Discovery of key regulators in classical monocyte phenotypes linked to COVID-19 severity using single-cell multi-omics sequencing

Summary

Dysregulated immune responses often accompany severe COVID-19, but the underlying epigenetic mechanisms driving monocyte heterogeneity and COVID-19 progression remain unclear. Here, we applied single-cell multi-omics profiling to peripheral blood mononuclear cells across healthy and four COVID-19 severity stages. We identified two severity-associated classical monocyte subtypes-IL7R+ and CD163+, These subtypes exhibit distinct transcriptional and epigenetic landscapes. By constructing a gene regulatory network and in silico perturbations, we revealed that ETS1 is a key driver of IL7R + monocytes with T cell-like signaling features and that JDP2 is a repressor that maintains the profibrotic, anti-inflammatory identity of CD163+ monocytes by suppressing AP-1 activity. These subtypes were enriched in the moderate-to-critical stages and exhibited signaling pathways associated with tissue remodeling and immune suppression. Our findings define monocyte heterogeneity as linked to COVID-19 severity and identify ETS1 and JDP2 as central regulators, offering insights into immune dysregulation, potential therapeutic targets for fibrosis, and long-term sequelae.

Graphical abstract

Highlights

• •Single-cell multi-omics reveals biological changes of classical monocytes in COVID-19
• •Two severity-associated subtypes (IL7R+ cMono and CD163+ cMono) are identified
• •In IL7R+ cMono, ETS1 drives T cell-like signaling and immune suppressive programs
• •JDP2 suppresses AP-1 activity to maintain an anti-inflammatory CD163+ cMono state

Molecular biology; Immunology; Microbiology

Introduction

Coronavirus disease (COVID-19) causes a wide range of symptoms, from asymptomatic cases to mild symptoms, such as fever, fatigue, and shortness of breath, to severe illness requiring intensive care unit (ICU) admission and mechanical ventilation.¹ While most individuals recover within a few days to two weeks, 10%–35% of patients experience prolonged symptoms lasting for months or even years.² In severe COVID-19 cases, increased inflammatory response due to excessive immune activation can lead to tissue damage and multi-organ failure.³ In infectious diseases such as COVID-19, immune cells undergo epigenetic changes—such as histone modifications and chromatin structural changes—in response to cytokines, pathogen-associated molecular patterns (PAMPs), and damage-associated molecular patterns (DAMPs), and are known to play a particularly important role in regulating inflammatory responses.^4,5

These epigenetic changes also lead to differentiation and functional alterations in various immune cells. Emerging evidence highlights epigenetic factors as critical regulators of COVID-19 severity, influencing gene expression, and immune cell function.⁶ A previous study comparing histone modifications (H3K4me3 and H3K27me3) and gene expression profiles in peripheral blood mononuclear cells (PBMCs) from healthy individuals and patients with COVID-19 demonstrated that the expression of genes involved in inflammatory pathways, including transforming growth factor beta (TGF-β), interleukin (IL)-1β, IL-6, and IL-17, was associated with histone methylation patterns.⁷ Furthermore, single-cell assay for transposase-accessible chromatin with sequencing (ATAC-seq) analysis of PBMCs from healthy individuals and convalescent patients with COVID-19 revealed widespread chromatin remodeling across almost all immune cell compartments following recovery. Epigenetically reprogrammed trained monocytes, particularly the CD14⁺ and CD16⁺ subsets, were enriched in convalescent individuals. In CD8⁺ T cells, COVID-19-specific chromatin changes in CD8⁺ T cells were associated with altered transcription factor (TF) activity and clonal expansion.⁸ Furthermore, a recent study has demonstrated that severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) mRNA vaccination induces a significant increase in H3K27ac at promoters of pro-inflammatory innate immune genes in human monocyte-derived macrophages. These epigenetic changes persisted for up to six months following a two-dose regimen, indicating that dynamic and durable epigenomic remodeling underlies sustained inflammatory innate immune activation.⁹ Despite these advances, a comprehensive understanding of the gene regulatory networks, including dynamic epigenetic alterations, key regulatory factors, and downstream gene expression programs, remains limited, in part due to the heterogeneity of the immune cell population and variability in COVID-19 severity.

Among various immune cells, monocytes are particularly important because of their functional diversity. Due to their high plasticity, monocytes respond rapidly to environmental signals and serve as important mediators of immune responses during and after viral infection. During viral infection, monocytes accumulate in the bloodstream and secrete pro-inflammatory cytokines, such as tumor necrosis factor-α (TNF- α), IL-1β, and IL-6. In addition, monocytes migrate to infected or damaged tissues, contributing to local inflammation and recruiting immune cells via chemokine secretion.^10,11 Dysregulated monocyte/macrophage differentiation can result in abnormal immune responses, chronic inflammation, or excessive fibrosis.^12,13 Recent COVID-19 studies have highlighted monocytes and macrophages as central immune cell types in the severity of COVID-19, accompanied by upregulation of genes involved in immune activation and inflammation, along with epigenetic reprogramming.^14,15 Epigenetic changes in hematopoietic stem and progenitor cells following COVID-19 infection transmit enduring epigenetic memory to monocytes.¹⁶ Furthermore, depending on the severity of COVID-19 infection, monocytes with distinct phenotypes not only accumulate in the blood but also infiltrate lung tissues while retaining their unique characteristics.^17,18 Despite these efforts, there is still a lack of understanding regarding the key regulators and epigenetic changes that control the differentiation and cellular function of monocytes/macrophages specific to COVID-19 severity.

Single-cell sequencing technologies have been widely used to dissect the heterogeneity of diverse cell populations, including immune cells, and to identify previously unrecognized cell types.^19,20,21 More recently, the advent of single-cell epigenomic mapping techniques, coupled with the development of multi-omics platforms that integrate transcriptomic and epigenomic data from the same cells, has further advanced our ability to investigate cellular states and regulatory mechanisms at unprecedented resolutions.^22,23 These multi-omics approaches have been particularly powerful for reconstructing gene regulatory networks (GRNs) and identifying key transcriptional regulators that control cell fate decisions, immune responses, and disease-associated cellular programs.^24,25

To address the poorly understood cellular mechanisms and key epigenetic regulators underlying monocyte phenotypes in severe COVID-19, we generated single-cell multi-omics data from the PBMCs of patients with COVID-19 across healthy and four severity groups. Our analysis revealed significant multimodal changes in the circulating immune cells, particularly monocytes, and identified two monocyte phenotypes linked to increased severity. Through multi-omics-based gene regulatory network analysis and perturbation simulations, we identified ETS proto-oncogene 1, transcription factor (ETS1), and Jun dimerization protein 2 (JDP2) as key transcription factors (TFs) that regulate monocyte differentiation into severity-associated phenotypes. This study provides deep insights into the two monocyte phenotypes enriched at specific severity stages, highlights the key regulators of their differentiation, and suggests a potential link with COVID-19 sequelae.

Results

Single-nucleus multi-omics profiling of PBMCs reveals distinct immune landscapes across COVID-19 severity stages

