation of transcription factor (TF) activity between severity groups within particular cell types to avoid capturing differentially active TF related to lineage markers. Dorothea ( https://saezlab.github.io/dorothea/ ) ressource was used to infer TF activity. In this context of single cell level resolution, we constructed regulons based on the mRNA expression levels of each TF from a manually curated database, along with the expression level of its direct targets. We used Hence, TF activity is set to be a proxy of a transcriptional state of its targets. We created TF regulons using "dorothea_regulon_human" wrapper function from "dorothea" library version 0.99.10, and chose "A" and "B" high confidence TF selection. Viper scores were computed on the dorothea regulons, scaled and added as "Dorothea" slot on the integrated Seurat object. To allow comparison of TF score activities, mean and standard deviation values of the scaled viper scores were computed per severity group. Transcription factors were ranked according the variance of their corresponding viper scores. Top 50 highly variable score scores per severity groupe (n=150 TF in total) were kept for visualization of their corresponding scores.

To evaluate the dysregulations occuring at the functional level for each APC subset from COVID-19 patients, we established a manually curated list of effector genes involved in specific APC functions, including: "Attraction", "Anti-viral effector molecules", "Cytotoxicity". The signature construction relied on a thorough mining of existence literature, using a combination of MeSH terms and keywords on the PubMed search tool. Each selected molecule was considered as "effector" of the related function if there is at least one experimental proof in a human setting. Overall, we outlined 12 "Cytotoxicity" effector molecules, 29 "Anti-viral" effector molecules along with 18 "Attraction" effector molecules.

"Innate sensing" effectors included 12 genes (DDX58 ,DHX58, CGAS,IFI16,AIM2,IRF3,   TMEM173,NLRP3,PYCARD,TLR7,TLR9,DHX9,DHX36) , and were from 56, 57 . Both "regulators of Interferon signalling" and "Anti-viral ISG" were implemented through a literature mining of a pioneer article, from 25 .

Communication scores were generated using ICELLNET R package ( https://github.com/soumelis-lab/ICELLNET/maste r ). This package allows computation of cell-cell communication scores between cell subsets, given their corresponding transcriptomic profiles from the same of different datasets. Considering severity groups separately, only clusters including more than 15 cells were considered for the analysis. The average gene expression profiles of APC subset clusters were given as input to ICELLNET package, to compute communication scores between APC subsets and T lymphocytes for each severity group. As our datasets did not include T cells for the analysis, we used as reference the T lymphocyte transcriptomic profile from the Human Primary Cell Altas included in ICELLNET package (n=39). From ICELLNET ligand-receptor interaction database, we only selected the 144 interactions known to be involved in DC-T communication 58 . Barplot and dotplot representations were generated to compare the proportions of communication type scores (checkpoint, cytokines, chemokines) between severity groups.

We thank our patients for participating into this study. We also wanted to acknowledge Dr 

Author Correction: A new coronavirus associated with human respiratory disease in China

A Novel Coronavirus from Patients with Pneumonia in China

Clinical Characteristics of Coronavirus Disease 2019 in China

The laboratory tests and host immunity of COVID-19 patients with different severity of illness

Pathogenic human coronavirus infections: causes and consequences of cytokine storm and immunopathology

SARS-CoV-2: a storm is raging

A Dynamic Immune Response Shapes COVID-19 Progression

Impaired type I interferon activity and exacerbated inflammatory responses in severe Covid-19 patients

Clinical and immunological features of severe and moderate coronavirus disease 2019

Endothelial cells orchestrate COVID-19 coagulopathy

The vascular endothelium: the cornerstone of organ dysfunction in severe SARS-CoV-2 infection

Sars-CoV-2 Infection in a Neutropenic Pediatric Patient with Leukemia: Addressing the need for universal guidelines for treatment of Sars-CoV-2 positive, immunocompromised patients

Tocilizumab for treatment of mechanically ventilated patients with COVID-19

Immunophenotyping of COVID-19 and influenza highlights the role of type I interferons in development of severe COVID-19

Dendritic cells are the principal stimulators of the primary mixed leukocyte reaction in mice

Immunobiology of dendritic cells

Human dendritic cell subsets: an update

DC subsets in positive and negative regulation of immunity

Dendritic cell subsets and lineages, and their functions in innate and adaptive immunity

IPC: professional type 1 interferon-producing cells and plasmacytoid dendritic cell precursors

Type I interferons (alpha/beta) in immunity and autoimmunity

A single-cell atlas of the peripheral immune response in patients with severe COVID-19

The Molecular Signatures Database (MSigDB) hallmark gene set collection

Negative regulation of the alpha interferon-induced antiviral response by the Ras/Raf/MEK pathway

Interferon-stimulated genes: a complex web of host defenses

Robustness and applicability of transcription factor and pathway analysis tools on single-cell RNA-seq data

Functional characterization of somatic mutations in cancer using network-based inference of protein activity

c-Rel is a myeloid checkpoint for cancer immunotherapy

Activation of the MHC Class II Transactivator CIITA by Interferon-γ Requires Cooperative Interaction between Stat1 and USF-1

ICELLNET: a transcriptome-based framework to dissect intercellular communication

Immunology of COVID-19: Current State of the Science

Sepsis-induced immune dysfunction: can immune therapies reduce mortality?

