From emergence to January 2024, coronavirus disease 2019 (COVID-19) has led to over 7 million deaths worldwide (1). Acute respiratory distress syndrome (ARDS) is one complication of severe COVID-19 and contribute to COVID-19 related death (2, 3). Among hospitalized patients with COVID-19, nearly one-third develop ARDS (4). Autopsy studies of COVID-19 reveal evidence of ARDS in most decedents (5, 6). Epidemiologic data from 2020 suggests at least a five-fold increase in ARDS-related deaths during the height of the COVID-19 pandemic (7). ARDS is a form of acute, non-cardiogenic, hypoxemic respiratory failure, characterized by bilateral lung infiltrates (8) that accounts for 10% of ICU admissions (9) with mortality rates ranging from around 30-50% (9). ARDS is notoriously heterogeneous, affecting varied patient populations with lung injury from various causes. One current focus of ARDS research is the identification of distinct patient subgroups that display varied outcomes and responses to targeted therapies (10–13). While clinical trials have demonstrated the efficacy of corticosteroids as treatment in COVID-19 ARDS (14, 15), studies evaluating immune modulating therapies in diverse ARDS populations have yielded mixed results (16, 17). Despite decades of clinical trials in ARDS, supportive care remains the primary treatment approach in diverse ARDS populations (18), and effective, targeted therapeutics for ARDS are lacking.

A better understanding of ARDS pathobiology may inform the limitations of current treatments, and the COVID-19 global pandemic has renewed the importance and urgency of new approaches to ARDS research. Recent studies have employed molecular profiling tools, such as RNA-sequencing, single-cell RNA-seq, and proteomics, with varied study design to better understand gene signatures that correlate with disease severity among COVID-19 ARDS patients. In this respect, multiple studies have highlighted the importance of type I interferon signaling and acute pro-inflammatory mediators in peripheral blood mononuclear cells (PBMCs) collected from patients early in disease course that correlate with COVID-19 disease severity (19–21), suggesting that PBMC analysis provides valuable insight into the dysregulated biology of COVID-19. As COVID-19 is dynamic with an evolving disease course over hours and days, other groups have examined transcriptomic data from patients at multiple time-points with varied sampling schema, including defined clinical stages (treatment, convalescence, and rehabilitation) (20), early and late ICU time points (22), and sampling from admission through two months of follow-up (23). Optimal strategies for patient sampling during the course of COVID-19 ARDS illness and recovery remain unclear. Our approach of longitudinal patient sampling provides opportunities to identify mechanisms that may contribute to disease progression or resolution.

To better understand how dynamic gene expression correlates with clinical outcomes among a group of patients with COVID-19 ARDS, we performed RNA-sequencing of patient PBMCs collected at 4 fixed time intervals across ICU admission. We hypothesized that longitudinal analysis may identify distinct transcriptomic changes compared to analysis at single time points to better characterize dynamic processes of injury and repair in COVID-19 ARDS. Here, we established three groups for comparison, including patients with COVID-19 ARDS that survived hospitalization, patients with fatal COVID-19 ARDS, and ARDS patients without COVID-19 as a control comparison group. We analyzed differential gene expression at each individual time point and performed longitudinal analysis to identify unique patterns of dynamic gene expression throughout acute illness.

Here we performed longitudinal RNA sequencing analysis of PBMCs from patients with ARDS at a single center during the COVID-19 pandemic (May 2020 – June 2021) and compared temporal gene expression changes between COVID-19 ARDS survivors and non-survivors, as well as non-COVID ARDS patients, across 10 days of ICU admission. Longitudinal analysis revealed 341 transcripts with significantly different patterns of dynamic gene expression over time with most significant differences in pathways of iron homeostasis, heme biosynthesis, and erythrocyte function, that remained upregulated throughout ICU course in fatal COVID-19 ARDS compared to survivors of COVID-19 ARDS. Enriched gene signatures of hemoglobin metabolism have previously been described in blood leukocytes from septic patients (30). In vitro studies have shown induction of hemoglobin genes during cellular stress in murine macrophages (31) and human PBMCs (32). The role of heme in macrophages is complex (32–34) and additional data are needed to support strategies targeting heme biosynthesis in ARDS patients. Notably, our study employing longitudinal sampling and analysis identified distinct pathway regulation throughout disease course that was not identified at single time points. This study highlights the dynamic nature of COVID-19 ARDS and represents a novel approach towards better understanding of COVID-19 ARDS pathobiology.