To investigate the biological mechanisms associated with the severity of COVID-19 infection, we obtained PBMCs from 3 healthy donors and 14 patients with COVID-19 for single-nucleus multi-omics analysis. Patients with COVID-19 were classified according to the World Health Organization severity criteria into the following groups: mild (n = 2), moderate (n = 3), severe (n = 6), and critical (n = 3) (Table S1).²⁶ We applied stringent filtration and batch correction procedures to multi-omics data, retaining a total of 47,851 individual cells (healthy: 11,417, COVID-19: 36,434 cells). These high-quality cells were annotated into 16 cell types based on gene expression, chromatin accessibility, and multimodality, and were evenly distributed across samples, infection status, and severity levels (Figures 1A and S1A). Since patients received different therapeutic regimens, we examined the impact of drug treatment and found that the differences were more significantly related to COVID-19 severity than to treatment (Figures S1B and S1C). These included T lymphocytes (CD4, CD8, and regulatory T cells), B lymphocytes (B and plasma cells), natural killer (NK) cells, and myeloid cells (monocytes and dendritic cells). We observed the expression of known markers for all 16 cell types along with chromatin accessibility near the transcription start site (TSS) of each marker and showed a dynamic pattern of gene expression and chromatin accessibility for each cell type (Figures 1B, 1C, and S1D). Next, we identified a total of 112,124 peaks across 16 distinct cell types. When we correlated gene activity scores using chromatin accessibility of the gene body and promoter with the corresponding genes, they showed an overall positive correlation in each cell type (Figure 1D). Moreover, the transcriptome and epigenome patterns showed a high correlation between cell types with the same immune cell lineages (lymphoid and myeloid lineages) (Figure S1E). We performed differential abundance testing based on the infection status to identify the cell type proportion changes due to COVID-19 infection.²⁷ We identified 440 differentially abundant Nhoods (a local cluster of cells; see STAR Methods for details), including 203 that increased and 237 that decreased during COVID-19 infection (Figure 1E, FDR <0.1). Increased Nhoods were located in myeloid cells, including 145 classical monocytes (cMono) and 10 non-classical monocytes (ncMono). In contrast, decreased Nhoods were primarily found in T lymphocytes, including 115 CD4 naive T cells, 44 CD4 central memory T cells (CD4 TCM), 65 CD8 naive T cells, and 5 CD8 effector memory T cells (CD8 TEM), confirming that these findings are consistent with those of previous studies.^14,28 Additionally, we investigated whether the severity stage affected the changes in the immune cell population (Figures 1F, S2A, and S2C). CD4/8 naive cells and mucosal-associated invariant T (MAIT) cells decreased proportionally with increasing severity, whereas cMono and plasma cells showed a proportional increase with severity. The proportion of ncMono significantly increased in the mild and moderate stages but gradually decreased with increasing severity. Both previous studies also showed the similar changes in cell populations according to severity and infection status (Figure S2B).^29,30 Consistent with the changes in cell populations, cell-cell communication analysis revealed that interactions were enriched among T lymphocyte populations in healthy stage, whereas increasing COVID-19 severity was accompanied by a progressive rise in interactions involving myeloid lineages (Figure S2D). Severity-specific interactions were associated with differentiation, activation, and survival of each cell type. For example, in the mild stage, the HLA-E–CD94:NKG2C interaction in NK cells has been reported to induce NK cell activation.³¹ In the moderate stage, an increase in Mac-1/LFA-1 signaling was observed in ncMonos, which is associated with cell activation, trafficking, tissue infiltration, and inflammation induction.³² Next, in the severe stage, IFNGR1 signaling activates JAK1/JAK2, leading to the transcription of interferon-stimulated genes (ISGs) and the promotion of a pro-inflammatory response in cMonos.³³ Finally, in the critical stage, increased interactions mediated by TNFSF13B were detected in plasma cells, which bind to B cell surface receptors TACI (TNFRSF13B) and BCMA (TNFRSF17), playing a crucial role in B cell survival, maturation, and antibody production.³⁴ These interactions were closely related to the observed changes in cell population ratios depending on severity. To identify the cell types contributing to severity-associated biological changes in COVID-19, we identified differentially expressed genes (DEGs) and differentially accessible regions (DARs) from all cells as pseudo-bulk (Table S2, p < 0.05, Log₂Fold Change >0.5). Using these severity signature genes, principal-component analysis (PCA) of individual samples revealed clear separation according to severity (Figure S3A). We next analyzed the DEG and DAR modules across cell types. Myeloid, B, and plasma cells showed the highest COVID-19 severity-specific DEG module scores, particularly at the more severe stages (severe and critical) (Figure 1G). This trend is similar to the changes in the cell population that increased with COVID-19 infection and severity. In contrast, most lymphocytes had high scores only in the HC stage, whereas CD8 TEM and NK cells, which have cytotoxic functions, showed increased scores in the mild stage. The representative severity signature genes exhibited severity-specific expression patterns and were associated with the activation, survival, and differentiation of cell types that increased with severity (Figures S3B and S3C). In particular, the severe stage signature genes IL1R1 and IL6R, which encode receptors for the immune-activating cytokines IL-1β and IL-6, respectively, are closely related to the activation of cMono.³⁵ Moreover, we calculated module scores in publicly single-cell datasets according to infection status and severity using severity signature genes.^29,30,36 The results showed that these genes effectively distinguished both COVID-19 infection and severity stages (Figure S3D). Similar to the COVID-19 severity-specific DEGs module score, cMono, ncMono, and cDC were the cell types with the highest COVID-19 severity-specific DAR module scores from the moderate-to-critical stages. Taken together, myeloid cells, particularly cMono, exhibited enhanced activation of severity-related GRNs at higher severity stages, promoting further investigation of cMono.

Figure 1.

Multi-omics landscape of PBMCs and dynamic changes following COVID-19 severity

(A) UMAP visualization of single-nucleus multi-omics sequencing data from COVID-19 and healthy PBMCs, based on RNA (left), ATAC (middle), and multimodal (right) profiles. The dataset includes 47,851 individual cells obtained from 17 donors and is classified into 16 major immune cell types.

(B) Dot plot showing the expression of representative marker genes across 16 major immune cell types. Each dot represents the average expression level (color intensity) and the percentage of cells expressing the gene (dot size) within each cell type.

(C) Genome browser visualization displaying chromatin accessibility profiles at promoter regions of representative marker genes across 16 major immune cell types.

(D) Correlation matrix comparing gene expression (RNA) and gene activity (ATAC) profiles across 16 major immune cell types. The size and color of each circle indicate the strength of correlation (red: positive, blue: negative).

(E) The UMAP plot displaying neighborhood (Nhood) groups by conditions defined using MiloR (left). The size of each dot corresponds to the number of cells in the neighborhood (Nhood size) and the thickness of the edges connecting Nhoods indicates the number of shared cells between two different Nhoods (overlap size). Nhoods are colored by the log fold change (logFC) in abundance between COVID-19 and healthy. Scatterplot showing the log fold change (logFC) in abundance for each neighborhood. Each dot represents a neighborhood within each cell type. Colors indicate the condition in which the neighborhood is enriched (red: Nhoods enriched in COVID-19, blue: Nhoods enriched in healthy, gray: Nhoods non-enriched).

(F) Boxplots showing the proportions of representative cell types that decrease (top row: CD4 naive, CD8 naive, MAIT) or increase (bottom row: cMono, ncMono, and plasma cells) according to COVID-19 disease severity (sample numbers: HC = 3, mild = 2, moderate = 3, severe = 6, critical = 3). Boxes indicate the interquartile range (25th–75th percentiles), the center line denotes the median, and whiskers extend to 1.5x the interquartile range.

(G) Heatmaps showing module scores calculated from severity-enriched differentially expressed genes (Left) and chromatin accessibility regions (right) across severity (HC, mild, moderate, severe, critical, log2fold change >0.5, p < 0.05 by Wilcoxon Rank-Sum test). Scores are scaled across cell types to visualize relative enrichment.

Multimodal sub-clustering of classical monocytes reveals functionally distinct subtypes associated with COVID-19 severity

To elucidate COVID-19 severity-associated changes in classical monocytes, we performed multimodal sub-clustering on cMono. The 9,949 cells were divided into four clusters, each accounting for more than 20% of the total cells (Figures 2A and S4A). The clusters were distinguishable based on COVID-19 infection and severity stages (Figure S4B). When we identified sub-cluster-specific DEGs (Table S3), two sub-clusters exhibited the upregulation of M1 phenotype-associated genes that play important roles in the inflammatory response, antiviral response, and cytokine secretion. Therefore, we designated these sub-clusters as M1-like cMono1 and cMono2 (Figure 2B).^37,38 In contrast, another sub-cluster showed an increase in the expression of genes related to T cell receptor (TCR) signaling, T cell activity, and development, including CD247, IL7R, LTB, TCF7, and CCL5.³⁹ Notably, these genes exhibited increased expression exclusively in this sub-cluster, at levels comparable to T lymphocytes (Figure S4C). As previous studies have reported the presence of IL7R-positive monocytes in inflammatory diseases, such as COVID-19 and rheumatoid arthritis, we annotated this sub-cluster as IL7R + cMono.^17,40 The last sub-cluster, CD163+ cMono, showed a significant increase in the expression of M2 phenotype-associated genes, including CD163, MERTK, PPARG, and TGFBR2.⁴¹ CD163 has been reported as a marker for dysfunctional monocytes in response to COVID-19 infection.^18,42 Chromatin accessibility near the TSS of marker genes also showed a sub-cluster-specific increase (Figure S4D). We performed a differential abundance analysis between infection statuses to identify the relationship between each sub-cluster and COVID-19 infection severity. We identified 139 differentially abundant Nhoods, of which 89.2% (124 Nhoods) increased during COVID-19 infection (Figure 2C). The proportion of M1-like cMonos was abundant in the healthy and mild states, whereas the proportions of IL7R+ and CD163+ cMonos increased significantly in the moderate stage and above (Figure 2D). Previous studies have reported that COVID-19 vaccination can increase the overall proportion of circulating monocytes in blood and induce functional changes, accompanied by alterations in subpopulation distributions.^43,44 In this study, 10 of the 14 COVID-19 donors had received COVID-19 vaccination, with varying numbers of doses across individuals (Table S1). However, PCA did not reveal differences within individual sub-clusters according to vaccination status or number of vaccine doses (Figure S4E). Consistently, no significant differences in the proportions of monocyte sub-clusters were observed based on vaccination status. Instead, changes in monocyte subpopulation composition were more strongly associated with COVID-19 disease severity than with vaccination history (Figure S4F). To validate the severity-specific expression of IL7R and CD163 in patients with COVID-19, we examined their expression in single-cell transcriptome datasets of bronchoalveolar lavage fluid (BALF) from patients infected with COVID-19.⁴⁵ In BALF macrophages classified into three severity stages, IL7R was expressed only in the moderate stage, while CD163 expression increased with severity (Figure 2E). Also, previous studies of PBMCs from COVID-19 patients also identified IL7R+ and CD163+ cMonos. Consistent with our findings, CD163+ cMono sub-cluster was notably enriched in in patients with higher severity (Figure S4G).^29,30 Additionally, the IL7R+ sub-cluster was identified in classical monocytes from the synovial fluid mononuclear cells of patients with spondyloarthritis,⁴⁰ and the module score of its sub-cluster-specific genes was enriched in IL7R + cMono (Figure 2F; Table S4). Similarly, the CD163+ sub-cluster has been observed in classical monocytes from patients with early severe COVID-19,¹⁸ and the module score of its sub-cluster-specific genes was enriched in CD163+ cMono. To investigate the cellular functions of the monocyte sub-clusters, we performed a gene ontology (GO) enrichment analysis of the biological processes associated with each sub-cluster (Figure 2G). IL7R + cMono was significantly enriched in processes such as the interleukin-2-mediated signaling pathway, positive regulation of the receptor signaling pathway via STAT, and negative regulation of apoptotic processes. The IL-2 signaling pathway is primarily known for its involvement in T cell proliferation and activation; however, it also promotes the activation of the JAK-STAT pathway, particularly STAT5 activation in monocytes. STAT5 activation is crucial for regulating various processes like monocyte survival and differentiation.^46,47 Additionally, pathways like fibroblast activation and positive regulation of TGF-β receptor signaling were enriched in CD163+ cMono, suggesting a role in tissue remodeling and repair.^48,49 The activation of these signaling pathways in monocytes can lead to an overproduction of extracellular matrix and the formation of fibrotic tissue, which may result in lung fibrosis.^50,51 This process also weakens the immune response by promoting the differentiation of monocytes into the M2 phenotype, which is associated with tissue repair and anti-inflammatory effects but also impairs the ability to defend against infections. These observations support the association of IL7R+ and CD163+ cMono cells with COVID-19 infection and inflammatory diseases, highlighting their potential roles in severity, immune regulation, and fibrosis.

