key: cord-0815065-m9cyhmrl
authors: Reyes, Miguel; Filbin, Michael R.; Bhattacharyya, Roby P.; Sonny, Abraham; Mehta, Arnav; Billman, Kianna; Kays, Kyle R.; Pinilla-Vera, Mayra; Benson, Maura E.; Cosimi, Lisa A.; Hung, Deborah T.; Levy, Bruce D.; Villani, Alexandra-Chloe; Sade-Feldman, Moshe; Baron, Rebecca M.; Goldberg, Marcia B.; Blainey, Paul C.; Hacohen, Nir
title: Induction of a regulatory myeloid program in bacterial sepsis and severe COVID-19
date: 2020-09-02
journal: bioRxiv
DOI: 10.1101/2020.09.02.280180
sha: 6e741573aba6865578eb23a85c2b84f9979f80e7
doc_id: 815065
cord_uid: m9cyhmrl

A recent estimate suggests that one in five deaths globally are associated with sepsis1. To date, no targeted treatment is available for this syndrome, likely due to substantial patient heterogeneity2,3 and our lack of insight into sepsis immunopathology4. These issues are highlighted by the current COVID-19 pandemic, wherein many clinical manifestations of severe SARS-CoV-2 infection parallel bacterial sepsis5–8. We previously reported an expanded CD14+ monocyte state, MS1, in patients with bacterial sepsis or non-infectious critical illness, and validated its expansion in sepsis across thousands of patients using public transcriptomic data9. Despite its marked expansion in the circulation of bacterial sepsis patients, its relevance to viral sepsis and association with disease outcomes have not been examined. In addition, the ontogeny and function of this monocyte state remain poorly characterized. Using public transcriptomic data, we show that the expression of the MS1 program is associated with sepsis mortality and is up-regulated in monocytes from patients with severe COVID-19. We found that blood plasma from bacterial sepsis or COVID-19 patients with severe disease induces emergency myelopoiesis and expression of the MS1 program, which are dependent on the cytokines IL-6 and IL-10. Finally, we demonstrate that MS1 cells are broadly immunosuppressive, similar to monocytic myeloid-derived suppressor cells (MDSCs), and have decreased responsiveness to stimulation. Our findings highlight the utility of regulatory myeloid cells in sepsis prognosis, and the role of systemic cytokines in inducing emergency myelopoiesis during severe bacterial and SARS-CoV-2 infections.

Sepsis is associated with profound alterations in the peripheral immune compartment, including marked reduction in lymphocyte counts [10] [11] [12] and phenotypic alteration of myeloid cells [13] [14] [15] . Monocytes from sepsis patients have decreased responsiveness to stimuli [15] [16] [17] and have lower expression of HLA-DR [18] [19] [20] [21] [22] [23] characteristic of monocytic MDSCs. In line with these findings, we recently reported an expanded CD14+ monocyte state in sepsis patients, MS1, reminiscent of monocytic MDSCs 9 . Previous studies have reported opposing effects of MDSCs on sepsis outcomes [24] [25] [26] [27] , warranting further investigation into their function and association with patient prognosis. Signatures similar to MS1 have also been described recently in severe SARS-CoV-2 infections 28, 29 , but have not been systematically analyzed across multiple cohorts. Here, we identify the gene expression programs associated with the MS1 cell state and determine their relationship with disease severity in both bacterial sepsis and COVID-19. We identify circulating host factors that result in the induction of the MS1 gene module, and characterize the response of MS1 cells to stimulation and their effects on other cell types.

To characterize the gene expression program associated with the MS1 cell state, we analyzed existing monocyte scRNA-seq data from sepsis patients and controls 9 using consensus non-negative matrix factorization 30 (cNMF; Figure 1a , Supplementary Fig. 1a-d ) . We found a gene expression program that includes the MS1 marker genes RETN , ALOX5AP , and IL1R2 among the top 20 genes with highest loadings ( Supplementary Fig. 1d ), has higher usage in sepsis patients ( Supplementary Fig. 1e , FDR < 0.001), and correlates with the fractional abundance of MS1 cells in each patient ( Supplementary Fig. 1f , r = 0.73, p < 0.001). This module negatively correlates with the usage of a class II major histocompatibility complex module (MHC-II module, Supplementary Fig. 1g ), consistent with our observation that MS1 cells have lower surface expression of HLA-DR 9 . Co-expression analysis within the MS1 module reveals that most genes are highly correlated with S100A8 ( Figure 1b ), suggesting its role as a core driver of the MS1 gene expression program. This gene and its partner S100A9 have been implicated in the development of MDSCs in cancer 31 and sepsis 24, 32 , indicating a similarity between monocytic MDSCs and the MS1 cell state.

Given its marked expansion, we hypothesized that expression of the MS1 program may be associated with worse outcome in bacterial sepsis. Using the transcriptional data from our earlier study 9 , the usage of the MS1 program correlates with sepsis severity in ICU patients but not in patients presenting to the emergency department with milder disease ( Supplementary Fig. 1h ) . Using gene expression deconvolution, we estimated the usage of the monocyte gene expression modules in 15 datasets included in a recent meta-analysis examining sepsis mortality 33 ( Methods , Supplementary Figure 1c, Supplementary Table 1 ). We found that expression of the MS1 program is negatively associated with patient survival (effect size = -0.32, FDR < 0.01), whereas MHC-II module usage has the opposite effect (effect size = 0.53, FDR < 0.01; Figure  1c ). These results show that expansion of MS1 cells is associated with worse infection outcomes, suggesting its prognostic value for sepsis patients.

