In this study, patient samples were collected from critically ill individuals aged 60 years or older who were undergoing treatment in the emergency intensive care units (ICUs) at Zhongshan Hospital and Minhang Hospital, both affiliated with Fudan University, China. We prospectively enrolled 17 patients diagnosed with sepsis-induced ARDS, following the diagnostic criteria established in the 2016 international Sepsis 3.0 consensus proposed by sepsis experts and the Berlin definition and diagnostic criteria for ARDS (9, 10). These patients did not have SARS-CoV-2 infection. Table S1 presents an overview of the demographic and clinical features of the study population. For three of these patients, blood samples were collected on Day 1 and Day 7 and subsequently subjected to scRNA-seq. Based on their clinical manifestations, Day 1 and Day 7 were designated as the onset and recovery phases, respectively. For additional information, please refer to Table S2.

To investigate the shared genetic correlations between COVID-19 and sepsis, we utilized RNA-seq and microarray datasets from the Gene Expression Omnibus (GEO) database. Specifically, we analyzed the GSE171110 dataset, which comprises whole-blood RNA-seq profiles from COVID-19 patients and healthy donors, and the GSE137342 dataset, which includes whole blood cells from sepsis patients and healthy volunteers and was sequenced using microarrays (11). Further information regarding the datasets is outlined in Table S3.

DEGs were determined from expression values utilizing the “limma” package in R software (version 4.2.0), applying Benjamini-Hochberg correction to regulate the false discovery rate. The DEGs were considered significant if they met the cutoff criteria (adjusted P-value < 0.05 and |logFC| ≥ 1.0). The common DEGs between COVID-19 and sepsis were identified using an online VENN analysis tool called Jvenn (12).

Utilizing the “clusterProfiler” package in R, potential functions and pathways associated with DEGs were identified. GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were performed, and a standardized metric (P-value < 0.05, Q-value < 0.25) was used to prioritize the top functional items and pathways.

The PPI network was generated based on the proteins encoded by the common DEGs between COVID-19 and sepsis, as determined through the STRING database (13). The PPI network was further processed and analyzed using Cytoscape software, and gene clusters were identified using the Markov cluster method. The prominent nodes in the PPI network modules were predicted using the CytoHubba plugin to identify hub genes.

Hub gene-microRNA (Hub-miRNA) interaction networks and hub gene-transcription factor (Hub-TF) interaction networks were analyzed using the NetworkAnalyst tool (14). TarBase (15) and miRTarBase (16) databases were used to identify Hub-miRNA interactions, while the JASPAR database (17) was used to analyze Hub-TF interactions. The hub genes common to COVID-19 and sepsis were used in the gene regulatory networks analysis to identify the transcriptional elements and miRNA that regulate hub genes at the post-transcriptional level.

The composition of immune cells in blood samples was analyzed utilizing the CIBERSORT (18), a deconvolution algorithm designed to quantify the representation of 22 distinct subpopulations of infiltrating lymphocytes. GraphPad Prism 8.0.2 software was utilized to contrast immune cell proportions between the COVID-19 and healthy control samples, as well as between the sepsis and healthy control samples.

The Enrichr web server (19) and the DSigDB database (20) were used to identify pharmacological compounds connected to the common hub genes of COVID-19 and sepsis, based on a statistical threshold of adjusted P-value < 0.05.

The DisGeNET database (21) was utilized to investigate gene-disease associations and identify diseases and chronic issues linked to the common hub genes of COVID-19 and sepsis. The NetworkAnalyst tool (22) was also used in this analysis.

Peripheral blood samples were collected from each patient in a 5 mL EDTA tube on the day of enrollment and processed within two hours to isolate all peripheral blood mononuclear cells (PBMCs). The PBMCs were either subjected to scRNA-seq or cryopreserved and stored at -80°C for subsequent analysis.