Figure 2.

Heterogeneity and severity associations of cMono sub-clusters

(A) The UMAP plot presents the sub-clustering results of cMono based on multimodal profiling, along with the corresponding sub-cluster labels.

(B) Dot plot showing the average expression and percentage of cells expressing selected marker genes across cMono sub-clusters. Color scale indicates scaled average expression, and dot size reflects the percentage of cells expressing each gene.

(C) The UMAP plot displaying Nhood groups by conditions defined using MiloR (Left). And the scatterplot showing logFC of Nhood between condition for each cMono sub-clusters (Right).

(D) Stacked bar chart showing the proportions of cMono sub-clusters-M1-like cMono1, M1-like cMono2, IL7R + cMono, and CD163+ cMono-across individual severity stages.

(E) Violin plots showing the expression of IL7R and CD163 in lung tissue macrophages across three COVID-19 severity groups (healthy, moderate, severe) in published single-cell RNA data.⁴⁵

(F) Feature plot showing gene module scores for cMono DEGs of IL7R+ (top) and CD163+ (bottom) in published cMono single-cell RNA data.^18,40 Color intensity reflects the normalized module score.

(G) Dot plot of module scores for genes belonging to the top enriched Gene Ontology (GO) biological processes across cMono sub-cluster. Dot color indicates the average module score for the genes in that GO term, and dot size reflects the percentage of cells within the sub-cluster expressing at least one gene from the term’s gene set.

Distinct gene regulatory programs define classical monocyte subtypes and their association with COVID-19 severity

We observed changes in transcriptional and chromatin accessibility in each cMono sub-cluster. IL7R + cMono exhibited the highest number of DEG and DAR changes, with 2,633 DEGs and 11,157 DARs (Figure S5A). Additionally, more than 25% of the DARs in all sub-clusters were located in the promoter regions within 3 kb of the TSS (Figure S5B). Consistent with the large number of DEGs and DARs, 63.5% (7,081 regions) of the chromatin regions in IL7R + cMono were located in promoter regions, indicating a strong association between these chromatin regions and gene transcription. To identify potential key regulators of each sub-cluster, we analyzed the overrepresented TF-binding motifs in the top 1,500 DARs of each sub-cluster and computed motif activity scores per cell based on chromatin accessibility variability (Figure 3A; Table S5). Highly accessible DARs in M1-like cMono 1 and 2 were enriched with C/EBP, SPI, and STAT sub-family motifs with elevated motif activities, key regulators of monocyte differentiation, inflammation, innate immunity, and cytokine production.^52,53,54 In contrast, in IL7R + cMono, the motif activities of RUNX1, ETS1, KLF12, NRF1, and ZBTB7A were increased, while in CD163+ cMono, the motif activity of the SMAD family, a downstream factor of TGF-β signaling, as well as members of the AP-1 family, such as FOS, JUN, and JDP2, was significantly increased.^55,56 Subsequently, we performed correlation analysis between TF motif activity and own gene expression in each sub-cluster to identify TFs showing positive correlations (Figures 3B and S5C). Several TFs, such as STAT1/2, JUN, CEBPB/D, ETS1, NRF1, SMAD3, and JDP2, were positively correlated, suggesting the presence of distinct gene regulatory mechanisms among the sub-clusters. To understand the gene regulation mechanisms in each sub-cluster, we constructed GRNs at the single-cell level by integrating motif activity, chromatin accessibility, and gene expression data using SCENIC+.⁵⁷ SCENIC+ defines regulons based on two relationships: (1) TF-to-gene correlation and (2) region-to-gene correlation. These are represented using a combination of positive and negative signs (e.g., +/+ and −/+). For example, +/+ regulon denotes an activator regulon that opens the chromatin of the target regions and upregulates target gene expression, whereas −/+ regulon represents a repressor regulon that closes the chromatin of the target regions and represses the target gene expression. A total of 303 active TFs identified from all cMono cells, 40 activator regulons, and 8 repressor regulons were classified as sub-cluster-specific (Figures 3C and S5D). A substantial number of TFs identified through motif activity and enrichment analyses (Figures 3A and 3B) were also identified as activator regulons by SCENIC+. Although FOS and JUN exhibited high motif activity and enrichment in CD163+ cMono, SCENIC+ identified them as repressors (Figures 3A and S5D). TFs that overlapped between the SCENIC+ and motif analyses were ranked among the top 10 within each sub-cluster (Figure S5E). Furthermore, the regulon scores of overlapping TFs in each sub-cluster changed dynamically across different severity levels (Figure 3D). Among the top regulons in M1-like, cMono1 and 2 were STAT1/2, NFIL3, JUNB, FOS, and CEBPD/B, which were predominantly upregulated in the mild stage. In contrast, the top regulons in IL7R + cMono, including ETS1, KLF12, NRF1, and ZBTB7A, were mainly upregulated in the moderate and critical stages, whereas those in CD163+ cMono, such as SMAD3, ELF1, and NR3C1, exhibited the highest scores in the severe stage, suggesting an association between these regulons and COVID-19 severity. Given that a regulon is defined by TF motif activity, chromatin accessibility of target regions, and expression of target genes, we assessed each factor separately to ensure that the identification of these regulons was not biased by any single element. We focused on the regulons of IL7R + cMono and CD163+ cMono, which showed associations with COVID-19 severity and showed that all elements were upregulated in each sub-cluster (Figure 3E). Moreover, severity-specific regulons were also associated with cellular functions in each sub-cluster. Representatively, BACH1, a TF responsive to heme and reactive oxygen species (ROS) signaling, was highly expressed in the M1-like cMono1 (Figure S6A). 86.2% of BACH1 downstream target genes overlapped with DEGs of M1-like cMono1 and showed elevated expression (Figures S6B and S6C; Table S6). Along with the increased expression of BACH1, genes involved in the ROS response were upregulated, while HMOX1, a key transcription factor regulating ferroptosis, and genes related to iron metabolism were downregulated, consistent with BACH1-mediated transcriptional repression under oxidative stress (Figure S6D).⁵⁸ Furthermore, GO term analysis of BACH1 target genes revealed biological processes closely associated with the M1 phenotype (Figure S6E). To confirmed the regulatory network of BACH1, we performed in silico TF perturbation analysis using SCENIC+. In the simulated gene expression matrix generated based on the BACH1 perturbation, the expression of M1 phenotype related pathway genes was downregulated (Figure S6F). Following BACH1 perturbation, UMAP shift was observed in several cells, particularly toward M1-like cMono1 to M1-like cMono2 and CD163+ cMono (Figure S6G). In the M1-like cMono2, the activity of CEBPB/D, the TFs associated with inflammatory stimulation and induction, was prominently elevated.⁵⁹ Consistently, TF motif activity, chromatin accessibility at target regions, and expression of downstream target genes were all increased in M1-like cMono2 (Figure S7A; Table S7). Among the downstream targets, 83.6% of CEBPB targets and 94.9% of CEBPD targets overlapped with DEGs in M1-like cMono2 (Figure S7B). Notably, 31 genes were identified as co-regulated targets, and GO analysis of these genes revealed enrichment for inflammatory response and defense-related processes (Figure S7C). In addition, we confirmed that the expression of these genes was reduced in M1-like cMono2 upon in silico perturbation of CEBPB/D, and that they were shifted from M1-like cMono2 to CD163+ cMono (Figures S7D and S7E). These results indicate that each sub-cluster possesses distinct key gene regulatory mechanisms accompanied by dynamic epigenomic changes, which are also associated with severity and specific cellular functions in each sub-cluster.