The COVID-19 pandemic increased the global incidence of severe viral pneumonia, acute respiratory failure, and ARDS (7). Clinicians and researchers alike observed unique features of COVID-19 ARDS and raised the question: Is COVID-19 ARDS somehow different than “regular” ARDS (35, 36)? As a syndrome, ARDS is known for its heterogeneity and efforts to identify subtypes and subgroups of ARDS patients predate the pandemic (10, 13). Our study identified a cohort of ARDS patients with COVID-19 and varied clinical outcomes and compared to a group of patients with ARDS and bacterial sepsis as the primary ARDS risk factor. Our differential expression analysis on Day 1 of study enrollment was characterized by significant activation of immune system pathways and interferon signaling consistent with prior reports in studies of severe COVID-19 (37–40). Dysregulated interferon signaling has been implicated in COVID-19 as an important driver of pathology (21). Compared to healthy controls, interferon responses are elevated in COVID-19 (41). However, when compared to other viral infections or among COVID-19 patients with varied clinical course, interferon responses are more variable (42). Subjects in our comparison group of non-COVID-19 ARDS had sepsis from bacterial sources as their primary risk factor for ARDS. Host responses to bacterial versus viral infection yield different transcriptomic signatures (43–45), which may account for the significant differences in interferon responses in our analysis. While multiple interferon-stimulated genes showed significant upregulation in our analysis, one specific gene, interferon-inducible protein 27 (IFI27), showed significant upregulation at study enrollment in COVID-19 ARDS compared to non-COVID-19 ARDS. This recapitulates findings of other study that suggest robust and specific upregulation of IFI27 in COVID-19. A re-analysis of nine independent cohort studies that analyzed peripheral blood gene expression across multiple infections, including COVID-19, influenza, and bacterial pneumonia, described a COVID-19-specific gene signature comprised of 149 genes, among which IFI27 was the sole gene directly associated with the interferon response (46). A single-center cohort study comparing peripheral blood transcriptomes at one time point showed IFI27 was highly upregulated in COVID-19, even when compared to influenza and seasonal coronaviridae (42). Despite reports of varied interferon responses, robust induction of IFI27 may represent a peripheral blood gene signature unique to COVID-19 infection. Notably, our study, like the above referenced COVID-19 transcriptomics studies, focused on mixed populations of peripheral blood cells, which may reflect transcriptomic signatures driven by a single cell type or changes in relative cell populations. Further, PBMCs represent transcriptional signatures in peripheral blood, which may differ from the lung transcriptome (47).

Despite multiple prior reports of transcriptomics in COVID-19, our study is unique in our sampling and analysis strategy. In opposition to single time-point studies or limited longitudinal sampling, we analyzed peripheral blood gene expression throughout acute illness at multiple short intervals. While our study is limited by the small sample size and single-center design, it provides proof of concept that longitudinal molecular profiling can be valuable for identifying dynamic molecular mechanisms that function during ARDS disease course. In ARDS studies, analysis at individual timepoints allows for identification of relative differences in gene expression between groups. However, this design is vulnerable to baseline patient heterogeneity and lead-time bias. While longitudinal sampling is subject to the same obstacles, our approach allows each patient to function as their own baseline control and focuses on dynamic gene expression during acute illness. We identified gene signatures related to iron homeostasis, erythropoietin signaling, heme biosynthesis, and iron uptake that we grouped into processes contributing to erythropoiesis, as well as upregulation of erythrocyte function in fatal COVID-19 compared to COVID-19 survivors. Among ARDS controls, we observed early upregulation of these genes compared to the other groups with decreased rate of change between days 7 and 10. Among the COVID-19 non-survivors, gene expression related to erythropoiesis and erythrocyte function showed late increased compared to COVID-19 survivors, which showed relatively low expression throughout ICU course. Other groups have employed sampling at multiple time points with various strategies and have also identified gene expression correlating with pathways of hemopoiesis, reactive oxygen species, and erythrocyte functioning (20, 22, 23). Zheng et al. examined longitudinal transcriptomes but defined three ad hoc clinical stages (treatment, convalescence, and rehabilitation) as opposed to consistent time intervals; they identified early downregulation of genes related to humoral immunity and type I interferon response and upregulated gene expression related to hemopoiesis, regulation of inflammatory response, mRNA splicing via spliceosome, and epithelial cell proliferation during COVID-19 recovery (20). Another study applied longitudinal multi-omics with 2 – 7 times points from 0 – 55 days after admission and demonstrated increased protein catabolism, erythrocyte differentiation, ferroptosis, and organelle disassembly in clusters primarily corresponding to COVID non-survivors compared to COVID survivors (23). As our study and others have identified common pathways of iron homeostasis and erythrocyte function, it is intriguing to hypothesize that interventions targeting effective erythropoiesis, iron handling, and erythrocyte function represent a novel therapeutic strategy in COVID-19 ARDS. Indeed, studies have demonstrated perturbations in iron handling and ferroptosis (iron-dependent cell death) due to COVID-19 in humans (48), animal models (49), and human cells (50), suggesting that COVID-19 may uniquely impact iron homeostasis and downstream pathways during disease course. Further, observational human studies, including ours, may be capturing a compensatory mechanism of increased erythropoiesis and erythrocyte function in severe COVID-19. Steroid treatment with dexamethasone is a common and accepted treatment in severe COVID-19 (14, 15) and may impact iron metabolism pathways (51). In our cohort, steroid exposure was high in the COVID-19 groups, as 8 of 9 subjects received dexamethasone. However, ARDS controls did not receive steroid treatment. All patients in our cohort met sepsis criteria, recognizing the frequent overlap between critical illness syndromes, such as ARDS and sepsis. Among septic patients, heme metabolism and iron homeostasis genes are enriched in white blood cells (neutrophils and PBMCs) compared to controls (30), and transcriptomic differences in heme biosynthesis genes correlated with sepsis subgroups or endotypes (52). In non-erythroid cells, hemoglobin scavenges free radicals and functions in nitric oxide metabolism (53). Cellular damage through reactive oxygen species is a known pathologic mechanism in both sepsis and ARDS. Our study and others (23, 52) demonstrate increased heme-related transcript expression correlated with worse outcomes in certain critically ill populations. We postulate that upregulation of these transcripts in non-erythroid cells represents a compensatory mechanism due to ongoing oxidative stress. As iron and heme metabolism are foundational biologic processes, it is unclear if these pathways represent viable interventional targets during COVID-19 ARDS.