To determine whether MS1 cells are similarly expanded in severe COVID-19, we analyzed four COVID-19 scRNA-seq datasets 7, 28, 34, 35 independently and identified gene expression modules in CD14+ monocytes from each dataset using an unbiased cNMF method ( Figure 1a, Supplementary Figure 2 ). In each of the four studies, we found gene expression modules corresponding to the MS1 and MHC-II modules, as evidenced by their strong correlation (Pearson r > 0.8) with modules from our sepsis data ( Figure 1d ). Similar to the trends we observed in bacterial sepsis, CD14+ monocytes from patients with severe SARS-CoV-2 or influenza A infections have significantly higher and lower usage of the MS1 and MHC-II modules, respectively (FDR < 0.05; Figure 1e ), and have increased MS1 scores compared with controls (p < 0.01; Supplementary Figure   2 ). These findings show that the MS1 cell state is expanded in both bacterial and viral sepsis syndromes, suggesting its utility as a signature for infection severity regardless of etiology.

We previously demonstrated that MS1-like cells can be derived from immature progenitors through stimulation of total bone marrow mononuclear cells (BMMCs) with LPS or Pam3CSK4 9 . Because BMMCs contain a heterogeneous mix of hematopoietic stem and progenitor cells (HSPCs) and mature immune cells, potential paracrine interactions among the cell types confound identification of the precise factors that cause MS1 induction. Given this limitation, we sought a method for inducing the MS1 program directly from CD34+ HSPCs purified from bone marrow. We hypothesized that cytokines circulating in the blood of sepsis patients might induce the differentiation of MS1 cells directly from HSPCs. Upon culturing HSPCs isolated from healthy human bone marrow with plasma from urosepsis (URO) patients or healthy controls (Control) ( Figure 2a ), we found that sepsis plasma significantly stimulated the production of monocytes and neutrophils compared with healthy plasma ( Figure 2b ; p = 0.025 and 0.004 for CD34-CD11b+CD14+ and CD34-CD11b+CD15+ cells, respectively). Single cell analysis of the differentiated cell populations showed clear trajectories of myeloid differentiation ( Figure 2d, Supplementary Fig. 3a-b, Methods ) ; importantly, we observed that incubation of HSPCs in plasma from URO patients resulted in the emergence of CD14+ cells with high MS1 scores compared with Control ( Figure 2d ; p < 0.01). cNMF analysis of the scRNA-seq data identified modules similar to the MS1 and MHC-II modules derived from patient PBMC data, as evidenced by strong correlations between their gene loadings (Pearson r = 0.73 and 0.78, respectively; Figure 2f, Supplementary Fig. 3c ). The MS1 and MHC-II modules are also significantly up-or down-regulated (p < 0.01), respectively, in CD14+ cells derived from HSPCs incubated with sepsis plasma ( Figure 2f ). These data support our hypothesis that cytokines circulating in the blood of sepsis patients can induce the differentiation of MS1 cells.

Due to the expansion of MS1 in severe COVID-19, we hypothesized that similar effects may be observed when incubating HSPCs with plasma from severe COVID-19 patients. We performed the same experiments with heat-inactivated plasma from SARS-CoV-2 infected patients with varied disease severity (C1-C4) and uninfected controls (CN; Methods, Supplementary Fig. 3e-f, Supplementary Table 2 ). Indeed, plasma from patients who eventually died of COVID-19 (C4) stimulated the production of monocytes more strongly than non-hospitalized COVID-19 patients (C1, FDR = 0.02; Figure 2c ). Compared with C1 plasma, C4 plasma induced the production of CD14+ cells with higher MS1 scores (FDR < 0.01; Figure 2e ) and caused increased and decreased usage of the MS1 and MHC-II modules, respectively (p < 0.01; Supplementary Fig. 3g-h ) . These findings further highlight the similarities between bacterial sepsis and severe COVID-19, and support our hypothesis that circulating factors from patients with severe disease stimulate the induction of MS1 cells.

Interestingly, upon incubation of HSPCs with URO plasma, we observed cells with high MS1 scores and a module of genes similar to MS1 in the emerging neutrophil population ( Figure 2d, Supplementary Fig. 3c ). Moreover, neutrophils sorted from critically-ill patients (ICU) and bacteremic (BAC) patients express a module which includes S100A8 among the genes with highest loadings ( Supplementary Fig. 4a -c ) and correlates with the MS1-like module in neutrophils generated from incubating HSPCs with URO plasma (Pearson r = 0.58, Supplementary Fig. 4d ). MS1 marker genes are also among the top differentially expressed genes in neutrophils between healthy control subjects and ICU or BAC patients ( Supplementary Fig. 4e ). These findings are consistent with previous reports of MDSCs having both granulocytic and monocytic subtypes 36 , and further highlight the similarity between the MS1 cell state and MDSCs.

Analysis of the monocyte differentiation trajectory shows different pseudo-temporal dynamics of the MS1 genes ( Figure 2g ). Of note, a number of genes, including S100A8, VCAN, and MNDA , are expressed early and remain up-regulated during the differentiation trajectory ( Figure 2h ) . Short term stimulation (24 h) of CD34+ HSPCs with sepsis plasma results in the up-regulation of these genes ( Figure 2i , Supplementary Table 3 ) and an increase in the fraction of S100A8+ cells in the CD34+ population ( Figure 2j ). These genes are among the core genes in the MS1 module ( Figure 1b ) , and their up-regulation early is in line with our hypothesis that S100A8 is an important factor driving the induction and differentiation of MS1 cells from HSPCs.