Figure 3.

Dynamic and severity-related gene regulatory networks of cMono sub-clusters

(A) Motif enrichment and activity of each cMono sub-cluster. Dot plot showing top TF motifs enriched in the top 1,500 differentially accessible regions (DARs) for each sub-cluster (Left). Dot size indicates statistical significance (-log10 adjusted p by Fisher’s exact test and Benjamini-Hochberg) and color indicates fold enrichment. Heatmap displaying motif activity scores calculated by ChromVAR (right).

(B) Scatterplots showing the correlation between TF motif activity and their corresponding gene expression in severity-associated cMono sub-clusters (left: IL7R + cMono, right: CD163+ cMono). Each dot represents a TF, with colored dots indicating TFs that show both significantly increased motif activity and upregulated expression (adjusted p < 0.05 by Wilcoxon Rank-Sum test and Benjamini-Hochberg).

(C) This heatmap displaying the specific activator regulons of cMono sub-clusters. Each row represents a TF regulon, while each column corresponds to a specific cMono sub-cluster. The color represents the correlation between the TF and its target genes. The dot size indicates the correlation between the TF and its binding regions.

(D) This heatmap shows the activity of selected TF regulons across COVID-19 severity groups, ranging from healthy controls (HC) to critical stage. Each row represents an activator regulon, with the number of target genes indicated in parentheses (e.g., 129g: 129 genes). The color scale indicates the eRegulon score within each severity stage group.

(E) Dot plot showing motif activity, TFs binding region accessibility, and target gene expression of activator regulons in severity-associated cMono sub-clusters (IL7R+ and CD163+ cMono). The left panel (“motifs”) displays regulon activity based on motif enrichment scores, the middle panel (“regions”) indicates chromatin accessibility at predicted TF binding regions, and the right panel (“genes”) shows average expression levels of target genes regulated by each TF.

IL7R + classical monocytes exhibit T cell-like signaling and ETS1-driven regulatory programs

Gene set enrichment analysis (GSEA)⁶⁰ revealed significant enrichment of gene sets associated with cellular metabolism and survival, such as MYC targets, oxidative phosphorylation, and fatty acid metabolism, in IL7R + cMono (Figure S8A). Several signaling pathways were also highly enriched, including the MTORC1, IL2-STAT5, and PI3K/Akt pathways. Additionally, GO term analysis using IL7R + cMono DEGs confirmed the GSEA results that the module score of IL2 signaling was elevated in IL7R + cMono, along with an increase in IL4 and IL7 receptor signaling (Figure 4A). Consistent with the increased expression of CD247 and LTB observed in IL7R + cMono (Figure 2B), the module score for TCR signaling was markedly higher than that in other clusters. The TCR signaling module score in IL7R + cMono was comparable to that in T lymphocytes (Figure 4B). The expression of the components of the TCR complex was elevated, along with an increase in the expression of JAK-STAT pathway genes and interleukin receptors (Figure 4C). The expression and motif activity of NFAT and RUNX family members, ETS1, which are downstream TFs of TCR signaling,^61,62,63 and STAT5B and GATA3, which are downstream TFs of interleukin receptors signaling,^64,65 were all significantly increased in IL7R + cMono (Figures 4C and 4D). ETS1, which is known to be activated by TCR signaling, showed the highest correlation between its own expression and motif activity and was also identified as a top regulon in IL7R + cMono (Figures 3B and S5E). The expression of target genes and the motif activity of ETS1 were increased in IL7R + cMono (Figures 4E and S8B). ETS1 acted as an activator of 898 genes, regulating them through an average of two binding regions per gene (1,825 regions). These binding regions exhibited high chromatin accessibility in IL7R + cMono (Figure S8C). More than 50% of ETS1 binding regions were located adjacent to the TSS (Figure S8D). Moreover, 85.6% of the target genes and 85.0% of the binding regions overlapped with the DEGs and DARs of IL7R + cMono (Figure S8E; Table S8). To understand the regulatory network of ETS1, we performed in silico TF perturbation analysis using SCENIC+. In the simulated gene expression matrix generated based on the ETS1 perturbation, the expression of IL7R+ cMono-specific gene sets was downregulated (Figure 4F). Upon ETS1 perturbation, only IL7R + cMono cells showed a shift in UMAP, with a particularly prominent movement toward M1-like cMono2 (Figure 4G). Conversely, in silico ETS1 overexpression induced a dynamic shift from M1-like cMono1 to IL7R + cMono (Figure S8F). These shifts suggest that ETS1 acts as a key regulator of gene expression profiles in IL7R + cMono and a driver of differentiation from M1-like cMono2 to IL7R + cMono. Next, we determined chromatin accessibility patterns at the actual ETS1 binding regions that were identified by published ETS1 chromatin immunoprecipitation sequencing (ChIP-seq) data across the cMono sub-clusters (GEO: GSE138516).⁶⁶ IL7R + cMono had higher chromatin accessibility in IL7R + cMono than other sub-clusters (Figure 4H). In summary, our results demonstrate that TCR and interleukin receptor signaling are upregulated in IL7R + cMono, leading to the activation of ETS1. ETS1 acts as a core regulator of the IL7R+ cMono-specific transcriptional pattern, governing the cell type identity of IL7R + cMono.

Figure 4.

ETS1, a key regulator of the severity-associated sub-cluster IL7R + cMono

(A) Dot plot showing module scores for genes enriched in IL7R + cMono, associated with signaling-related Gene Ontology pathways Each dot is colored based on the average module score of its associated GO term, and its size reflects the percentage of sub-cluster cells expressing at least one gene from the term’s gene set.

(B) Comparison of module scores for the TCR signaling pathway gene set across T lymphocytes and cMono sub-clusters. The colored dashed lines within each violin plot indicate the mean module score of the respective cell population.

(C) Dot plot illustrating the expression patterns of genes involved in IL7R + cMono enriched signaling pathways across cMono sub-clusters. Genes are categorized into functional groups: T cell receptors (red), TCR signaling molecules (green), PI3K-Akt pathway (orange), interleukin receptors (purple), JAK-STAT pathway (cyan), and transcription factors (pink).

(D) Dot plot showing the motif activity of TFs associated with IL7R + cMono enriched signaling pathways across cMono sub-clusters. Dot color indicates the average scaled motif activity, while dot size represents the percentage of cells within each sub-cluster that activity the motif.

(E) Kernel density estimation (KDE) plot illustrating the relationship between ETS1 motif activity and the expression of its predicted target gene module across cMono sub-clusters. Each color corresponds to a distinct cMono sub-cluster. Horizontal dashed lines represent the mean motif activity of each sub-cluster.

(F) Line plot showing predicted expression change of IL7R + cMono-specific gene sets across perturbation iterations in IL7R + cMono. Each colored line represents an individual gene.

(G) UMAP embedding of cMono sub-clusters with simulated transcriptional dynamics following ETS1 perturbation. Arrows represent the predicted direction and magnitude of transcriptional state transitions after ETS1 perturbation. The arrow shading reflects the degree of simulated state change, with darker arrows indicating greater displacement.

(H) Genomic heatmaps displaying fragment counts at ETS1 binding sites cMono sub-clusters. Signal intensity reflects the density of chromatin accessibility, with darker red indicating higher enrichment.

