key: cord-0314635-afvxwbu5
authors: Batra, R.; Wahlen, W.; Alvarez-Mulett, S.; Hoffman, K.; Simmons, W.; Harrington, J.; Chetnik, K.; Buyukozkan, M.; Benedetti, E.; Choi, M. E.; Suhre, K.; Schmidt, F.; Schenck, E.; Choi, A. M. K.; Cho, S. J.; Krumsiek, J.
title: Multi-omic comparative analysis of COVID-19 and bacterial sepsis-induced ARDS
date: 2022-05-17
journal: nan
DOI: 10.1101/2022.05.16.22274587
sha: 78938cdeee5e3617bea9818067999db864ce3cb8
doc_id: 314635
cord_uid: afvxwbu5

Background: Acute respiratory distress syndrome (ARDS), a life-threatening condition characterized by hypoxemia and poor lung compliance, is associated with high mortality. ARDS induced by COVID-19 has similar clinical presentations and pathological manifestations as non-COVID-19 ARDS. However, COVID-19 ARDS is associated with a more protracted inflammatory respiratory failure compared to traditional ARDS. Therefore, a comprehensive molecular comparison of ARDS of different etiologies groups may pave the way for more specific clinical interventions. Methods and Findings: In this study, we compared COVID-19 ARDS (n=43) and bacterial sepsis-induced (non-COVID-19) ARDS (n=24) using multi-omic plasma profiles covering 663 metabolites, 1,051 lipids, and 266 proteins. To address both between- and within- ARDS group variabilities we followed two approaches. First, we identified 706 molecules differently abundant between the two ARDS etiologies, revealing more than 40 biological processes differently regulated between the two groups. From these processes, we assembled a cascade of therapeutically relevant pathways downstream of sphingosine metabolism. The analysis suggests a possible overactivation of arginine metabolism involved in long-term sequelae of ARDS and highlights the potential of JAK inhibitors to improve outcomes in bacterial sepsis-induced ARDS. The second part of our study involved the comparison of the two ARDS groups with respect to clinical manifestations. Using a data-driven multi-omic network, we identified signatures of acute kidney injury (AKI) and thrombocytosis within each ARDS group. The AKI-associated network implicated mitochondrial dysregulation which might lead to post-ARDS renal-sequalae. The thrombocytosis-associated network hinted at a synergy between prothrombotic processes, namely IL-17, MAPK, TNF signaling pathways, and cell adhesion molecules. Thus, we speculate that combination therapy targeting two or more of these processes may ameliorate thrombocytosis-mediated hypercoagulation. Conclusion: We present a first comprehensive molecular characterization of differences between two ARDS etiologies: COVID-19 and bacterial sepsis. Further investigation into the identified pathways will lead to a better understanding of the pathophysiological processes, potentially enabling novel therapeutic interventions.

Acute respiratory distress syndrome (ARDS), a severe form of respiratory failure that is associated with high mortality, emerged as a frequent complication of coronavirus disease 2019 [1] . ARDS may be induced by other infections (sepsis, influenza), major traumatic injury, or inhalation of toxic chemicals [2] . Clinical presentations and pathological manifestations of COVID-19 ARDS overlap with non-COVID-19 ARDS, including decreased static lung compliance, hypoxemia, hypercarbia, inflammation, thrombosis, and endothelial injury [3] [4] [5] [6] [7] [8] [9] .

However, COVID-19 ARDS is specifically characterized by a protracted hyperinflammatory state, and may lead to higher rates of thrombosis as well as fibroproliferative lung remodeling [10] [11] [12] .

These differences in ARDS etiologies have not yet been fully characterized to an extent that would enable timely and tailored clinical care. Moreover, ARDS is a heterogeneous disorder with substantial molecular differences even within a specific ARDS group [13] . Thus, to provide deeper insight into disease pathophysiology and enable etiology-specific therapeutic interventions, a comprehensive molecular characterization of variations between and within ARDS groups is needed.

Previously, ARDS groups have been studied in comparison to non-ARDS reference groups, such as healthy controls or hospitalized patients without ARDS [5] [6] [7] [8] [9] [14] [15] [16] . Here, we present the first detailed comparative multi-omic analysis between COVID-19 ARDS (n=43) and bacterial sepsisinduced (non-COVID-19) ARDS (n=24). The comprehensive measurement panel included 1,980 molecules, including 663 metabolites, 1,051 lipids, and 266 proteins. We followed a two-step analysis workflow to elucidate the differences between the two ARDS groups. In the first part, we directly compared patients from the two groups to identify differentially abundant molecules.