To determine which circulating factors induce the increased production of myeloid cells from HSPCs, we analyzed the levels of inflammatory cytokines implicated in cytokine storms 37 in the plasma of COVID-19 patients ( Supplementary Figure 5 ). We found that IL-6 strongly correlated with higher production of CD14+ cells from HSPCs ( Figure 3a ), suggesting its involvement in the induction of emergency myelopoiesis. To test this hypothesis, we added plasma from urosepsis or COVID-19 patients, or corresponding controls, to IL6R-knockout or control CD34+ HSPCs and observed a marked reduction in CD14+ cell production across all conditions for IL6R-knockout CD34+ HSPCs ( Figure 3b ). We also observed a reduction, albeit weaker, specifically for URO and C4 plasma when wild-type HSPCs are differentiated in the presence of IL-6 neutralizing antibodies ( Figure 3c ). These findings highlight the role of circulating IL-6 in inducing emergency myelopoiesis during severe infections.

To determine the specific cytokines that induce the differentiation of MS1 cells from HSPCs, we measured the levels of inflammatory cytokines in the plasma of sepsis patients and controls from a previous study 9 . Whereas several cytokines displayed increased concentration in the plasma of sepsis patients ( Supplementary Fig. 6 ), we found that both IL-6 and IL-10 levels specifically correlated with the usage of the MS1 module in monocytes ( Figure 3d ), suggesting their specific involvement in its induction, and consistent with the role of both cytokines in MDSC expansion in cancer 38, 39 . We found that short-term incubation of HSPCs with URO plasma, but not Control plasma, resulted in phosphorylation of STAT3 ( Figure 3e ), the transcription factor downstream of both cytokines 40, 41 . In addition, short term treatment of HSPCs with recombinant IL-6 and/or IL-10 show a dose-dependent up-regulation of the early MS1 genes S100A8 , VCAN , and MNDA ( Supplementary Fig. 7a ) and an increase in S100A8+ progenitor cells ( Supplementary Fig 7b ) , similar to the effects observed with sepsis plasma. To further test this dependence, we used CRISPR-Cas9 to knock out the surface receptors for IL-6 ( IL6RA and IL6ST ) and IL-10 ( IL10RA and IL10RB ) in CD34+ HSPCs before incubation in URO plasma ( Methods, Supplementary Fig. 7c-d ) . Knockout of either receptor resulted in the production of CD14+ cells with lower MS1 scores ( Figure 3g ) and reduced expression of several, but not all, MS1 genes ( Figure 3h ). In addition, in these experiments, the CD14+ cells displayed increased surface expression of HLA-DR, consistent with the absence of the MS1 phenotype ( Supplementary Fig. 7e ) . Altogether , these results demonstrate that the cytokines IL-6 and IL-10 are necessary for the induction of the MS1 program from HSPCs with sepsis plasma.

To test whether these cytokines are sufficient to induce the MS1 program, we differentiated CD34+ HSPCs with IL-6, IL-10, or both, with and without GM-CSF and M-CSF, which are known growth factors that support the differentiation of HSPCs into monocytes 42 ( Supplementary Fig 8 ) . We found that addition of GM-CSF and M-CSF, and to a lesser extent, IL-6 or IL-10, increased the fraction of CD14+ cells produced by HSPCs, but IL-10 decreased the absolute number of cells that are produced ( Supplementary Fig 8g ) . Importantly, we found increased expression of several MS1 genes in CD14+ cells generated from HSPCs treated with IL-6 and IL-10 in the presence of GM-CSF and M-CSF ( Figure 3h, Supplementary Fig 8b ) . cNMF analysis reveals modules resembling the MS1 and MHC-II modules, both of which correlate significantly with the modules derived from patient PBMC data (Pearson r = 0.57 and 0.66, p < 0.01; Figure 3i , Supplementary Fig 8e-f ). The usage of these modules have opposite trends with cytokine treatment ( Figure 3j ), consistent with their anti-correlation in the PBMC dataset ( Supplementary Fig. 1g ). These results suggest that MS1-like cells can be induced by incubation of HSPCs with the cytokines IL-6 and IL-10 in the presence of GM-CSF and M-CSF, enabling generation of large numbers of these cells for functional studies.

Given their similarity to MDSCs, we hypothesized that MS1 cells suppress the activation of T cells 43 . We generated monocytes from CD34+ HSPCs either with GM-CSF and M-CSF alone (iMono), or with IL-6 and IL-10 (iMS1), and co-incubated the cells with PBMCs activated with anti-CD3 and anti-CD28. We found that incubation with iMS1 cells suppressed the proliferation of T cells, as evidenced by a reduction in the number of cell divisions undergone by CD4 T cells after 4 days ( Figure 4a ). The suppression of T cell proliferation by iMS1 also depended on the ratio of monocytes added to culture ( Figure 4b ) . Consistent with this phenomenon, we found that usage of the MS1 module in monocytes correlated negatively with the fraction of CD4+ T cells among total PBMCs in sepsis patients and controls (Pearson r = -0.59, p < 0.01; Figure 4c ). A similar suppressive effect was observed in vitro for CD8 T cells, but CD8 T cell levels were not correlated with MS1 module usage ( Supplementary Fig. 9a-c ) . These results are consistent with previous studies showing that MDSCs in sepsis patients have immunosuppressive effects on T cells 22, 44 .