Suppression of AP-1 regulon by JDP2 defines the immunoregulatory identity of CD163+ cMono

In CD163+ cMono, TGF-β signaling increased (Figure 2G), and both the module score of the SMAD3 target genes and motif activity were elevated in CD163+ cMono (Figure S9A). Moreover, 77.8% of the SMAD3 target genes overlapped with the DEGs of CD163+ cMono (Figure S9B; Table S9). Through in silico SMAD3 perturbation, we observed a significant shift from CD163+ cMono to M1-like cMono (Figure S9C), accompanied by a marked reduction in the expression of SMAD3 target genes in CD163+ cMono (Figure S9D). In contrast to the SMAD3 regulon, positively regulated target genes of AP-1 family regulons were expressed at lower levels in CD163+ cMono than in the other cMono sub-clusters (Figure 5A). Interestingly, genes that are negatively regulated by FOS, FOSB, and JUN, which were identified as repressor regulons in CD163+ cMono by SCENIC+ (Figure S5D), were upregulated in this population (Figure 5A). GO enrichment analysis of AP-1 family target genes (Table S10) revealed that they were significantly associated with inflammatory response, innate immune response, and cytokine production, which are closely linked to the M1 phenotype. Furthermore, the module scores of these GO gene sets were the highest in M1-like cMono2 and lowest in CD163+ cMono (Figure 5B), which is consistent with the SCENIC+ analysis results. Furthermore, 89.1% of the negatively regulated target genes of AP-1 overlapped with DEGs in CD163+ cMono (Figure 5C). These results suggest the presence of a specific mechanism that interferes with the transcriptional activity of AP-1 family members in CD163+ cMono. Among the CD163+ cMono TFs identified in previous analyses, JDP2 is a member of the AP-1 family that forms dimers with other AP-1 members to suppress transcriptional activity.^67,68 AP-1 members showed the highest motif activity in CD163+ cMono, but only JDP2 showed increased expression in CD163+ cMono (Figure 5D). Members of the AP-1 family share highly similar motif sequences, allowing them to bind to the same chromatin regions (Figure 5E). Therefore, we performed a TF footprinting analysis on the JDP2 motif positions and observed higher occupancy scores in CD163+ cMono and M1-like cMono2 (Figure 5F). This increase in M1-like cMono2 may be attributed to the binding of other AP-1 family members rather than JDP2. To mimic the AP-1 repression mechanism of JDP2, we performed in silico TF co-perturbation analysis across all AP-1 family members. A remarkable shift from M1-like cMono2 to CD163+ cMono was observed (Figure 5G). Conversely, in silico TF co-overexpression of the AP-1 family (FOS, FOSB, and JUN) acting as repressors in CD163+ cMono induced a regression from CD163+ cMono to M1-like cMono2 (Figure S9E). To validate the role of JDP2 as an AP-1 repressor in monocytes, we knocked down JDP2 in THP-1 cells using shRNA and confirmed significantly reduced JDP2 transcript and protein levels compared with the control (Figure 5H). Next, we observed the changes in the expression of genes that were positively regulated by AP-1 following LPS and R848 stimulation. Importantly, JDP2 knockdown led to the upregulation of AP-1 positively regulated genes, including JUNB, an AP-1 family member (Figure 5I). Additionally, the expression of the M1 phenotype-associated genes, IFNGR1 and IFI35, increased. Taken together, the CD163+ cMono-specific TF JDP2 plays a central role in the transcriptional repression of the AP-1 regulon, suggesting its potential role in regulating monocyte polarization and immune responses in severe COVID-19.

Figure 5.

JDP2 acts as an AP-1 family inhibitor in the severity-related sub-cluster CD163+ cMono

(A) Dot plot showing the module scores of AP-1 family regulon target genes across cMono sub-clusters. Symbols (+/+) and (−/+) indicate activator and repressor regulons, respectively. Dot color represents the average module score of target genes within each regulon, while dot size reflects the percentage of cells in the sub-cluster expressing at least one gene from the corresponding gene set.

(B) Dot plot summarizing the GO term analysis of AP-1 family activator regulon target genes. The left panel displays the top GO biological processes enriched among the AP-1 activator regulon target genes. Dot size indicates the number of genes associated with each GO term, and color represents statistical significance (-log10 adjusted p by Fisher’s exact test and Bonferroni). The right panel shows module scores of genes included in each GO term across cMono sub-clusters.

(C) Venn diagram showing overlap between AP-1 regulons negative regulated genes and DEGs of CD163+ cMono.

(D) Dot plot showing AP-1 family motif activity (Left) and their own expression (Right) across cMono sub-clusters. The left panel visualizes motif activity scores derived from chromatin accessibility data. The right panel shows the AP-1 family’s own expression.

(E) Motif plots of FOS, JUNB, and JDP2.

(F) Foot printing analysis of the JDP2 motif across cMono sub-clusters. The top panel shows Tn5 insertion enrichment centered around the JDP2 binding motif, with each line representing a distinct sub-cluster. The bottom panel displays the expected Tn5 insertion profile as a background.

(G) UMAP embeddings of cMono sub-clusters simulating transcriptional dynamics after AP-1 family perturbation. Arrows indicate the predicted direction and magnitude of transcriptional state transitions after AP-1 family perturbation. Arrow shading indicates the magnitude of the simulated state change, with darker arrows indicating larger displacements.

(H) JDP2 knockdown efficiency in THP-1 cells. Left: qPCR analysis shows significantly reduced JDP2 mRNA levels in shJDP2 cells compared to control (shCtrl) (n = 3, ∗∗adjusted p < 0.01 by two-sided Student’s t test and Bonferroni). Bars represent mean ± SD. Right: western blot confirms decreased JDP2 protein expression upon knockdown, with GAPDH as a loading control.

(I) Expression of AP-1 target genes in THP-1 cells following JDP2 knockdown. qPCR analysis shows significant upregulation of multiple AP-1 positively regulated genes in shJDP2 cells compared to shCtrl (n = 3, statistical significance: ∗ adjusted p < 0.05, ∗∗ adjusted p < 0.01, ∗∗∗ adjusted p < 0.001 by two-sided Student’s t test and Bonferroni). Bars represent mean ± SD.

Discussion

In this study, we investigated the gene expression changes and epigenomic alterations in immune cells associated with COVID-19 severity. Using a comprehensive approach to gene expression and chromatin accessibility, we identified significant changes in the immune populations at different stages of COVID-19 severity. Our findings highlight the important role of classical monocytes in mediating the severity and immune responses during COVID-19 infection. Through the sub-clustering of classical monocytes, we identified four distinct subpopulations. Each sub-cluster had different functional characteristics and was transcriptionally and epigenetically distinct, correlating with specific COVID-19 severity stages. Notably, IL7R + cMono and CD163+ cMono displayed unique GRNs and were highly enriched in the moderate and critical, or severe stages.

IL7R + cMono showed a unique gene expression profile, with significant upregulation of genes involved in TCR signaling and IL2/4/7-mediated immune responses. Surprisingly, several T cell-associated genes, including CD3E/D/G, CD247, and IL7R, were highly expressed in IL7R + cMono. These changes have also been observed in several studies on patients with COVID-19, rheumatoid arthritis, and LPS-stimulated monocytes.^17,40 IL7R^high monocytes exhibit a hypo-inflammatory phenotype, and STAT5, a downstream TF of IL7R signaling, has emerged as a key regulator. Similarly, we also observed increased STAT5 activity in IL7R + cMono, along with enhanced activation of downstream TFs of TCR and IL2/4/7 receptor signaling. These results suggest that IL7R + cMono adopt a T cell-like signaling pathway, which may influence immune regulation in COVID-19. ETS1 has emerged as a key transcriptional regulator of IL7R + cMono, directly controlling the expression of numerous target genes involved in immune regulation. Through in silico TF perturbation analysis, we demonstrated that ETS1 deficiency significantly altered the gene expression profile, shifting toward M1-like cMono, which has the M1 phenotype. This suggests that ETS1 is a key factor in monocyte differentiation and functional heterogeneity during COVID-19, particularly in the transition from a pro-inflammatory to a hypo-inflammatory state. Moreover, ETS1 directly regulated the expression of IL7R+ cMono-specific TFs, including LEF1 (Table S8), a key regulator of the increased Wnt signaling pathway in IL7R + cMono. The Wnt signaling pathway is known as the main signaling pathway for developmental processes, differentiation, renewal, and repair in various tissues and immune cells.^69,70 Previous studies have shown that Wnt ligands are upregulated in monocytes/macrophages of patients with COVID-19 and acute respiratory distress syndrome caused by COVID-19.^71,72 The Wnt signaling pathway triggers macrophages to differentiate into the M2 phenotype in idiopathic pulmonary fibrosis and lung infection.^73,74 Furthermore, activation of the Wnt signaling pathway in M2 phenotype macrophages promotes myofibroblast differentiation and pulmonary fibrosis.⁷⁵ The expression of TGFB1, an additional factor associated with fibrosis, was increased in IL7R + cMono. TGF-β1 is a key factor in fibrosis and wound healing. Previous studies have shown that it induces persistent fibrosis through fibroblast activation in various tissues, including lung tissue, leading to chronic sequelae after COVID-19 infection.^76,77 Moreover, increased TGFB1 expression in monocytes/macrophages has been observed in patients with severe COVID-19.⁴² Additionally, we confirmed that KLF2/12, ZBTB7A, and NFATC3, which are upregulated by ETS1, are TFs involved in the positive regulation of TGFB1 in IL7R + cMono. In summary, the hypo-inflammatory phenotype, Wnt signaling activation, and TGF-β secretion induced by ETS1 can reduce the excessive inflammation triggered by COVID-19 infection, but they may also drive excessive fibrotic processes.