These molecules were involved in various biological processes and may highlight the differences in the pathological manifestation of the groups. Furthermore, we analyzed a set of ARDSassociated biological processes with therapeutic relevance, covering various signaling and metabolic pathways. In the second part of the study, we associated clinical manifestations, including acute kidney injury (AKI), thrombocytosis (platelet count), patient's oxygen in arterial blood to the fraction of the oxygen in the inspired air (PaO2/FiO2), and mortality, with the molecular profiles to identify molecular signatures in each ARDS group. For a systematic comparison of these molecular changes, we performed multi-omic network analysis and identified subnetwork-based signatures. To the best of our knowledge, this is the first study addressing both between and within ARDS group variabilities in a large-scale multi-omic setting.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 17, 2022 

We analyzed 67 patients admitted to the intensive care unit at Weill Cornell Medical Center . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint Figure 1 : Study overview. This study was based on 67 ARDS patients, 43 with COVID-19 and 24 with bacterial sepsis group. Profiling of plasma samples resulted in 1,906 measured molecules, including 663 metabolites, 1,051 lipids, and 266 proteins. For inter-ARDS comparison, we identified molecules and pathways differently regulated between the two ARDS groups. In addition, focusing on several selected pathways with therapeutic relevance, we constructed a cascade of biological processes starting from sphingosine metabolism. For intra-ARDS comparison, we identified molecules associated with clinical manifestations, including acute kidney injury (AKI), thrombocytosis (platelet count), PaO2/FiO2 ratio, and mortality, within each ARDS group. Further, we constructed a data-driven multi-omic network based on the Gaussian graphical model (GGM). This network was used to generate subnetworks associated with clinical manifestations.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

To assess the molecular differences between COVID-19 and bacterial sepsis-induced ARDS, we analyzed three molecular layers -metabolic, lipidomic, and proteomic. A total of 175 out of 663 metabolites, 437 out of 1,051 lipids, and 94 out of 266 proteins were differentially abundant between the groups at a 5% false discovery rate (FDR) (Figure 2a) . Detailed results of this analysis are available in Supplementary Table 2 .

To identify the biological processes underlying the differences between the ARDS groups, the differentially abundant molecules were functionally annotated. Metabolites were annotated using Metabolon's 'sub-pathways', lipids were annotated by lipid classes, and proteins were annotated using signaling pathways from KEGG [17] (Supplementary Table 3 ). Top ranking pathways are depicted in Figure 2b . Interestingly, several of these pathways from each of the three omics have previously been implicated in COVID-19 ARDS or non-COVID-19 ARDS and will be further discussed per omics and pathway in the following. To corroborate our findings, we used previous studies which have compared these ARDS groups with healthy controls or less severe COVID-19 cases. We followed this route to provide general evidence for the importance of the respective pathway in ARDS, and since parallel studies comparing COVID-19 and non-COVID-19 ARDS at the molecular level in a high-throughput setting were unavailable.

Branched-chain amino acids (BCAAs): In our analysis, 10 metabolites from this pathway were differentially abundant between the ARDS groups. Of these, 8 had higher levels in COVID-19 ARDS compared to bacterial sepsis-induced ARDS, and 2 had lower levels. A previous study based on bronchoalveolar lavage (BAL) fluid reported BCAAs to be higher in non-COVID-19 ARDS groups compared to non-ARDS groups [18] . Further corroborating the role of this metabolite class, BCAAs were recently found to be differently regulated in severe COVID-19 cases compared to mild COVID-19 cases [19] . Glutamate metabolism: 7 metabolites from this pathway were found to be differentially abundant between the ARDS groups, of which 4 had higher levels in COVID-19 ARDS compared to bacterial sepsis-induced ARDS and 3 had lower levels. Previous studies have reported elevated glutamate levels in BAL fluid of non-COVID-19

ARDS patients compared to healthy controls, as well as elevated levels of metabolites involved in glutamate metabolism in serum of patients with severe COVID-19 disease compared to healthy controls [19] .