Sepsis is a systemic disease; thus, we tested the effects of co-incubating iMS1 cells with cell types from other organs that are commonly dysfunctional in sepsis 45 Table  6 ). Co-incubation with iMS1 also resulted in decreased expression of IL6 , CSF1, and CSF2 in HREs ( Figure  4g ), suggesting a systemic negative feedback loop wherein MS1 suppresses its own induction. These effects were not observed in HUVECs or in HREs incubated in conditioned media from iMS1 or iMono cells, suggesting a cell-contact dependent effect ( Supplementary Fig. 9f ). Of note, co-incubation of HREs with iMS1 cells resulted in the up-regulation of MMP1 , an important regulator of tissue remodeling and extracellular matrix homeostasis that has been previously shown to be up-regulated during the development of sepsis 46 . Similarly, PROS1 , the gene encoding protein S, an important regulator of the clotting cascade, was also upregulated in HREs co-incubated with iMS1, suggesting a possible involvement of MS1 cells in sepsis-related coagulopathy 47, 48 . These findings demonstrate broad anti-inflammatory effects of MS1 cells and, importantly, highlight other potential functions of monocytes in sepsis beyond their response to pathogens and interaction with other immune cells.

To examine the potential role of MS1 cells in COVID-19, we generated CD14+ cells from HSPCs using plasma from mild (C1) or severe (C4) COVID-19 patients. As expected, the core MS1 gene S100A8 was the most significantly up-regulated gene in CD14+ cells generated with severe COVID-19 plasma ( Figure 4h , Supplementary Table 5 ). Among the positively up-regulated genes, we found enriched pathways related to inflammation and coagulation ( Figure 4i ), both of which contribute to the pathogenesis of severe COVID-19 49, 50 . Next, we stimulated C1-and C4-treated populations with high molecular weight (HMW) poly-(I:C), a synthetic RNA analogue, or IFN-β, an important mediator of antiviral responses, and observed weaker induction of various cytokines and interferon-stimulated genes in cells generated with C4 plasma in response to poly-(I:C) ( Figure 4j ), whereas no such effect was observed in response to IFN-β. These results reflect our findings in bacterial sepsis 9 , wherein MS1 cells showed a diminished response to pathogen-associated molecular patterns, and are consistent with a recent report demonstrating functional impairment of myeloid cells in COVID-19 patients 7 . Altogether, these findings suggest that the expansion of MS1 cells in severe SARS-CoV-2 infection is detrimental to the antiviral response and may play a role in the pathogenesis of severe COVID-19.

In this study, we show that expression of the MS1 program is associated with poor outcome in both bacterial and SARS-CoV-2 infections. In addition, we show that systemic cytokines induce emergency myelopoiesis and differentiation of HSPCs into myeloid cells that express the MS1 program, providing a potential explanation for their expansion in bacterial sepsis and severe COVID-19, and a model system through which these cells can be generated and studied in greater detail. Importantly, our study reveals that induction of myelopoiesis and the MS1 program depends on systemic IL-6, a current therapeutic candidate in COVID-19 infection 51 . Our results suggest that the role of IL-6 in infection is complex and more extensive than just the induction of an acute phase response 52 . Given the similarity between MS1 and monocytic MDSCs, we propose that the MS1 gene expression program provides a precise definition for monocytic MDSCs in peripheral blood. In fact,~50% genes expressed specifically in monocytic MDSCs vs monocytes, as found in a recent study 5 , overlap with MS1-specific marker genes ( Supplementary Fig. 10 ), though our MS1 signature contains many additional genes, suggesting that we have identified a more pure MDSC-like population; however, analysis of additional contexts in which MDSCs are found is warranted 43, [54] [55] [56] . We also demonstrate that MS1 cells have an anti-inflammatory effect on endothelial and epithelial cells, raising the possibility that MS1 contributes to the pathology of these cell types during bacterial sepsis and severe COVID-19. Whereas these findings suggest several potential roles for MS1 during infections, whether it plays a causal role in the pathogenesis of bacterial sepsis or severe COVID-19 must still be tested. Our study highlights the importance of hematopoietic reprogramming in bacterial sepsis and severe COVID-19 and supports the hypothesis that the interaction of MS1 cells with other cells and tissues impacts pathogenesis of severe infections. Cas9-RNPs and treated with sepsis plasma for 7 days. (g) Expression of the top 30 MS1 module genes in sorted CD14+ cells generated from CD34+ HSPCs electroporated with Cas9-RNPs and treated with sepsis plasma for 7 days. Asterisks indicate a significant difference (FDR < 0.1, Wilcoxon rank-sum test, corrected for testing of multiple genes) compared with the non-targeting condition (NTA). In (f,g), n = 4 experiments were performed for each condition (2 biological and 2 technical replicates). (h) Expression from scRNA-seq of the top 30 MS1 module genes in CD14-expressing cells generated from CD34+ HSPCs treated with the indicated cytokines (all at 100 ng/mL). Asterisks indicate a significant difference (FDR < 0.1, Wilcoxon rank-sum test, corrected for testing of multiple genes) compared with the non-treated condition (NT). (i) Gene weight correlation between the MS1 modules detected in the cytokine treatment (x-axis) and patient PBMC datasets . Line and shadow indicate linear regression fit and 95% confidence interval, respectively. Significance of the correlation (Pearson r ) was calculated with a two-sided permutation test. (d) Experimental schematic of the co-incubation of iMS1 or iMono with primary human umbilical vein endothelial cells (HUVECs) or human renal epithelial cells (HREs) . (e-f) Volcano plot showing differential expression analysis results (exact test) between TNF α-activated HUVECs (e) or HREs (f) co-incubated with either iMono or iMS1 cells generated from CD34+ HSPCs. Genes with FDR < 0.1 are highlighted in red, and the top 10 genes with lowest FDR values are shown. (g) Expression of MS1-inducing cytokines in TNF α-activated HREs co-incubated with either iMono or iMS1 cells generated from CD34+ HSPCs. P values are calculated with an exact test. In (e-g), n = 8 experiments were performed for each condition (2 bone marrow donors with 2 biological and 2 technical replicates). (h) Volcano plot showing differential expression analysis results (exact test) between CD14+ cells generated from HSPCs using pooled plasma from C1 or C4 patients. Genes with FDR < 0.1 are highlighted in red, and the top 15 genes with lowest FDR values are shown. (i) Dotplot showing enrichment of pathways (KEGG database) for upregulated genes in CD14+ cells generated with C4 plasma (FDR < 0.1, exact test). (j-k) Scatterplots showing the log2 fold-change of each gene after high molecular weight (HMW) poly-I:C (j) or IFN-β (k) treatment of CD14+ cells generated with pooled plasma from C1 (x-axis) or C4 (y-axis) patients. Genes with FDR < 0.1 in C1 plasma-generated cells are highlighted in red, and the top 10 genes with highest fold-change values are shown. In (h-k), n = 4 experiments were performed for each condition (2 bone marrow donors with 2 technical replicates).