In contrast, CD163+ cMono showed the typical gene expression profile of M2-like monocytes/macrophages associated with tissue remodeling, fibrosis, and anti-inflammatory responses. CD163^high monocytes have been observed in lung tissue and blood of severe COVID-19, COVID-19 acute respiratory distress syndrome, and idiopathic pulmonary fibrosis and have been proposed as a severity marker for COVID-19.^18,42,78,79 Similarly, CD163+ cMono was enriched in severe and critical stages, and functional enrichment analysis revealed an increase in pro-fibrotic phenotypes, characterized by elements of TGF-β and fibroblast growth factor signaling pathways. Additionally, SMAD3, a downstream regulator of TGF-β signaling and known to regulate the transition to the M2 phenotype, was highly activated in CD163+ cMono. The upregulated secretion of TGF-β in IL7R + cMono may have also contributed to the expansion of the M2-like monocytes (CD163+ cMono). Furthermore, the target genes of the AP-1 family, which are key regulators of M1 polarization, were downregulated in CD163+ cMono. Although FOS, FOSB, and JUN act as repressors of CD163+ cMono, their negatively regulated target genes were upregulated in CD163+ cMono. To explain this, we focused on JDP2, a known repressor of the AP-1 family.^80,81 JDP2 binds to the consensus sequence 5′-TGAG/CTCA-3′, similar to other AP-1 family members, thereby inhibiting AP-1-mediated transcriptional activation. Additionally, it suppresses AP-1 activity by forming heterodimers with members of the Jun, Fra, and ATF families. Consistent with JDP2-mediated AP-1 inhibition, in silico perturbation of AP-1 family members resulted in a drastic shift toward CD163+ cMono. Furthermore, among the AP-1 family members, only JDP2 increased both motif activity and expression in CD163+ cMono. Validation using JDP2 knockdown THP-1 monocyte cell lines revealed significant upregulation of AP-1 target genes and pro-inflammatory genes in JDP2 KD cells. Importantly, JDP2-mediated reduction of CD14 impairs intrinsic immunity mediated by monocytes/macrophages, preventing the recognition of LPS and PAMPs, which may increase the likelihood of persistent infection. Taken together, increased TGF-β signaling in CD163+ cMono promoted recovery, wound healing, and epithelial-mesenchymal transition. Furthermore, in CD163+ cMono, JDP2 sustained the M2 phenotype by inhibiting AP-1, which not only contributed to excessive fibrosis but also suppressed antiviral function, potentially leading to the accumulation of SARS-CoV-2.

Collectively, our study provides a detailed characterization of the monocyte heterogeneity associated with COVID-19 severity and identifies the subtypes that accumulate at different stages of severity. Additionally, we observed suppressed inflammatory responses and multiple fibrosis-related signaling pathways in severely affected patients, highlighting ETS1 and JDP2 as key transcriptional regulators. These findings offer insights into monocyte/macrophage-mediated tissue fibrosis in COVID-19 and suggest potential therapeutic targets. In line with these observations, persistent transcriptional reprogramming has been reported in circulating monocytes and regulatory T cells of convalescent COVID-19 patients.⁸² By integrating single-cell transcriptomic and flow cytometric analyses of 100 individuals, they identified multiple monocyte and regulatory T cell clusters that retained frequencies and molecular signatures associated with the initial severity even months after recovery, suggesting a potential contribution of hematopoietic progenitor cell (HPC) priming. Similarly, durable epigenetic changes and hyperactivation in IL7R + classical monocytes from patients who recovered from severe COVID-19 were shown to persist for several months to up to one year.¹⁶ They further showed that such epigenetic alterations were transmitted to descendant innate immune cells through reprogramming of hematopoietic stem and progenitor cells (HSPCs). These findings underscore the importance of investigating whether the regulatory programs and cMono subtypes identified in our study persist beyond the acute infection phase, and whether epigenetic imprinting in HSPCs contributes to the long-term reprogramming of monocyte lineages in post-COVID-19 conditions.

Limitations of the study

This study provides meaningful insights by identifying two cMono subtypes associated with COVID-19 severity and their key regulators, ETS1 and JDP2. However, several limitations remain.

First, we did not identify the cytokine donors in the upstream signaling pathways that induced these phenotypes. Future analyses of cell-to-cell interactions based on blood or lung tissue data are needed to verify the infection- or severity-specific increase in cytokine signaling associated with key regulators to complement this.

Second, we did not experimentally recapitulate the differentiation of monocyte subtypes or validate the expression of IL7R and CD163 at the transcript and protein levels. Further studies using primary human monocytes or the THP-1 cell line are required to reproduce subtype differentiation and elucidate the mechanisms of key regulators through in vitro assays.

Finally, this study lacked follow-up of patients with COVID-19. In particular, it is necessary to investigate the sustained accumulation of IL7R+ and CD163+ cMono subtypes in severe cases, as well as their potential contribution to sequelae such as impaired viral defense due to immunosuppression and increased fibrosis.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jihwan Park (jihwan.park@gist.ac.kr).

Materials availability

The THP-1 shJDP2 cell line generated in this study is available from the corresponding author upon reasonable request.

Data and code availability

• •The processed single-cell multiome data generated in this study have been deposited in the Gene Expression Omnibus (GEO) under accession number GEO: GSE297529 and are publicly available.
• •All code used for data processing and analysis, together with the required files to reproduce the results, is available on Zenodo: https://doi.org/10.5281/zenodo.15385493.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon reasonable request.

Acknowledgments

J.P. was supported by the Korea National Institute of Health (2022-ER1603-00, 2024-ER1605-00, and 2025-ER1606-00), Korea Basic Science Institute (National research Facilities and Equipment Center) grand funded by Ministry of Science and ICT (RS-2024-00403622), Korea-US Collaborative Research Fund (KUCRF) funded by the Ministry of Science and ICT and Ministry of Health & Welfare, Republic of Korea (grant number: RS-2024-00466906), and the National Research Foundation of Korea (NRF) funded by the Korean government (RS-2024-00335026). E.T.K. was supported by grants from the National Research Foundation of Korea (NRF), funded by the Ministry of Science and ICT (RS-2024-00352590) and the Ministry of Education (RS-2023-00270936).

Author contributions

This study was designed by J.P. and E.T.K.; J.R.Y. collected PBMCs from COVID-19 patients.; S.Y.C. performed the single-cell multi-omics sequencing library construction.; Data analysis was performed by D.K.; J.P. guided the process of data analysis.; C.Y.K. performed the experiments under the supervision of E.K.; J.P. and D.K. wrote the original draft of the manuscript. All authors read and approved the final manuscript.

Declaration of interests

S.Y.C. is an employee of Eyeoncell, and J.P. holds a leadership position at Eyeoncell.

Declaration of generative AI and AI-assisted technologies in the writing process

No generative AI or AI-assisted technologies were used in the writing or preparation of this manuscript.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit polyclonal anti-JDP2	Abcam	Cat# ab40916, RRID:AB_943893
Mouse monoclonal anti-GAPDH	Santa Cruz Biotechnology	Cat# sc-365062, RRID:AB_10847862
Peroxidase AffiniPure Goat Anti-Rabbit IgG (H+L)	Jackson Laboratories	Cat# 111-035-045, RRID: AB_2337938
Peroxidase AffiniPure Goat Anti-Mouse IgG (H+L)	Jackson Laboratories	Cat# 115-035-003, RRID: AB_10015289
Biological samples
Human Peripheral Blood Mononuclear Cells from COVID-19 patients and Healthy	Jenu National University Hospital	IRB# JEJUNUH 2022-09-011-001
Chemicals, peptides, and recombinant proteins
Lipopolysaccharide (LPS)	InvivoGen	Cat# tlrl-b5lps
R848 (Resiquimod)	InvivoGen	Cat# tlrl-r848-1
SuperSignal West Pico PLUS Chemilumiescent Substrate	Thermo Fisher Scientific	Cat# 34577
jetOPTIMUS DNA Transfection Reagent, jetOPTIMUS® Buffer	Polyplus	Cat# 101000006, 201000001
NUPAGE™ LDS Sample Buffer (4X)	Invitrogen	NP0007
Critical commercial assays
Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Kits	10x Genomics	Cat# PN-1000283
Chromium Next GEM Chip J Single Cell Kit	10x Genomics	Cat# PN-1000234
Library Construction Kit	10x Genomics	Cat# PN-1000190
Library Construction Kit B	10x Genomics	Cat# PN-1000279
Monarch® Total RNA Miniprep Kit	NEB	Cat# T2010S
TOPscript™ RT DryMIX (dT18)	Enzynomics	Cat# RT200
TOPreal™ SYBR Green qPCR PreMIX	Enzynomics	Cat# RT500M
Deposited data
Processed single-cell multiome data	GEO (This study)	GEO: GSE297529
Processed single-cell RNA-seq data	GEO	GEO: GSE145926
Processed single-cell RNA-seq data	GEO	GEO: GSE174072
Processed single-cell RNA-seq data	GEO	GEO: GSE264196
Processed single-cell RNA-seq data	GEO	GEO: GSE206284
ChIP-seq data	GEO	GEO: GSE138516
Experimental models: Cell lines
THP-1	ATCC	Cat# TIB-202, RRID: CVCL_0006
THP-1 shCTRL	This paper	N/A
THP-1 shJDP2	This paper	N/A
293FT	Invitrogen	Cat# R70007, RRID: CVCL_6911
Software and algorithms
Cell Ranger ARC (v2.0.1)	10x Genomics	https://www.10xgenomics.com/support/software/cell-ranger-arc/
R (v4.3.1)	R Core Team	https://www.R-project.org
Seurat (v5.0.3)		https://satijalab.org/seurat/
Signac (v1.13.0)		https://stuartlab.org/signac/
Demuxlet		https://github.com/statgen/demuxlet
DoubletFinder (v2.0.4)		https://github.com/chris-mcginnis-ucsf/DoubletFinder
scDblFinder (1.17.2)		https://github.com/plger/scDblFinder
Harmony (v1.2.0)		https://github.com/immunogenomics/harmony
MACS2 (v2.2.9.1)		https://pypi.org/project/MACS2/
MiloR (v1.10.0)		https://github.com/MarioniLab/miloR
genekitr (v1.2.5)		https://github.com/GangLiLab/genekitr
ChIPseeker (v1.38.0)		https://github.com/YuLab-SMU/ChIPseeker
chromVAR (v1.24.0)		https://github.com/GreenleafLab/chromVAR
CellPhoneDB (v5.0.0)		https://github.com/Teichlab/cellphonedb
SCENIC+ (1.0a1)		https://github.com/aertslab/scenicplus
Other
6well Plate	SPL	Cat# 30006
RPMI-1640 medium	WELGENE	Cat# LM011-03
QuantStudio 1 Real-Time PCR system	Thermo Fisher Scientific	Cat# A40426
Chemiluminescent Imaging System	Azure Biosystems	Model c300