We observed substantial lipidomic changes between the two ARDS groups, with the greatest differences observed in the triacylglycerols (260) and diacylglycerols (32) lipid classes. Triacylglycerols and diacylglycerols levels have been previously associated with . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint mortality in ARDS [20] , highlighting the role of lipid metabolism in the prognosis of patients.

However, previous studies have reported inconsistent results, with both higher and lower levels of triacylglycerols in the COVID-19 compared to a control group [8, 21] . In our study, 158 TAGs and 29 DAGs had higher levels in COVID-19 ARDS compared to bacterial sepsis-induced ARDS while 102 TAGs and 3 DAGs had lower levels.

Proteomic pathways: PI3K-AKT signaling: In our analysis, 12 proteins from the PI3K-AKT pathway were differentially abundant between the ARDS groups, of which 8 molecules had higher levels in COVID-19 ARDS compared to bacterial sepsis-induced ARDS and 4 had lower levels.

PI3K-AKT signaling plays a pivotal role in the induction of a hyperinflammatory state [22] and the propagation of acute lung injury [23] . Previous studies found the pathway to be elevated in COVID-19 [24] compared to influenza patients. MAPK signaling: 11 proteins from the MAPK pathway were differentially abundant between the ARDS groups, of which 8 proteins had higher levels in COVID-19 ARDS compared to bacterial sepsis-induced ARDS, and 3 had lower levels.

The MAPK signaling pathway has previously been reported to promote ARDS [25] , and its inhibition has been discussed as a potential therapeutic approach for COVID-19 [26] .

Overall, we identified 706 molecules differently abundant between the two ARDS etiologies, revealing more than 40 biological processes (Supplementary Table 3 ) differently regulated between the two groups.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

Pathway annotations provide functional relevance to the molecules differently abundant between the two ARDS etiologies. However, such an analysis does not provide any insights into the interplay of these pathways in the context of ARDS pathology. From the list of differently regulated pathways (Supplementary Table 3 ), we selected a few ARDS-associated processes for detailed investigation at the molecular level. We built a cascade of pathways that have been reported to be of pharmaceutical interest in inflammatory or infectious diseases and used literature-based evidence of their interactions (Figure 3) . Each of the pathways in this cascade and its therapeutic potential in ARDS is discussed in the following.

The cascade is built downstream of sphingosine metabolism, involving sphingosine-1 phosphate (S1P) and its receptors (S1PRs) (Figure 3a) . In our analysis, sphingosine and sphingosine-1 phosphate levels were higher in COVID-19 compared to bacterial sepsis-induced ARDS.

Sphingosine metabolism plays an important role in immune and vascular systems [27, 28] . S1P and S1PRs have gained considerable attention in the treatment of various inflammatory conditions. For instance, Fingolimod (FTY720), an agonist of S1PR1, is already in clinical use for multiple sclerosis (MS), a chronic autoimmune inflammatory disorder [29] . Moreover, S1P analogs have been studied for the treatment of cytokine storms [30] and pulmonary infections induced by influenza H1N1 and paramyxovirus [31] . Consequently, and owing to the life-threatening hyperinflammatory syndrome induced by SARS-COV2 infections [32], three clinical trials were launched to use S1P-S1RPs agonists (Fingolimod, Opaganib) against COVID-19 (Clinicaltrials.gov identifiers: NCT04280588, NCT04467840, NCT04414618). Arginine metabolism: It has been reported that the PI3K-AKT and JAK-STAT signaling pathways induce nitric oxide (NO) production via arginine [42] (Figure 3c) . NO production at higher levels mediates lung injury via the formation of toxic oxidants [43] . In our data, arginine levels were higher in COVID-19 ARDS compared to bacterial sepsis-induced ARDS. Notably, NO was not measured in our data.

Arginine depletion strategies that block its conversion to NO and citrulline are effective in inhibiting viral replication (HCV, HIV) [44, 45] and have thus been discussed as a therapeutic approach in the context of the COVID-19 [46] . Furthermore, Karki et. al [47] , reported higher levels of the NOS2 gene (coding for iNOS, one of the enzymes catalyzing nitric oxide production) in severe and critical COVID-19 cases compared to controls [47] .