Purified CD34+ bone marrow cells from healthy individuals were either purchased directly from StemCell Technologies or isolated from fresh human bone marrow. Bone marrow aspirates anticoagulated with EDTA were purchased from StemExpress and processed within 24 h of isolation. To remove red blood cells (RBC) from the bone marrow, 1X RBC lysis buffer (eBioscience) was added directly at a 10:1 ratio to the sample. After 5 min, the cells were centrifuged at 400g for 5 min and resuspended in 1X RBC lysis buffer to further clear the sample of RBCs. The cells were then centrifuged, resuspended in FACS buffer (1X PBS, 2.5% FBS, 2 mM EDTA, Invitrogen), and purified using human CD34 Microbeads (Miltenyi Biotec). Isolated CD34+ cells were validated using flow cytometry (CD34-BV650, clone 561; BioLegend) to be of >90% purity. Cells were cryopreserved in Cryostor CS10 (Stemcell Technologies) in aliquots of 200,000 cells each. The tubes were kept at −80 °C overnight, then transferred to liquid nitrogen for long-term storage.

For each experiment, CD34+ cells were first thawed and rested for 48 h in SFEM II (StemCell Technologies) with 75 nM StemRegenin 1 (StemCell Technologies), 3.5 nM UM171 (StemCell Technologies), 40 ng/mL SCF, TPO, Flt3L (PeproTech), and 1X penicillin-streptomycin (Gibco). Cells were subsequently cultured in the same medium supplemented with either 20% plasma sterilized through a 0.2 μm filter (Millipore) or 100 ng/mL of the cytokines IL-6, IL-10, GM-CSF, and M-CSF (PeproTech). iMS1 and iMono cells were isolated after 7 d of culture using human CD14 Microbeads (Miltenyi Biotec). Isolated CD14+ cells were validated using flow cytometry (CD14-FITC, clone M5E2; BioLegend) to be of >90% purity.

To assess the number of myeloid cells from plasma incubation experiments, cells were stained with the following panel: CD3-APC (clone HIT3a), CD19-APC (clone HIB19), CD56-APC (clone 5.1H11), CD14-FITC (clone M5E2), CD15-AF700 (clone HI98), CD11b-PE-Cy7 (clone ICRF44), CD34-BV650 (clone 561), and CD38-PE/Cy5 (clone HIT2) (BioLegend). After staining, cells were resuspended in FACS buffer with 2% CountBright beads (Invitrogen) to allow determination of absolute counts during analysis. Flow cytometry data were acquired on a Cytoflex LX (Beckman Coulter) and analyzed using FlowJo v10.1.

Single cell RNA-seq combined with cell hashing 57 was performed as previously described 9 . Briefly, cells from multiple culture conditions were labeled with HTO antibodies (BioLegend) and loaded on the Chromium platform using the 3' v3 profiling chemistry (10X Genomics). Libraries were sequenced to a depth of~25,000 reads per cell on a Nextseq 550 (Illumina). The data were aligned to the GRCh38 reference genome using cellranger v3.1 (10X Genomics). Due to their technical incompatibility with droplet-based platforms, scRNA-seq of neutrophils was performed using plate-based Smart-seq2 as previously described 58 .

Single cell data analysis was performed using scanpy 59 with the same pre-processing and filtering parameters described in a prior publication 9 . To identify the major cell types in the differentiation experiments, we assessed the expression of the following marker genes: HSPCs: CD34 and CD38 , monocytes: CD14 and LYZ , neutrophils: ELANE and MPO , megakaryocytes: PPBP and PF4 . MS1 scores were calculated for the top 30 genes from the MS1 module derived from the sepsis dataset using the 'score_genes' in scanpy (ctrl_size = 50, n_bins = 25). RNA velocity analysis was performed using the scVelo package 60 using the default parameters.

Consensus non-negative matrix factorization analysis was performed as detailed in a previous publication 30 . Briefly, the top 3,000 variable genes for each dataset was first selected to filter the gene expression matrix. NMF was then performed with k=5 to 25 (10 iterations for each k). The number of modules (k) for downstream analysis was selected based on biological interpretability of the modules and stability of the cNMF solution. To ensure that no modules from technical artifacts were analyzed, only gene programs with mean usage >10 across all cells were included for further analysis.