Blood-derived PBMCs were subjected to single-cell gel bead-in-emulsion generation using the Chromium Controller Instrument (10×Genomics) and the Single Cell 3’ Library and Gel Bead Kit V3.1 (10×Genomics) according to the manufacturer’s recommended protocol. The resulting GEM libraries were then sequenced on an Illumina Novaseq 6000 using a custom paired-end sequencing mode of 150 bp for the first read and 150 bp for the second read. The raw data was processed using the Cell Ranger Single-Cell Software Suite (v5.0.0) with default parameters and aligned to the genomic reference. To maintain data quality, cells were discarded if they had gene counts below 200 or over 5000, or mitochondrial gene expression greater than 15%. The most variable genes among the single cells were identified, and their log-transformed gene-barcode matrices were subjected to principal component analysis to reduce their dimensionality. The resulting data was visualized in a t-SNE plot constructed with Seurat (v3.1.1) to differentiate among the PBMC cell types.

To induce sepsis-induced ARDS in mice, we used the cecal ligation and puncture (CLP) method (23, 24). In brief, mice underwent an 8-hour fast and 4-hour water deprivation before surgery. Under 2% isoflurane anesthesia, a sterile abdominal incision provided access to the cecum. The cecum was ligated 1 cm from its end, punctured twice with a 22-gauge needle, repositioned, and the incision closed in two layers. Control mice experienced a similar procedure without CLP. At 24 hours post-CLP, lungs were collected, and RNA was isolated from the tissue using a Tiangen kit. RNA was reverse transcribed with the Tiangen kit, and relative gene expression was determined via the ΔΔCt method, using actin as an internal control.

Data are displayed as mean ± standard error of the mean (SEM). Statistical analyses were executed using GraphPad Prism and R software. For normally distributed data, a two-tailed, unpaired Student’s t-test was applied to assess differences between groups. Non-normally distributed data were evaluated using the Mann-Whitney test. Correlation analyses were performed employing the Pearson method. Statistical significance was established for P-values below 0.05 (*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001).

To reveal shared genetic pathways between COVID-19 and sepsis, we comprehensively analyzed human transcriptomic data from the GEO database (Figure S1). Our results showed that COVID-19 and sepsis patients had 4082 and 551 DEGs, respectively, compared to healthy controls (Table S3). The DEGs of the highest significance for COVID-19 and sepsis are depicted in heatmaps and volcano plots in Figures 1A–D. We then utilized Jvenn to uncover 189 DEGs that were shared between the sepsis and COVID-19 datasets (Figure 1E). The complete list of these 189 DEGs can be found in Table S4.

To understand the functional significance of the shared DEGs between COVID-19 and sepsis, we conducted a functional enrichment analysis using the “clusterProfiler” package. The analysis consisted of GO enrichment, linking genes to GO terms, and KEGG pathway enrichment, which revealed gene-pathway associations. The GO analysis was categorized into biological process, cellular component, and molecular function (25). Our results showed that the DEGs were significantly enriched in defense against fungus in the biological process category, specific granule lumen in the cellular component category, and MHC class II receptor activity in the molecular function category (Figures 2A, B). The KEGG pathway analysis revealed the top 10 pathways, including asthma, inflammatory bowel disease, hematopoietic cell lineage, intestinal immune network for IgA production, legionellosis, staphylococcus aureus infection, leishmaniasis, Th1 and Th2 cell differentiation, systemic lupus erythematosus, and cytokine-cytokine receptor interaction (Figures 2C, D).

The shared DEGs between COVID-19 and sepsis were subjected to a PPI analysis using the STRING database, aiming to reveal functional networks and associated pathways. The PPI network is illustrated in Figure 3A, where highly interconnected nodes represent hub genes. The maximal clique centrality method of cytoHubba in Cytoscape was employed to identify the top 15 most influential DEGs, which include CD4, CD3E, IL7R, CD5, CD247, CD2, CCR7, CD40LG, ITK, KLRB1, MPO, MMP9, TLR2, LCN2, and RETN. These hub DEGs hold potential as biomarkers and therapeutic targets for COVID-19 and sepsis, presenting novel opportunities for therapeutic intervention. A submodule network was constructed to facilitate a better understanding of the relationships and positions of these genes, as shown in Figure 3B.