JAK/STAT signaling pathway: In our analysis, 4 proteins from the JAK/STAT signaling pathways had higher levels in COVID-19 ARDS compared to bacterial sepsis-induced ARDS, and 6 proteins had lower levels (Figure 3d ). Two recent clinical trials have shown improved outcomes in COVID-19 patients that received JAK inhibitors [48, 49] .

Taken together, in this section we investigated the interplay of metabolic and signaling events that potentially lead to and propagate the pathophysiology of ARDS. Some of our observations could be corroborated using ARDS-specific literature. The rest of the findings are potentially novel and can be further investigated in a targeted manner to establish a mechanistic understanding of specific pathways/molecules in the context of ARDS. Overall, our analysis suggests that ARDS is an inflammatory process coordinated by multiple cellular processes with severe physiological implications.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 17, 2022 [44, 45] , and JAK inhibition was found to be effective in improving the outcome of COVID-19 [48, 49] .

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

In this part of our study, we compared the differences in omics associations with clinical manifestations across the two ARDS groups. These included acute kidney injury (AKI), thrombocytosis (determined by a pathological increase in platelet count), PaO2/FiO2 ratio (low ratio indicates severe hypoxia), and mortality. For PaO2/FiO2 ratio, we only found significant correlations in the COVID-19 ARDS group and there were no molecules associated with mortality in our data. Thus, these two clinical parameters were not used for the comparison of ARDS groups.

In total, 249 molecules were associated with AKI and 111 molecules associated with platelet count To obtain a systematic view of these dysregulated multi-omic molecules across ARDS groups, we adopted a network-based approach. To this end, we generated a data-driven Gaussian graphical model (GGM, Figure 4b ) [50] , which is a partial correlation-based approach to identify interactions between molecules. GGMs have previously been shown to reconstruct biochemical pathways from omics data [51-53] and therefore add biological relevance to results aside from the predefined pathway annotations. We then extracted subnetworks for AKI and thrombocytosis (see methods for details), which will be discussed in the following subsections. An interactive Cytoscape version of the full network and all subnetworks for further exploration are available in Supplementary File 1.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Metabolites from this subnetwork were from two main metabolic groups: amino acid metabolism and fatty acids from the acylcarnitine class. In critical illness, protein catabolism leads to the production of excess amino acids [56, 57] , whereas in lung injury, fatty acid oxidation is altered, resulting in the production of acylcarnitines [58] . Amino acid dysregulation has been reported in correlation with COVID-19 severity [59] [60] [61] , further highlighting the role and relevance of amino acid metabolism. Both amino acids and acylcarnitines are known to be involved in bioenergetic processes mediated by mitochondria [62, 63] and have also been associated with kidney injury [64] . Mitochondrial dysfunction often leads to oxidative stress [65] , which is a characteristic feature of COVID-19 as well as ARDS from other etiologies [66, 67] . Therefore, the dysregulation of acylcarnitines and amino acid metabolism observed in plasma suggests widespread mitochondrial dysfunction in AKI associated with ARDS.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

Thrombocytosis is marked by increased production of thrombocytes (platelets), which can be triggered by an underlying condition, such as infection. Platelet activation is crucial for various normal physiological and pathophysiological processes, including hemostasis, thrombosis, and immune response [68] . Thrombosis or coagulopathy is associated with poor prognosis in ARDS patients [11, 69] . Previous studies have reported high incidences of thrombotic complications in COVID-19 ARDS as compared to non-COVID-19 ARDS patients [11] . It has also been postulated that thrombotic manifestation in COVID-19 ARDS is atypical, i.e., despite increased platelet consumption, circulating platelet count is maintained via a compensatory platelet production [11, 69] . Taken together, our finding of correlations between thrombocytosis (platelet count) and molecules involved in cell adhesion, IL-17, TNF, MAPK signaling pathways imply a coordinated effort of these pathways toward thrombocytosis-mediated coagulopathy during ARDS.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint

In this study, we performed a comprehensive multi-omic comparative analysis of COVID-19 and bacterial sepsis-induced ARDS. We profiled plasma samples from 67 patients hospitalized at WCMC/NYP using untargeted metabolomics, untargeted lipidomics, and targeted proteomics profiling, resulting in the quantification of 1,980 molecules. To perform the comparison of molecular differences between these ARDS groups, we followed two approaches. First, to identify differently regulated molecules and biological processes in the two groups, we directly compared the molecular profiles between COVID-19 ARDS and bacterial sepsis-induced ARDS. Second, to obtain an overview of the similarities and differences in molecular presentation of severity in both groups, we compared molecular associations with clinical manifestation within each group.