The gene loading matrix from cNMF analysis of monocytes was used to construct a reference matrix for gene expression deconvolution. To reduce the number of genes in the reference matrix, only the top 1,000 variable genes within the monocyte data were included. Sepsis datasets with survival annotation were obtained from a previously published meta-analysis 33 . Gene expression deconvolution was performed using CIBERSORT 61 with a no-sum-to-one constraint and absolute scoring. The resulting score matrix was then used as an input to MetaIntegrator 62 , where the effect size of each gene module was visualized using forest plots.

Plasma samples from sepsis patients and controls were obtained from an existing cohort of patients described in a prior publication 9 . Plasma samples were isolated by obtaining the top layer from Ficoll gradient separation of whole blood (diluted 1:1 with 1X PBS) and were centrifuged again at 1000g for 10 min to remove cell debris. Samples were immediately stored at −80 °C.

COVID-19 plasma samples were obtained from patients entering the Massachusetts General Hospital Emergency department. This study was approved by the Partners Healthcare Institutional Review Board under protocol 2017P001681. Patients presenting to the MGH ED from March through May 2020 with respiratory distress suspected or known to be due to COVID-19 were enrolled. Inclusion criteria were age 18 years or older, clinical concern for COVID-19 upon Emergency Department presentation, and acute respiratory distress with at least one of the following: 1) tachypnea ≥ 22 breaths per minute, 2) oxygen saturation ≤ 92% on room air, 3) a requirement for supplemental oxygen, or 4) positive-pressure ventilation. Patients were categorized based on disease outcomes as follows: C1: non-hospitalized, C2: hospitalized without intensive care, C3: hospitalized and admitted to intensive care, C4: hospitalized and eventually died, and CN: SARS-CoV-2 negative patients. Blood samples were collected upon hospital presentation. Plasma was obtained as described above, but using whole blood diluted 1:2 in RPMI-1640. Prior to use in experiments, plasma samples were quickly thawed at 37°C and incubated for 1 h at 53°C to inactivate viral particles. For the knockout and scRNA-seq experiments, plasma samples were pooled across each patient category.

Samples for neutrophil scRNA-seq were from patients enrolled in the Brigham and Women's Hospital's (BWH); the criteria for patient recruitment for this cohort are described elsewhere 63, 64 . Control samples consisted of blood samples from age, gender, and ethnicity-matched healthy controls obtained from Research Blood Components (MA, USA).

To maintain viability of neutrophils for scRNA-seq, fresh blood was collected in EDTA Vacutainer tubes (BD Biosciences) and processed within 4 h. To remove red blood cells (RBC) from the sample, 1X RBC lysis buffer (eBioscience) was added directly at a 10:1 ratio to the sample. After 5 min, the cells were centrifuged at 400g for 5 min and resuspended in 1X RBC lysis buffer to further clear the sample of RBCs. The remaining cells were stained with a general panel: DAPI, CD3-APC (clone HIT3a), CD19-APC (clone HIB19), CD56-APC (clone 5.1H11), CD14-FITC (clone M5E2), CD15-AF700 (clone HI98), CD11b-PE-Cy7 (clone ICRF44) (BioLegend). Single neutrophils were sorted using an SH800 cell sorter (Sony) into 10 μL TCL buffer (Qiagen) with 1% β-mercaptoethanol (BME, Sigma) in individual wells of a 96-well plate.

Bulk RNA-seq was performed using Smart-Seq2 65 with minor modifications, as described previously 66 , using 1,000 cells as input. All RNA-seq libraries were sequenced with 38 x 38 paired-end reads using a NextSeq (Illumina). RNA-seq libraries were sequenced to a depth of >2 million reads per sample. STAR was used to align sequencing reads to the UCSC hg19 transcriptome and RSEM was used to generate an expression matrix for all samples. Both raw count and transcripts per million data were analyzed using edgeR and custom python scripts.

Intracellular staining with S100A8-PE (clone REA917, Miltenyi Biotec) was performed using the Cytofix/Cytoperm kit (BD Biosciences) following the manufacturer's protocol. For staining of phosphorylated STAT3 (Y705), cells were first fixed with 4% paraformaldehyde (Electron Microscopy Sciences) for 15 min at room temperature. The cells were then washed twice with 1X PBS and resuspended in 95% ice-cold methanol and left at -20°C overnight. The permeabilized cells were then stained with a pSTAT3-Y705 antibody (clone 13A3-1, BioLegend) for 30 mins on ice. Flow cytometry data were acquired on a Cytoflex LX (Beckman Coulter) and analyzed using FlowJo v10.1.

Cas9 protein, pre-designed guide RNAs targeting IL6ST, IL6R, IL10RA, and IL10RB and non-targeting guide RNAs (from GeCKO v2 library) were purchased from Integrated DNA Technologies. RNP complexes were assembled by combining 2.1 μL 1X PBS, 1.2 μL 100 μM gRNA, and 1.7 μL 10 μg/mL Cas9 protein and incubating at room temperature for 15 min. The complexes were added to 50,000 -100,000 CD34+ HSPCs resuspended in 20 μL P3 (Lonza) and electroporated (program code DZ-100) using the 4D-Nucleofector system (Lonza). After electroporation, the cells were immediately transferred to 500 μL of HSPC media and rested for 48 h. Knockout efficiency in CD34+ HSPCs was assessed after 48 h via flow cytometry using the following panel: CD34-BV650 (clone 561), CD38-PE/Cy5 (clone HIT2), CD126-APC (clone UV4), CD210-PECy7 (clone 3F9) (BioLegend).