Our study supports a novel patient sampling strategy and demonstrates a unique analysis of dynamic gene expression. We focused on patients with severe COVID-19 ARDS as a subgroup of ARDS patient to gain insights into disease pathology. Our findings suggest an approach for future studies, generalizable to larger COVID-19 and ARDS cohorts. Our study is limited by several factors. First, all patients were enrolled from a single center for analysis, which may not be representative of the general population. As we analyzed multiple samples from each subject, the sample size of the cohort was small. While we attempted to select age- and sex-matched subjects, the sample size does not allow for proper control of these potentially confounding variables. Despite the small cohort size, we were able to identify several biologically relative pathways, reflected in correlation with prior studies, and provide new insights via our longitudinal approach. We defined outcomes by mortality in this study. While our COVID-19 ARDS non-survivors demonstrate impaired oxygenation throughout ICU stay, consistent with worsened ARDS, this group also had baseline higher SOFA scores that increased over time and increased vasopressor requirements, which may confound our findings. Further, this study used ARDS as a primary inclusion criterion, but severe COVID-19 is a systemic disease. All patients in our cohort met criteria for sepsis, and we did not attempt to differentiate COVID-19 ARDS, sepsis, or both as independent groups. Our transcriptomic findings in peripheral blood cells may be related to systemic responses instead of lung-specific pathology related to COVID-19 or ARDS. Disease classification in critical illness is currently limited by often overlapping syndrome-based paradigms (54, 55). However, transcriptomic studies have the potential to identify dysregulated host response pathways during critical illness that inform clinical disease course beyond established definitions of ARDS or sepsis (56).

Patients with ARDS from COVID-19 may represent a subgroup of ARDS patients with distinct molecular pathophysiology that drive disease outcomes. Our study identified differences in dynamic expression of genes related to iron homeostasis and erythrocyte function that correlate to survival in COVID-19 ARDS. Our findings are supported by prior studies of molecular profiling in COVID-19 and suggest that iron handling and ferroptosis may be putative mechanisms of ongoing lung injury during SARS-CoV-2 infection. Additional research is needed to examine the therapeutic potential of these pathways. Finally, our study suggests that short interval longitudinal sampling during acute illness may uncover novel mechanisms of injury, repair, and resolution during COVID-19 ARDS. This approach should be considered further for future ARDS molecular profiling study design.

The data presented in the study are deposited in the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/), accession number GSE273149.

The studies involving humans were approved by The Ohio State University Biomedical Sciences Institutional Review Board. The studies were conducted in accordance with the local legislation and institutional requirements. The human samples used in this study were acquired from an existing biorepository that collects and stores human biosamples and data from critically-ill patients at Ohio State University. The biorepository project utilizes broad consent to collect and store samples for use in secondary analyses. Written informed consent for participation was not required from the participants or the participants’ legal guardians/next of kin in accordance with the national legislation and institutional requirements.

ME: Formal Analysis, Validation, Visualization, Writing – original draft, Writing – review & editing, Data curation, Investigation. FJ: Data curation, Investigation, Writing – review & editing, Methodology. DF: Investigation, Methodology, Writing – review & editing, Supervision. LL: Investigation, Writing – review & editing, Data curation. SC: Data curation, Investigation, Writing – review & editing. KH: Data curation, Investigation, Writing – review & editing. SK: Data curation, Investigation, Writing – review & editing. GS: Data curation, Investigation, Writing – review & editing. SP: Data curation, Investigation, Writing – review & editing, Conceptualization, Methodology. JH: Methodology, Writing – review & editing, Funding acquisition, Project administration, Resources, Supervision. RM: Writing – review & editing, Funding acquisition, Methodology, Project administration, Resources, Supervision. JE: Methodology, Writing – review & editing, Data curation, Investigation, Writing – original draft. JB: Methodology, Writing – original draft, Writing – review & editing, Conceptualization, Formal Analysis, Funding acquisition, Project administration, Visualization.