For the first part of the study, we identified 706 molecules (metabolites, lipids, proteins) differently abundant between COVID-19 ARDS and bacterial sepsis-induced ARDS. These molecules spanned various biological processes (Figure 2b) and may drive the pathological manifestation of the two etiologies. To further contextualize our findings, we built a cascade of ARDS-induced changes in a selected set of interrelated pathways with therapeutic relevance, including sphingosine metabolism, MAPK, RAS, PI3K/AKT signaling, arginine metabolism, and JAK-STAT signaling. This analysis suggested that ARDS is coordinated by multiple cellular processes with severe pathophysiological consequences and led to two main propositions: (1) We speculate that arginine metabolism plays a critical role in the long-term sequelae of ARDS, as arginine metabolism has previously been shown to be altered in the pulmonary fibrosis [83, 84] . (2) We postulate that blockage of JAK-STAT signaling may improve outcomes of bacterial sepsisinduced ARDS. JAK-STAT activation has been implicated in the pathogenesis of ARDS previously [85, 86] and its inhibition has already been shown to improve outcomes of COVID-19 ARDS [87] .

For the second part of our study, to examine within-ARDS heterogeneity, we compared molecular profiles within each of the two ARDS groups concerning clinical manifestations. We identified ARDS group-specific signatures for AKI, thrombocytosis (platelet count). Using a multi-omic network, we identified two network-based signatures for AKI and thrombocytosis. The AKI-related subnetwork included deregulated amino acids and acylcarnitines, hinting toward aberrations in bioenergetic processes mediated by mitochondria. Importantly, mitochondrial dysfunction is known to cause the progression of AKI to chronic kidney disease (CKD) [88, 89] . Thus, we . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint hypothesize that mitochondrial dysfunction associated with ARDS may lead to a worse prognosis of AKI. Renal sequelae have been studied in people suffering from severe AKI and requiring renal replacement therapy during COVID-19 infection [41] . The thrombocytosis-related subnetwork included deregulated molecules from IL-17, TNF, MAPK signaling pathways, and cell adhesion molecules. Our findings suggest a synergy between the above-mentioned prothrombotic processes [73, 90, 91] as a likely reason for hypercoagulation in ARDS. We speculate that combination therapy targeting two or more of these processes may ameliorate hypercoagulation. Our findings are encouraging and warrant further investigation to evaluate their potential for applicability in clinics and ARDS-specific therapeutic intervention.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 17, 2022 

The cohort was derived from the Weill Cornell Biobank of Critical Illness (WC-BOCI) at WCMC/NYP. The process for recruitment, data collection, and sample processing has been described previously [93] [94] [95] . In brief, the recruits in the WC-BOCI database were patients 

Definitions used to diagnose key clinical manifestations used in the study are described below.

Acute Respiratory Distress Syndrome (ARDS). We defined ARDS by the Berlin definition [96] , which was then adjudicated by two independent pulmonary and critical care attendings after a review of the subject's history, arterial blood gas, and chest X-ray. Bacterial sepsis-induced ARDS was defined if subjects met the criteria for ARDS in addition to meeting the definition for sepsis outlined in The Third International Consensus Definitions for Sepsis and Septic Shock [97] .

Subjects were diagnosed with COVID-19 if a viral swab of the nasopharynx tested positive exclusively for SARS-CoV-2 via RT-PCR.

Outcomes' definition (KDIGO). KDIGO requires a change of serum creatinine greater than or equal to 0.3 mg/dL within 48 hours, an increase in serum creatinine greater than or equal to 1.5 times baseline serum creatinine known or presumed to have occurred within the past 7 days, or urine output less than or equal to 0.5 mL/kg/hour for six hours [98] .

Sequential Organ Failure Assessment (SOFA). The SOFA score is a clinical tool used by clinicians in the ICU to determine the degree of a patient's organ failure. The SOFA score is composed of the following variables, with higher values being assigned for more severe alterations: PaO2/FiO2 (mm Hg), Platelets x 10 3 /µL, Glasgow Coma Scale, Bilirubin (mg/dL), Mean Arterial Pressure or administration of vasoactive required, and Creatinine (mg/dL) [99] .