Samples from sepsis patients and controls were thawed and analyzed in parallel using the Legendplex Human Inflammation for IFN-α2, IFN-γ, IL-1β, IL-6, IL-10, TNFα and IL-18, and Human Hematopoietic Stem Cell Panels for M-CSF and GM-CSF (BioLegend). Flow cytometry data were acquired on a Cytoflex LX (Beckman Coulter) and analyzed using FlowJo v10.1.

Samples from COVID-19 patients and controls were analyzed using a commercially available multiplexed proximity extension assay (Olink Proteomics).

T cell co-culture was performed as previously described 67 with minor modifications. Briefly, tissue-culture plates were coated with 5 μg/mL purified anti-CD3 (clone HIT3a, Biolegend) at 4°C overnight and subsequently washed twice with 1X PBS. PBMCs from a healthy donor (Research Blood Components) were labelled with CFSE (Invitrogen) following the manufacturer's protocol. PBMCs were resuspended in SFEM II (StemCell Technologies) with 5 μg/mL purified anti-CD28 (clone CD28.2, BioLegend) and plated at a density of 1M cells/mL. Isolated iMS1 or iMono cells were added at different ratios as indicated. The cells were left in culture for 3-4 days, with media replenished after 2 days. At the end of incubation, the cells were stained with CD3-AF700 (clone OKT3), CD4-APC (clone OKT4), and CD8a-PE (clone RPA-T8) (BioLegend) to determine the amount of CFSE dilution within the T cells.

Primary HUVECs and HREs were purchased from Lonza and cultured in EGM-2 and REGM, respectively. To improve cell viability, HUVECs were cultured in tissue culture plates pre-coated with Matrigel (Corning) diluted 1:100 in EBM-2 (Lonza) . Cells were used for experiments within 3-5 passages. Prior to co-culture, HUVECs and HREs were treated with 10 ng/mL TNF α (PeproTech) for 24 h. For the co-culture experiments, CFSE-labelled iMono or iMS1 cells were added at a 1:1 ratio to a confluent monolayer of HUVECs or HREs and incubated for 2 h. The cells were washed 3x with 1X PBS and detached by adding 1X Accutase (Innovative Cell Technologies). The cells were transferred to FACS buffer after 15 mins, and 1,000 CFSE-negative cells were sorted into 10 μL TCL buffer (Qiagen) with 1% BME (Sigma) for bulk RNA-seq. Conditioned media was prepared by incubating iMS1 or iMono cells at 0.5 M cells/mL in EGM-2 or REGM for 24 h overnight.

C1 and C4-plasma generated cells were isolated after 7 d of culture using human CD14 Microbeads (Miltenyi Biotec). Isolated CD14+ cells were validated using flow cytometry (CD14-FITC, clone M5E2; BioLegend) to be of >90% purity. Cells were stimulated for 4h with 100 ng/mL IFN-β (PeproTech) or 1 μg/mL HMW poly-I:C (Invivogen). After stimulation, cells were washed twice with FACS buffer and resuspended in 20 μL TCL buffer (Qiagen) with 1% BME (Sigma) for bulk RNA-seq.

Global, regional, and national sepsis incidence and mortality, 1990-2017: analysis for the Global Burden of Disease Study

Presenting Symptoms Independently Predict Mortality in Septic Shock: Importance of a Previously Unmeasured Confounder

The Search for Efficacious New Therapies in Sepsis Needs to Embrace Heterogeneity

The immunopathology of sepsis and potential therapeutic targets

Surviving Sepsis Campaign: Guidelines on the Management of Critically Ill Adults with Coronavirus Disease 2019 (COVID-19)

SARS-CoV-2 and viral sepsis: observations and hypotheses

Systems biological assessment of immunity to mild versus severe COVID-19 infection in humans

Extrapulmonary manifestations of COVID-19

An immune-cell signature of bacterial sepsis

Persistent lymphopenia after diagnosis of sepsis predicts mortality

Sepsis-Induced T Cell Immunoparalysis: The Ins and Outs of Impaired T Cell Immunity

Impact of sepsis on CD4 T cell immunity

Sepsis Triggers a Late Expansion of Functionally Impaired Tissue-Vascular Inflammatory Monocytes During Clinical Recovery

Differential Gene Expression in Circulating CD14+ Monocytes Indicates the Prognosis of Critically Ill Patients with Sepsis

Human monocytes undergo functional re-programming during sepsis mediated by hypoxia-inducible factor-1α

Monocyte deactivation in septic patients: restoration by IFN-gamma treatment

Bacterial reprogramming of PBMCs impairs monocyte phagocytosis and modulates adaptive T cell responses

Human leucocyte antigen (HLA-DR) gene expression is reduced in sepsis and correlates with impaired TNFα response: A diagnostic tool for immunosuppression

Proof of concept study of mass cytometry in septic shock patients reveals novel immune alterations

Downregulation of Blood Monocyte HLA-DR in ICU Patients Is Also Present in Bone Marrow Cells

Immunosuppression in patients who die of sepsis and multiple organ failure

A high frequency of MDSCs in sepsis patients, with the granulocytic subtype dominating in gram-positive cases

Myeloid-derived suppressor cell function and epigenetic expression evolves over time after surgical sepsis

Intracellular S100A9 Promotes Myeloid-Derived Suppressor Cells during Late Sepsis