Immunosuppression in sepsis: a novel understanding of the disorder and a new therapeutic approach

Differential Redistribution of Activated Monocyte and Dendritic Cell Subsets to the Lung Associates with Severity of

Early cellular responses to Salmonella infection: dendritic cells, monocytes, and more

The role of dendritic cells in immunity against primary herpes simplex virus infections

Antigen presentation and T cell stimulation by dendritic cells

Complex Immune Dysregulation in COVID-19

Patients with Severe Respiratory Failure

Aberrant pathogenic GM-CSF T cells and inflammatory CD14 CD16 monocytes in severe pulmonary syndrome patients of a new coronavirus

Type III interferons: Balancing tissue tolerance and resistance to pathogen invasion

Production of interferons by dendritic cells, plasmacytoid cells, natural killer cells, and interferon-producing killer dendritic cells

SARS-CoV-2 induces activation and diversification of human plasmacytoid pre-dendritic cells

Plasmacytoid dendritic cell precursors/type I interferon-producing cells sense viral infection by Toll-like receptor (TLR) 7 and TLR9

DDX1, DDX21, and DHX36 Helicases Form a Complex with the Adaptor Molecule TRIF to Sense dsRNA in Dendritic Cells

STAT2 Is Required for TLR-Induced Murine Dendritic Cell Activation and Cross-Presentation

Zika Virus Antagonizes Type I Interferon Responses during Infection of Human Dendritic Cells

Regulation of MHC class II genes: lessons from a disease

Prostaglandin E2 Leads to the Acquisition of DNMT3A-Dependent Tolerogenic Functions in Human Myeloid-Derived Suppressor Cells

CD8 T cell responses to lymphocytic choriomeningitis virus in early growth response gene 1-deficient mice

Immunosuppression in patients who die of sepsis and multiple organ failure

Profound and persistent decrease of circulating dendritic cells is associated with ICU-acquired infection in patients with septic shock

Low monocyte human leukocyte antigen-DR is independently associated with nosocomial infections after septic shock

Association between mRNA expression of CD74 and IL10 and risk of ICU-acquired infections: a multicenter cohort study

Dendritic cells as cancer therapeutics

Turbocharging vaccines: emerging adjuvants for dendritic cell based therapeutic cancer vaccines

DHX9 pairs with IPS-1 to sense double-stranded RNA in myeloid dendritic cells

The DNA Inflammasome in Human Myeloid Cells Is Initiated by a STING-Cell Death Program Upstream of NLRP3

A Quantitative Multivariate Model of Human Dendritic Cell-T Helper Cell Communication

TNFA signaling Hypoxia IFNG response **** **** **** **** **** **** **** **** P53 pathway IL6-JAK-STAT3 signaling Coagulation Complement **** **** **** **** **** **** **** **** Severe   ZC3HAV1  PML  MX2  ISG20  IRF7  IRF1  IFIT5  IFIT3  IFIT2  IFIT1  IFIH1  HPSE  GBP2  GBP1  DDX60  DDX58  RSAD2  BST2  ADAR  ISG15  TRIM5  APOBEC3G  MX1  OAS1  EIF2AK2  IFITM2  IFITM1 MYC  ZNF263  RXRA  USF1  E2F4  USF2  CEBPD  NFKB1  STAT5B  GATA1  STAT2  SP3  FOXP1  STAT3  IRF1  ATF1  NFYA  EPAS1  PRDM14  NFE2L2  GATA3  JUN  YY1  BACH1  RELB  TCF7L2  TP53  SOX2  STAT4  FOXL2  FOXM1  PAX8  CEBPB  RUNX1  TP63  GLI2  HNF4A  ATF4  RARA  CEBPA  IRF9  SREBF1  SMAD3  FOSL1  SRF  ATF2  TWIST1  RELA GZMB  TNFSF10  ITGB2  MX2  ISG20  IRF7  IFIT2  BST2  ISG15  MX1  OAS1  IFITM2  TRIM5  ADAR  CGAS  DHX36  DHX9  PYCARD   TMEM173   TLR7  TLR9  CCL20  CCL5  CCL3  IL18