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint

Standard serum collection and storage practices at the New York-Presbyterian/Weill Cornell Medical College include collecting venous blood into a serum-separating tube (SST). Serum was obtained by centrifuging at 1,500g for 7 minutes as soon as possible after collection and latest 2 hours after sample collection. Specimens were stored at 4°C for 1 to 5 days before being coded/de-identified and then transferred into a -80°C freezer. Samples were thawed and inactivated in different ways: For the metabolic profiling, x3 sample volume of HPLC grade ethanol was added; for the proteomics analysis, the samples were heat-inactivated in a water bath of 56°C for 15 minutes. After these processes, the samples were again stored at -80°C until the omic profiling were performed.

This assay was performed using the Olink platform (Uppsala, Sweden) at the Proteomics Core of 

This assay was performed by Metabolon, Inc (Morrisville, NC) which utilizes ultrahigh performance liquid chromatograph-tandem mass spectroscopy (UPLC-MS/MS).

Sample preparation was performed using the automated MicroLab STAR® system from Hamilton

Company. For quality control, before extraction, several recovery standards were added. For . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint extraction, methanol with vigorous shaking followed by centrifugation was used to remove protein, dissociate small molecules bound to protein or trapped in the precipitated protein matrix, and recover chemically diverse metabolites. The resulting extract was placed briefly on a TurboVap® (Zymark) to remove the organic solvent and stored overnight under nitrogen before preparation for analysis.

For quality assurance /quality control (QA/QC), several types of controls were analyzed in concert with the experimental samples that allowed instrument performance monitoring and aided chromatographic alignment. Instrument variability was determined by calculating the median relative standard deviation (RSD) for the standards that were added to each sample before injection into the mass spectrometers. Overall process variability was determined by calculating the median RSD for all endogenous metabolites (i.e., non-instrument standards) present in 100% For metabolite identification, raw data was extracted, peak-identified and QC processed using Metabolon's hardware and software. Metabolon maintains a library based on authenticated standards that contain the retention time/index (RI), mass to charge ratio (m/z), and chromatographic data (including MS/MS spectral data) on all molecules present in the library. A variety of curation procedures were carried out to ensure that a high-quality data set was made available for statistical analysis and data interpretation.

This assay was also performed by Metabolon, Inc.

For sample preparation, lipids were extracted from the biofluid in the presence of deuterated internal standards using an automated BUME extraction according to the method of Lofgren et al [100] .

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

Metabolites, lipids, and proteins with more than 25% missing values were removed, leaving 663 out of 1,005 measured metabolites, 1,051 out of 1,218 measured lipids, and 266 out of 276 measured proteins. Sample-wise variation in the data was corrected using probabilistic quotient normalization [101] , followed by log2 transformation. The remaining missing values were imputed using a k-nearest-neighbor-based algorithm [102] . Ten proteins (CCL3, CXCL1, FGF-21, FGF-23, IL-18, IL-6, MCP-1, OPG, SCF, uPA) were measured in multiple Olink panels, and the replicate values for each sample were averaged prior to statistical analysis. All data processing was performed using the maplet R package [103] .

The metabolite, lipid, and protein associations were computed using linear models with the molecules as response variables and diagnosis/clinical manifestations (ARDS group, AKI levels, thrombocyte/platelet count, mortality status) as predictors. Since demographic factors including age, sex, and BMI are considered determinants of COVID-19 severity [104] , we did not consider them as covariates in the models. Multiple hypothesis testing was accounted for by correcting the p-values using the Benjamini-Hochberg (BH) method [105] . All of these analyses were performed using the maplet R package [103] .

Metabolites were annotated using Metabolon's 'sub-pathway' groups, lipids were annotated by lipid classes, and proteins were annotated using signaling pathways from Kyoto Encyclopedia of Genes and Genomes (KEGG) [17] . The complete list of pathways annotated to the significant . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Figure 2b , only pathways/classes with at least 4 significant molecules were included. Moreover, only Metabolon's sub-pathways with the term 'metabolism' and only KEGG pathways with the phrase 'signaling pathway' were considered for this analysis.