Experimental model and study participant details

Human specimens

Peripheral blood samples were collected from adult COVID-19 patients hospitalized at Jeju National University Hospital (Jeju, South Korea). All participants provided written informed consent prior to sample collection. The study protocol was approved by the Institutional Review Board of Jeju National University Hospital (IRB No. JEJUNUH 2022-09-011-001). Sex and age information of all human participants are provided in Table S1. No significant differences associated with sex were observed in the analyses performed in this study. Patient samples were classified into four severity groups based on the WHO Clinical Progression Score. The WHO COVID-19 Clinical Progression Scale is divided into 10 scores. Generally, the mild stage (scores: 1-3) includes patients with symptoms such as fever, cough, or shortness of breath, managed in an outpatient setting. The moderate stage (scores: 4-5) includes hospitalized patients who require medical care but do not need oxygen therapy. The severe stage (scores: 6-7) involves patients requiring low-flow oxygen therapy and additional medical treatments. Finally, the critical stage (scores: 8-9) includes patients hospitalized in an intensive care unit (ICU) who require mechanical ventilation.

Peripheral blood samples from COVID-19 patients

Whole blood (4–6 mL per sample) was collected into EDTA tubes and processed within 2 hours of collection. Peripheral blood mononuclear cells (PBMCs) were isolated using density gradient centrifugation with Histopaque®-1077 (Sigma-Aldrich, Cat. No. 10771). Briefly, whole blood was diluted 1:1 with pre-warmed Dulbecco’s phosphate-buffered saline (DPBS) and layered over 15 mL of Histopaque in 50 mL conical tubes. Samples were centrifuged at 800 × g for 15 minutes at room temperature with no brake or acceleration. The mononuclear cell layer was collected, washed with 10 mL PBS (400 × g, 10 minutes), followed by a second wash with 5 mL DPBS (400 × g, 5 minutes). The final PBMC pellet was resuspended in cryopreservation medium consisting of 90% newborn calf serum (NBCS) and 10% dimethyl sulfoxide (DMSO), and stored under controlled freezing conditions for subsequent analysis.

Method details

Preparation of human peripheral blood mononuclear cells (PBMCs) from COVID-19 patients

PBMCs were isolated by density gradient centrifugation using Histopaque®-1077 (Sigma-Aldrich, Cat. No. 10771). All reagents were equilibrated to room temperature before use. Whole blood (4–6 mL per sample) was diluted 1:1 with pre-warmed Dulbecco’s phosphate-buffered saline (DPBS) and gently layered over 15 mL of Histopaque in 50 mL conical tubes. Samples were centrifuged at 800 × g for 15 minutes at room temperature without brake and acceleration. The mononuclear cell layer at the interface was collected using a sterile Pasteur pipette and transferred to a new 50 mL tube. Cells were washed with 10 mL of PBS and centrifuged at 400 × g for 10 minutes at room temperature. A second wash was performed with 5 mL of DPBS followed by centrifugation at 400 × g for 5 minutes. The final PBMC pellet was resuspended in cryopreservation medium containing 90% newborn calf serum (NBCS) and 10% dimethyl sulfoxide (DMSO) and stored under controlled freezing conditions for subsequent analysis.

Single-cell multi-omics library preparation

The libraries were generated using the Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Kits (PN-1000283) (10x Genomics) in accordance with the manufacturer's guidelines. In summary, nuclei were incubated with Transposition Mix at 37°C for 60 minutes. The resulting mixture was loaded onto Chromium Next GEM Chip J Single Cell Kit (PN-1000234, 10x Genomics), and GEMs (Gel Beads-in-Emulsions) were formed following the manufacturer’s protocol. After GEM processing and cleanup, barcoded transposed DNA and barcoded full-length cDNA fragments underwent pre-amplification via PCR. For library construction, ATAC libraries were prepared using the Library Construction Kit (PN-1000190), while full-length pre-amplified cDNA fragments underwent additional PCR amplification. RNA libraries were assembled using the Library Construction Kit B (PN-1000279). Finally, the completed libraries were subjected to sequencing on an Illumina NovaSeq platform.

Preprocessing of single-cell multi-omics data and quality check for obtaining high-quality cells

Raw sequencing data were processed by 10x genomics Cell Ranger-arc⁸³ version 2.0.1 (https://www.10xgenomics.com/support/software/cell-ranger-arc/latest/release-notes/release-notes). FASTQ files were aligned to the GRCh38 human reference genome, and expression, fragment count matrices were generated. The two count matrices were created as individual objects following the 10x multi-omics process of Seurat⁸⁴ (v.5.0.3) and Signac⁸⁵ (v.1.13.0) in R (v.4.3.1). To remove low-quality cell and dying cells and obtain high-quality cells, we filtered cells based on the number of unique genes (200-2,500 genes), total UMI/fragments counts (1000-25,000 RNA counts and 1000-100,000 ATAC counts), and mitochondrial read percentage (less than 50%), as described in previous studies.^22,86,87 Furthermore, we calculated the strength of the nucleosome signal and transcriptional start site enrichment score for each cell using fragment matrix, and obtained cells with nucleosome signal < 2 and TSS enrichment score > 4.

Demultiplexing of donors using genetic variants and doublet filtering

To assign individual cells to their original donors in pooled single-cell multi-omics data, we utilized Demuxlet.⁸⁸ We sorted SNPs with a minor allele frequency of at least 5% from bulk DNA genotype data (VCF file) containing individual genetic variants using bcftools.⁸⁹ Next, we compared the GEX and ATAC BAM files and the sorted SNPs (MAF ≥ 0.05) to determine the donor origin of each cell and detected and removed doublets. Furthermore, doublet removal was also performed through DoubletFinder⁹⁰ and scDblFinder⁹¹ based on the transcriptomic and epigenetic profiles of each pooling object. All pooling objects containing high-quality singlets were merged into a single object.

Dimension reduction and batch effect correction

Gene expression data were processed using principal component analysis (PCA) after SCTransform normalization, while chromatin accessibility data were processed using latent semantic indexing (LSI) following term frequency-inverse document frequency (TF-IDF) transformation. To integrate data from multiple samples and correct for batch effects, Harmony⁹² was applied to PCA for GEX data and to LSI for ATAC data. To integrate multimodalities into a shared low-dimensional space, we applied the weighted nearest neighbor (WNN) method to identify the closest neighbors for each cell based on a weighted combination of the two modalities. After constructing a joint multimodal neighbor graph, dimensionality reduction was performed using uniform manifold approximation and projection (UMAP) (RNA dimensions = 1:50, ATAC dimensions = 2:50). The clustering was based on the weighted shared nearest neighbor (WSNN) graph (resolution = 0.5). Consequently, 21 clusters were identified, and a total of 16 cell types were annotated based on the expression of well-known marker genes.

Peak calling

Peak calling was performed using the “CallPeaks” function in the Signac package, which utilizes MACS2⁹³ (Model-based Analysis of ChIP-Seq) to identify cell type-specific regions of accessible chromatin. Pseudo-bulk peak calling was conducted by merging fragments from all cells and running MACS2 with the following default parameters. The identified peaks were filtered and combined based on enrichment and reproducibility across cells. As a result, a total of 112,124 peaks were identified across 16 cell types, and these were used to construct a chromatin accessibility matrix for downstream analyses, including differential accessibility testing and motif enrichment analysis.