S100-alarmin-induced innate immune programming protects newborn infants from sepsis

Myeloid-derived suppressor cells control microbial sepsis

Myeloid-derived suppressor cells evolve during sepsis and can enhance or attenuate the systemic inflammatory response

Severe COVID-19 is marked by a dysregulated myeloid cell compartment

Single-cell profiling of Ebola virus infection in vivo reveals viral and host transcriptional dynamics

Identifying gene expression programs of cell-type identity and cellular activity with single-cell RNA-Seq

Inhibition of dendritic cell differentiation and accumulation of myeloid-derived suppressor cells in cancer is regulated by S100A9 protein

IL-10 induces an immune repressor pathway in sepsis by promoting S100A9 nuclear localization and MDSC development

A community approach to mortality prediction in sepsis via gene expression analysis

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

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

Myeloid-derived suppressor cells coming of age

COVID-19: consider cytokine storm syndromes and immunosuppression

Ascites-derived IL-6 and IL-10 synergistically expand CD14+HLA-DR-/low myeloid-derived suppressor cells in ovarian cancer patients

Distinct myeloid suppressor cell subsets correlate with plasma IL-6 and IL-10 and reduced interferon-alpha signaling in CD4 + T cells from patients with GI malignancy

Cytokine response is determined by duration of receptor and signal transducers and activators of transcription 3 (STAT3) activation

Activation of STAT3 by IL-6 and IL-10 in primary human macrophages is differentially modulated by suppressor of cytokine signaling 3

Expansion and differentiation of CD14+CD16(-) and CD14+ +CD16+ human monocyte subsets from cord blood CD34+ hematopoietic progenitors

Tumor-induced tolerance and immune suppression depend on the C/EBPbeta transcription factor

Early Expansion of Circulating Granulocytic Myeloid-derived Suppressor Cells Predicts Development of Nosocomial Infections in Patients with Sepsis

Mechanisms and treatment of organ failure in sepsis

Role of plasma matrix-metalloproteases (MMPs) and their polymorphisms (SNPs) in sepsis development and outcome in ICU patients

Diagnosis and management of sepsis-induced coagulopathy and disseminated intravascular coagulation

Dynamic evolution of coagulopathy in the first day of severe sepsis: relationship with mortality and organ failure

Pathological inflammation in patients with COVID-19: a key role for monocytes and macrophages

Immune complement and coagulation dysfunction in adverse outcomes of SARS-CoV-2 infection

Focus shifts to antibody cocktails for COVID-19 cytokine storm

Interleukin-6: designing specific therapeutics for a complex cytokine

Identification of monocyte-like precursors of granulocytes in cancer as a mechanism for accumulation of PMN-MDSCs

Activated human hepatic stellate cells induce myeloid derived suppressor cells from peripheral blood monocytes in a CD44-dependent fashion

A human promyelocytic-like population is responsible for the immune suppression mediated by myeloid-derived suppressor cells

Generation and functional characterization of MDSC-like cells

Cell Hashing with barcoded antibodies enables multiplexing and doublet detection for single cell genomics

Single-cell RNA-seq reveals new types of human blood dendritic cells, monocytes, and progenitors

SCANPY: large-scale single-cell gene expression data analysis

Generalizing RNA velocity to transient cell states through dynamical modeling

Robust enumeration of cell subsets from tissue expression profiles

Empowering Multi-Cohort Gene Expression Analysis to Increase Reproducibility

Circulating mitochondrial DNA in patients in the ICU as a marker of mortality: derivation and validation

Inflammasome-regulated cytokines are critical mediators of acute lung injury

Full-length RNA-seq from single cells using Smart-seq2

Multiplexed enrichment and genomic profiling of peripheral blood cells reveal subset-specific immune signatures

Methods to Measure MDSC Immune Suppressive Activity In Vitro and In Vivo

We thank Matteo Gentili, Bingxu Liu, Thomas Eisenhaure, Anna Le, Mohammad Najia, Rebecca Carlson, Joshua Peters, Joshua Elacqua, and other members of the Blainey and Hacohen Labs (Broad Institute) for helpful discussions. We thank Alec Schmaier and Samir Parikh (Beth Israel Deaconess Medical Center) for advice on endothelial cell experiments. We also thank the Broad Flow Cytometry core for assistance in cell sorting experiments and the Broad Genomics Platform for assistance in sequencing. We are grateful to Drs. Laura Fredenburgh, Paul Dieffenbach, and Sam Ash for assistance with patient phenotyping at Brigham and Women's Hospital. 

MHC-II (right) modules in CD14-expressing cells across the different plasma treatment conditions. The experiment in (a-d) was performed on 2 healthy bone marrow donors with 2 plasma donors for each condition; a total of 3,039 and 5,254 cells were profiled for Control and URO plasma treatment, respectively. The experiment in (e-h) was performed on 2 healthy bone marrow donors with pooled plasma from all donors in (c); a total of 4,449, 4,591, 3,129 and 3,711 cells were profiled for C1-C4 plasma, respectively. that are also among the top 30 genes in the MS1 module are highlighted in red. (d) Gene weight correlation between the MS1-like modules detected in neutrophils from the plasma-incubation (x-axis) and patient datasets (y-axis). Significance of the correlations (Pearson r ) are calculated with a permutation test. (e) Heatmap showing the top 20 differentially expressed genes (Wilcoxon rank-sum test, FDR < 0.01) between neutrophils sorted from critically-ill patients and healthy controls. BAC, bacteraemic patient; ICU, patient in intensive care.