A partial correlation-based Gaussian graphical model (GGM) was computed using the GeneNet R package [106] to infer a multi-omic network. Partial correlations with FDR < 0.05 were used for network construction between molecules. This multi-omic network was annotated with a score which was computed for each molecule/outcome combination as follows:

is the adjusted p-value of the model, and d is the direction (-1/1) of the association based on test statistic (positive or negative correlation with the outcome).

This score was used to color the nodes in Figure 5 .

Subnetworks associated with clinical manifestations were generated from the multi-omic network by selecting the molecules significantly associated with the specific clinical manifestation at 5%

FDR and at 10% FDR that interact with the molecules significant at 5% FDR. Within this subnetwork, we focused on the largest connected component.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 17, 2022 Krumsiek J, Suhre K, Illig T, Adamski J, Theis FJ. Gaussian graphical modeling reconstructs pathway reactions from high-throughput metabolomics data. BMC Syst Biol.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 17, 2022 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint

Biological mechanisms of COVID-19 acute respiratory distress syndrome

The acute respiratory distress syndrome: from mechanism to translation

Clinical characteristics of SARS-CoV-2 infection in children with cystic fibrosis: An international observational study

COVID-19-versus non-COVID-19-related acute respiratory distress syndrome: Differences and similarities

Acute respiratory distress syndrome

Therapeutic targeting of metabolic alterations in acute respiratory distress syndrome

Proteomic study of acute respiratory distress syndrome: current knowledge and implications for drug development

Multi-omic Analysis of COVID-19 Severity

Proteomic and Metabolomic Characterization of COVID-19

Circuits between infected macrophages and T cells in SARS-CoV-2 pneumonia

High risk of thrombosis in patients with severe SARS-CoV-2 infection: a multicenter prospective cohort study

Uncontrolled Innate and Impaired Adaptive Immune Responses in Patients with COVID-19 Acute Respiratory Distress Syndrome

Understanding a Heterogeneous Syndrome

Hypolipidemia is associated with the severity of COVID-19

Integrative Metabolomic and Proteomic Signatures Define Clinical Outcomes in Severe

KEGG for integration and interpretation of large-scale molecular data sets

Therapeutic targeting of metabolic alterations in acute respiratory distress syndrome

Metabolomics analysis reveals a modified amino acid metabolism that correlates with altered oxygen homeostasis in COVID-19 patients

Omics-Driven Systems Interrogation of Metabolic Dysregulation in COVID-19 Pathogenesis

PI3K signalling in inflammation

PI3K-γ inhibition ameliorates acute lung injury through regulation of IκBα/NF-κB pathway and innate immune responses

Serum Protein Profiling Reveals a Landscape of Inflammation and Immune Signaling in Early-stage COVID-19 Infection

P38MAPK plays a pivotal role in the development of acute respiratory distress syndrome

p38 MAPK inhibition: A promising therapeutic approach for COVID-19

Sphingosine-1-phosphate receptors and innate immunity

Vascular and Immunobiology of the Circulatory Sphingosine 1-Phosphate Gradient

Fingolimod (FTY720): Discovery and development of an oral drug to treat multiple sclerosis

Suppression of cytokine storm with a sphingosine analog provides protection against pathogenic influenza virus

Battling COVID-19 Pandemic: Sphingosine-1-Phosphate Analogs as an z 42

The role of nitric oxide in metabolic regulation of Dendritic cell immune function

Nitric oxide in the lungtherapeutic and cellular mechanisms of action

Nitric Oxide Synthesis Enhances Human Immunodeficiency Virus Replication in Primary Human Macrophages

Arginine depletion as a therapeutic approach for patients with COVID-19

Synergism of TNF-α and IFN-γ Triggers Inflammatory Cell Death, Tissue Damage, and Mortality in SARS-CoV-2 Infection and Cytokine Shock Syndromes

Baricitinib plus Remdesivir for Hospitalized Adults with Covid-19

Tofacitinib in Patients Hospitalized with Covid-19 Pneumonia

A shrinkage approach to large-scale covariance matrix estimation

Phenotype-driven identification of modules in a hierarchical map of multifluid metabolic correlations

Network inference from glycoproteomics data reveals new reactions in the IgG glycosylation pathway

Acute respiratory distress syndrome and risk of AKI among critically ill patients

Epidemiology and Outcomes of Acute Kidney Injury in COVID-19 Patients with Acute Respiratory Distress Syndrome: A Multicenter Retrospective Study