To assess the potential of hub genes as biomarkers for predicting COVID-19 and sepsis-induced ARDS, we employed a murine model of sepsis-induced ARDS as a proxy for both COVID-19 and sepsis-related lung injuries. This choice was guided by the shared pathophysiological characteristics of ARDS in COVID-19 and sepsis, and their substantial mechanistic overlap. The use of the murine model was also influenced by the absence of a biosafety level 3 facility in our laboratory, which precluded the development of a specific SARS-CoV-2 infection model. Our results showed that genes such as IFN-γ, TNF-α, IL-6, MIP-2, IL-10, KC (CXCL1), CCL-2, MPO, MMP9, TLR2, and LCN2 were significantly upregulated in the lungs of sepsis-induced ARDS mice compared to control mice. In contrast, genes such as CD247, CD2, CD40LG, KLRB1, and RETN were significantly downregulated (Figures 4A, B).

TNF-α, IFN-γ, and IL-6 represent vital inflammatory mediators in the progression of lung injury in both sepsis and COVID-19 scenarios (1, 26). Positive correlations were observed between MPO, TLR2, LCN2 and TNF-α, and between MPO, MMP9, LCN2 and IL-6. Conversely, negative correlations were observed between CD2 and IFN-γ, and between CD40LG and IL-6 (Figures 4C, D, S2A–C).

KC and CCL-2 serve as significant chemokines that attract neutrophils and monocytes to infection sites, thereby playing a pivotal role in the pathophysiology of COVID-19 and sepsis-induced lung injury (1, 27, 28). The results showed that MMP9 and LCN2 were positively correlated with KC, while CD40LG and KLRB1 were negatively correlated with KC. Furthermore, MPO, MMP9, and LCN2 were positively correlated with CCL-2, while CD40LG was negatively correlated with CCL-2 (Figures 4E, F, S2D, E).

Finally, we assessed the potential of the discovered hub genes as targeted biomarkers in elderly sepsis-induced ARDS. We investigated the expression levels of these hub genes in PBMCs collected from elderly patients with sepsis-induced ARDS. Our findings revealed significant decreases in CD4, CD3e, IL-7R, CD5, CD247, CD2, CD40LG, ITK, and KLRB1, and significant elevations in MMP9, LCN2, and RETN in sepsis-induced ARDS compared to controls (Figure 5A). Similar results were also observed in COVID-19 and sepsis (Figure S3). Moreover, we found that CD3, CD247, CD2, and CD40LG were negatively correlated with KC, whereas MMP-9, LCN2, and RETN were positively correlated with KC (Figure 5B), suggesting that these candidate hub genes may regulate geriatric sepsis-induced ARDS by modulating the recruitment of neutrophils to the lung.

In order to provide additional support for the involvement of hub genes in the development of sepsis-induced ARDS among elderly individuals, we executed scRNA-seq on PBMCs procured from geriatric patients diagnosed with sepsis-induced ARDS. Drawing upon the insights gleaned from our murine sepsis-induced ARDS model and human subjects afflicted with sepsis-induced ARDS, we selected CD247, CD2, CD40LG, KLRB1, LCN2, and RETN as the foci of our subsequent investigation. Our findings revealed that CD247, CD2, and KLRB1 were predominantly expressed in natural killer (NK) cells, CD4+ T cells, and CD8+ T cells, respectively; CD40LG was chiefly expressed in CD4+ T cells; LCN2 was primarily expressed in monocytes; and RETN was principally expressed in monocytes and DCs (Figures 6A, S4). Additionally, we detected elevated expression of CD247 in NK cells, CD4+ T cells, and CD8+ T cells on Day 7 (recovery phase) in comparison to Day 1 (onset phase). CD2 expression was heightened in CD8+ T cells on Day 7 relative to Day 1. CD40LG expression demonstrated a decline in CD8+ T cells on Day 7 compared to Day 1. KLRB1 expression exhibited an increase in NK cells and CD8+ T cells and a decrease in CD4+ T cells on Day 7 relative to Day 1. LCN2 expression witnessed an augmentation in monocytes on Day 7 in comparison to Day 1. Lastly, RETN expression manifested a reduction in monocytes and DCs on Day 7 as opposed to Day 1 (Figure 6B). These findings indicate that the six identified hub genes could be closely related to the development of sepsis-induced ARDS in older individuals, with their expression patterns and alterations potentially having a crucial impact on disease progression and recovery.