Cell type and RNA-ATAC correlation

The gene activity matrix was calculated based on the chromatin accessibility of peaks mapped to the gene body and promoter regions. This matrix was normalized and log-transformed to compare with RNA expression data. The correlation between RNA and ATAC for each cell type was assessed by constructing gene expression and gene activity matrices through the summation of raw counts for 2,000 variable features. These matrices were normalized using Z-scores, and Pearson correlation coefficients were computed between the normalized RNA expression and gene activity matrices.

Differential abundance analysis of KNN graph in COVID-19 infection

The identification of differentially enriched cell populations according to COVID-19 infection was analyzed using the MiloR²⁷ framework. MiloR uses the KNN (k-nearest neighbor) graph to define neighborhoods (Nhoods) in single-cell data. First, a KNN graph was constructed from the single-cell dataset, where each node represents a cell and edges connect similar cells based on multimodal profiles. Next, cell counts within each neighborhood were modeled using a generalized linear model (GLM) framework to assess differential abundance. A statistical test was performed to compare neighborhood-wise cell proportions between condition, and multiple testing correction was applied to control the false discovery rate. Finally, significant neighborhoods were visualized on a UMAP embedding.

Gene ontology term and module score analysis

To assess the functional activity of specific Gene Ontology (GO)⁹⁴ terms at the single-cell level, a predefined set of GO terms was selected and mapped to corresponding gene sets using the org.Hs.eg.db database. The extracted gene sets were formatted as a list and used as input for “AddModuleScore”, which calculates a module score for each cell based on the expression levels of genes within the set.

Cell-cell interaction analysis using CellphoneDB

To identify ligand-receptor interactions across cell types according to severity, we performed cell-cell communication analysis using CellPhoneDB (v5.0.0).⁹⁵ For each severity group (Healthy, Mild, Moderate, Severe, and Critical), single-cell expression data and the corresponding metadata were analyzed separately. The statistical analysis method implemented in CellPhoneDB was applied to identify statistically significant ligand-receptor pairs enriched between specific cell type pairs based on permutation testing (iterations: 1000, threshold ≥10% of cells expressing the specific ligand/receptor, P-value threshold <0.05).

Gene set enrichment analysis

To perform Gene Set Enrichment Analysis⁶⁰ (GSEA), we retrieved gene sets for human biological pathways from the MSigDB database. Differentially expressed genes were filtered for a specific group and ranked based on their area under the curve (AUC) values. The ranked gene list was formatted with gene names mapped to their respective AUC scores. Significantly enriched pathways were identified using the “genGSEA” function in the genekitr⁹⁶ package.

Genomic annotation of DARs

To annotate the differentially accessible regions (DARs) of cMono sub-clusters on the genome, sub-cluster DAR files were loaded and annotated using ChIPseeker,⁹⁷ mapping peaks to genomic features within a ±3 kb region around transcription start sites (TSS) using TxDb. The distribution of annotated peaks was visualized with bar plots to assess feature composition.

Transcription factor (TF) binding motif enrichment and activity analysis

To identify transcription factor binding motifs enriched within differentially accessible regions (DARs), the top 1,500 DARs for each sub-cluster were selected and analyzed using the “FindMotifs” function in Signac. The number of overrepresented motifs within the given peak set was calculated and compared with the background peak set based on the hypergeometric test. Next, we performed chromVAR⁹⁸ analysis using the Signac package to assess transcription factor activity across sub-clusters based on ATAC-seq. The deviation matrix was computed by quantifying chromatin accessibility variability across sub-clusters using predefined motif position-weight matrices (PWMs) and was normalized for sequencing depth and GC content.

Gene regulatory network analysis in single-cell multi-omics data

To identify gene regulatory networks (GRNs in cMono sub-clusters, SCENIC+⁵⁷ was used to integrate single-cell transcriptomic and chromatin accessibility data. First, co-expression modules were identified to infer potential transcription factor (TF)-target relationships. Next, TF binding motifs and chromatin accessibility data were utilized to refine direct regulatory interactions within these modules. Finally, regulon activity scores were quantified using AUCell, enabling the identification of key TFs, their binding sites, and target genes within the inferred GRNs.

In silico TF perturbation and overexpression analysis

To evaluate the impact of TF activity within cMono sub-cluster-specific regulons, simulations of in silico TF perturbation and overexpression were performed. Specific TF expression within the regulons was artificially minimized or maximized. Through five iterations, both direct and indirect regulatory effects were captured. After the simulation, the changes in regulon activity and target gene expression were assessed by comparing the simulated conditions with the previous conditions. Additionally, changes in gene expression matrices resulting from TF perturbation or overexpression were embedded in UMAP to identify alterations in cell clusters or states.

Analysis of fragment counts in ETS1 binding regions (ChIP-seq)

The ETS1 ChIP-seq dataset (GSE138516⁶⁶) for THP-6 cells was used to identify ETS1 binding sites across the genome. The number of fragments within 1,000bp upstream and downstream of these binding sites was counted per sub-cluster, creating a region matrix. The chromatin accessibility data within 20 windows of 100bp (total 2,000bp) was then visualized using the “RegionHeatmap” function from the Signac package to examine the patterns of chromatin accessibility across different cMono subtypes.

Short hairpin RNA (shRNA) mediated JDP2 knockdown in THP-1 cells

Lentiviral vectors encoding shRNAs targeting human JDP2 were purchased from Sigma-Aldrich (shRNA ID: TRCN0000019001). A non-targeting shRNA control vector was also obtained from Sigma-Aldrich (SHC016). Lentiviral particles were produced in 6-well plates by co-transfecting 293FT cells with 1 μg of the transfer plasmid, 0.75 μg of the packaging plasmid pCMV-ΔR8.74, and 0.25 μg of the envelope plasmid pMD.G2, using 2 μL of jetOPTIMUS DNA Transfection Reagent (Polyplus, Cat. No. 101000006) in jetOPTIMUS® Buffer (Polyplus, Cat. No. 201000001). Supernatants were collected at 48 hours post-transfection, passed through a 0.45 μm syringe filter, and used directly for transduction. THP-1 cells were transduced with the filtered lentiviral supernatant by spin infection (800 × g, 50 min, 35°C) in the presence of 10 μg/mL polybrene. After 24 hours, cells were selected with 1 μg/mL puromycin (Santa Cruz, SC-1342320) to establish stable knockdown lines. Knockdown efficiency was validated by immunoblotting and quantitative RT-PCR (see section “RNA extraction and qPCR”). For immunoblotting, whole-cell lysates were prepared in NuPAGE™ LDS Sample Buffer (Invitrogen, Cat. No. NP0007) supplemented with β-mercaptoethanol. Equal amounts of protein were resolved on SDS-PAGE gels and transferred to nitrocellulose membranes (Whatman, E06-07-116). Membranes were probed with anti-JDP2 (Abcam, Cat. No. ab40916) and anti-GAPDH (Santa Cruz, sc-365062) antibodies, followed by HRP-conjugated secondary antibodies. Signals were detected using SuperSignal™ West Pico PLUS Chemiluminescent Substrate (Thermo Scientific, Cat. No. 34577) and visualized with a chemiluminescence imaging system (Azure Biosystems, Model c300).

THP-1 cell stimulation assay

THP-1 cells were seeded at a density of 0.3 × 10⁶ cells per well in a 6-well plate (SPL, Cat. No. 30006) and incubated for 24 hours at 37°C in a 5% CO₂ incubator in RPMI-1640 medium (WELGENE, Cat. No. LM011-03). Following incubation, the cells were stimulated with LPS (1 μg/μL, InvivoGen, Cat. No. tlrl-b5lps) and R848 (10 μg/μL, InvivoGen Cat. No. tlrl-r848-1) for 6 hours. A control group (n = 3) and a stimulation group (n = 3) were included to assess the LPS and R848 stimulation effects.

RNA extraction and qPCR

After stimulation, cells were harvested by centrifugation at 400 × g for 5 minutes. Next, total RNA was extracted using the Monarch® Total RNA Miniprep Kit (NEB, Cat. No. T2010S). For cDNA synthesis, 300 ng of total RNA per sample was reverse transcribed using TOPscript™ RT DryMIX (dT18) (Enzynomics, Cat. No. RT200) at 50°C for 1 hour and 95°C for 5 minutes. Quantitative real-time PCR (qPCR) was performed using TOPreal™ SYBR Green qPCR PreMIX (Enzynomics, Cat. No. RT500M) on a QuantStudio 1 Real-Time PCR system (Thermo Fisher Scientific). The relative gene expression levels were analyzed to assess the effects of stimulation.

Quantification and statistical analysis

All statistical details are described in the figure legends and in the STAR Methods.

Published: January 29, 2026