Metabolic aspects of muscle wasting during critical illness

Protein catabolism and requirements in severe illness

Impairment of fatty acid oxidation in alveolar epithelial cells mediates acute lung injury

Altered amino acid profile in patients with SARS-CoV-2 infection

Integrated cytokine and metabolite analysis reveals immunometabolic reprogramming in COVID-19 patients with therapeutic implications

Nutrient Metabolism, Human | Learn Science at Scitable. In: Nature Education

Circulating acylcarnitines as biomarkers of mitochondrial dysfunction after acetaminophen overdose in mice and humans

Serum metabolomic profiles from patients with acute kidney injury: A pilot study

Mitochondrial dysfunction and oxidative stress in metabolic disorders -A step towards mitochondria based therapeutic strategies

COVID-19: A Mitochondrial Perspective

Platelet gene expression and function in patients with COVID-19

The Impact of COVID-19 Disease on Platelets and Coagulation

Effects of Interleukin 17 on the cardiovascular system

Cell adhesion mechanisms in platelets

Vascular Injury and Arterial Thrombosis

Combination of IL-17 and TNFα induces a pro-inflammatory, pro-coagulant and pro-thrombotic phenotype in human endothelial cells

TNF-adriven inflammation and mitochondrial dysfunction define the platelet hyperreactivity of aging

COVID-19 and thrombosis: From bench to bedside

SARS-CoV-2 binds platelet ACE2 to enhance thrombosis in COVID-19

Thrombosis in COVID-19

Immunothrombosis in Acute Respiratory Distress Syndrome: Cross Talks between Inflammation and Coagulation

COVID-19: a case for inhibiting IL-17?

The pathogenesis and treatment of the 'Cytokine Storm'' in COVID-19

Accumulating evidence suggests anti-TNF therapy needs to be given trial priority in COVID-19 treatment

The Potential for Repurposing Anti-TNF as a Therapy for the Treatment of COVID-19

Metabolic heterogeneity of idiopathic pulmonary fibrosis: A metabolomic study

Induction of arginase I and II in bleomycin-induced fibrosis of mouse lung

Activation of the STAT pathway in acute lung injury

Protective effect of suppressing STAT3 activity in LPS-induced acute lung injury

Ruxolitinib rapidly reduces acute respiratory distress syndrome in covid-19 disease. Analysis of data collection from respire protocol

Mitochondrial dysfunction and the AKI-to-CKD transition

The role of mitochondria in acute kidney injury and chronic kidney disease and its therapeutic potential

COVID-19 pulmonary pathology: a multi-institutional autopsy cohort from Italy and New York City

Promotes Aortic Endothelial Activation Via Transcriptionally and Post Transcriptionally Activating p38 MAPK Pathway

the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted

Comparison of qSOFA and SIRS for predicting adverse outcomes of patients with suspicion of sepsis outside the intensive care unit

Inflammasome-regulated cytokines are critical mediators of acute lung injury

Circulating cell death biomarker TRAIL is associated with increased organ dysfunction in sepsis

Acute respiratory distress syndrome: The Berlin definition

The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3)

KDIGO clinical practice guidelines for acute kidney injury

The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure. Intensive Care Med

The BUME method: a novel automated chloroform-free 96-well total lipid extraction method for blood plasma

Probabilistic quotient normalization as robust method to account for dilution of complex biological mixtures. Application in1H NMR metabonomics

Characterization of missing values in untargeted MS-based metabolomics data and evaluation of missing data handling strategies

CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) an extensible R toolbox for modular and reproducible metabolomics pipelines

Features of 20 133 UK patients in hospital with covid-19 using the ISARIC WHO Clinical Characterisation Protocol: prospective observational cohort study

Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing

A shrinkage approach to large-scale covariance matrix estimation and implications for functional genomics

The study was approved by the institutional review board at Weill Cornell Medicine (#22-03024534). Written informed consent was received before participation by all patients, except when the institutional review board approved a waiver of informed consent (eg, for the use of discarded samples and deidentified patient data).

The data used in this study can be downloaded at https://doi.org/10.6084/m9.figshare.19775359All R scripts to generate the tables and figures of this paper are available at https://github.com/krumsieklab/covid-ards-plasma . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 17, 2022. ; https://doi.org/10.1101/2022.05.16.22274587 doi: medRxiv preprint