To delve deeper into the potential roles of the discovered hub genes in developing COVID-19 and sepsis, we examined the proportions of immune cells and their associations with these hub genes in patients affected by COVID-19 and sepsis. Our results revealed that memory B cells, CD8+ T cells, and CD4+ memory resting T cells were reduced in COVID-19, while plasma cells, CD4+ memory activated T cells, and neutrophils were increased when compared to healthy controls (Figures 7A, B). Similarly, memory B cells, CD8+ T cells, CD4+ memory resting T cells, and resting NK cells were decreased in sepsis, while gamma delta T cells, monocytes, M0 macrophages, activated DCs, and resting mast cells were increased when compared to healthy controls (Figures 7C, D). Neutrophil levels were also elevated in sepsis, although not to a statistically significant extent. Furthermore, our analysis showed strong associations between multiple immune cell types, including memory B cells, CD8+ T cells, CD4+ memory resting T cells, and neutrophils, in both COVID-19 and sepsis (Figures 7E, F). Specifically, in COVID-19, memory B cells, CD8+ T cells, and CD4+ memory resting T cells were negatively correlated with neutrophils (Figure 7E). In sepsis, CD8+ T cells and gamma delta T cells were negatively correlated with neutrophils (Figure 7F).

Subsequently, we employed the Pearson correlation coefficient to assess the association between immune cell abundance and hub gene expression in both COVID-19 and sepsis cases. Results revealed a negative correlation between neutrophils and CD247, CD2, and KLRB1, and a positive correlation with RETN in both COVID-19 and sepsis (Figures 8A, B). LCN2 was positively correlated with neutrophils in COVID-19 but not in sepsis cases (Figures 8A, B). In sepsis, monocytes were negatively correlated with CD247, CD2, and CD40LG, while positively correlated with LCN2 and RETN (Figure 8B). This correlation was weaker in COVID-19. CD247, CD2, CD40LG, and KLRB1 displayed a positive correlation with CD8+ T cells and CD4+ T cells in both COVID-19 and sepsis cases, while LCN2 and RETN showed a negative correlation (Figures 8A, B). These results demonstrate the key relationships between immune cells and hub gene expressions, highlighting differences and similarities in immune responses across COVID-19 and sepsis-induced ARDS cases.

To elucidate the regulatory molecules controlling the identified hub genes (including CD4, CD3E, IL7R, CD5, CD247, CD2, CCR7, CD40LG, ITK, KLRB1, MPO, MMP9, TLR2, LCN2, and RETN) at the transcriptional level, we utilized a network-based approach to identify the key TFs and miRNAs involved. The interaction between TF regulators and hub genes is depicted in Figure 9, while Figure 10 shows the interactions of miRNA regulators with the hub genes. Our analysis revealed a total of 62 TFs and 125 miRNAs that are potentially involved in regulating the hug genes, indicating a significant interplay between them. For more detailed information, Tables S5, S6 provide the regulatory network of target TF-genes and target miRNA-genes, as well as the topology table.

Protein-drug interaction analysis is critical for understanding the structural features of receptor sensitivity, which can aid in discovering new drugs (29). To identify potential therapeutic drugs for COVID-19, sepsis, and geriatric sepsis-ARDS, we focused on the hub genes common to these diseases. By utilizing the Enrichr tool and analyzing transcriptional characteristics from the DSigDB database, we identified ten candidate compounds. These drugs were selected based on their P-values and are listed in Table 1. Our findings suggest that these compounds may have promising therapeutic effects and could serve as the basis for developing new treatment options for these diseases.

The relationship between diseases can be established based on shared genetic factors, which is a crucial step toward developing effective treatments for various disorders (30). Using NetworkAnalyst, we thoroughly examined the associations between our identified hub genes and various disease states. Our analysis revealed that autosomal recessive predisposition, rheumatoid arthritis, ulcerative colitis, severe combined immunodeficiency, hepatomegaly, eosinophilia, COVID-19, and sepsis have the strongest connections with our reported hub genes. Figure 11 illustrates the gene-disease relationships, highlighting the potential correlations between these diseases. This emphasizes the potential impact of these diseases on each other and provides valuable insight into their complex relationships.