If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New YorkDepartment of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New YorkDepartment of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New YorkDepartment of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New YorkDepartment of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New YorkDepartment of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New YorkDepartment of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New YorkDepartment of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New York
Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New YorkDivision of Hematology and Medical Oncology, Department of Medicine, Weill Cornell Medicine, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New YorkDepartment of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New York
Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New YorkDivision of Nephrology and Hypertension, Department of Medicine, Weill Cornell Medicine, New York, New York
Division of Pulmonary and Critical Care Medicine, Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, Weill Cornell Medicine, New York, New YorkDepartment of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New York
Department of Medicine, New York-Presbyterian Hospital–Weill Cornell Medical Center, New York, New YorkAnsary Stem Cell Institute, Division of Regenerative Medicine, Department of Medicine, Weill Cornell Medicine, New York, New York
Vascular injury is a well-established, disease-modifying factor in acute respiratory distress syndrome (ARDS) pathogenesis. Recently, coronavirus disease 2019 (COVID-19)–induced injury to the vascular compartment has been linked to complement activation, microvascular thrombosis, and dysregulated immune responses. This study sought to assess whether aberrant vascular activation in this prothrombotic context was associated with the induction of necroptotic vascular cell death. To achieve this, proteomic analysis was performed on blood samples from COVID-19 subjects at distinct time points during ARDS pathogenesis (hospitalized at risk, N = 59; ARDS, N = 31; and recovery, N = 12). Assessment of circulating vascular markers in the at-risk cohort revealed a signature of low vascular protein abundance that tracked with low platelet levels and increased mortality. This signature was replicated in the ARDS cohort and correlated with increased plasma angiopoietin 2 levels. COVID-19 ARDS lung autopsy immunostaining confirmed a link between vascular injury (angiopoietin 2) and platelet-rich microthrombi (CD61) and induction of necrotic cell death [phosphorylated mixed lineage kinase domain-like (pMLKL)]. Among recovery subjects, the vascular signature identified patients with poor functional outcomes. Taken together, this vascular injury signature was associated with low platelet levels and increased mortality and can be used to identify ARDS patients most likely to benefit from vascular targeted therapies.
For decades, vascular injury has been recognized as a key element in the pathogenesis of acute respiratory distress syndrome (ARDS).
However, this has not translated into vascular targeted therapies for ARDS. This may, in part, be related to heterogeneity in the vascular response to injury among ARDS subjects, as well as to difficulty in selecting patients most at risk for ARDS vascular injury. Blood proteomics has been proposed as a novel translational approach to better match patients to precision therapies for ARDS.
A better understanding of the blood proteomic changes associated with ARDS vascular injury could therefore help identify patients likely to benefit from vascular therapies.
Previous targeted studies of circulating vascular proteins have greatly enhanced the understanding of ARDS vascular injury. For example, measurement of the plasma angiocrine factor angiopoietin 2 (ANGPT2) in patients at the early stages of ARDS demonstrates that vascular injury likely precedes mechanical ventilation
However, these ANGPT2-mediated vascular disruptions can be countered. In mice, systemic administration of platelet-derived pericyte chemokines, such as angiopoietin 1 (ANGPT1) and platelet-derived growth factor B (PDGFB), counter ANGPT2-mediated vascular disruption, demonstrating the homeostatic potential of the blood vascular proteome.
Improved understanding of the blood proteomic changes in subjects with ARDS with high or low vascular injury can build on these prior observations, shed further light onto disease pathogenesis, and identify protein targets for further investigation.
More recently, vascular injury has been associated with coronavirus disease 2019 (COVID-19) ARDS,
including the vascular complications of inflammation and thrombosis. In this context, COVID-19–induced injury to the vascular compartment has been associated with complement activation and microvascular thrombosis,
However, this focus on inflammation and thrombosis limits our insights into other disruptions associated with aberrant vascular activation, including angiogenesis, junctional barrier integrity, the role of activated platelets in vascular injury, and induction of vascular cell death, including specialized receptor-interacting protein kinase 3 (RIPK3)–mediated necrotic cell death. Specifically, although ANGPT2-mediated vascular disruption has been documented in COVID-19,
the association between ANGPT2 and induction of vascular cell death remains largely unexplored in ARDS investigations.
The purpose of this study was to assess whether aberrant vascular activation in COVID-19 was associated with the induction of necroptotic vascular cell death. To this aim, blood proteomics was performed in three independent COVID-19 cohorts, which enrolled patients at distinct time points in disease pathogenesis and included non–COVID-19 ARDS samples as control. Protein expression was linked to relevant clinical outcomes, vascular injury, and cell death markers in COVID-19 autopsy lung tissue.
Materials and Methods
Study Design
This study is an exploratory analysis of three cohorts that independently enrolled COVID-19 subjects at New York Presbyterian Weill Cornell Medical Center (WCM) between March 15 and August 17, 2020. Additional historic non–COVID-19 ARDS samples from influenza and bacterial ARDS patients prospectively enrolled into the Weill Cornell Biobank of Critical Illness from October 20, 2014, until May 24, 2020, were included as part of the ARDS cohort as ARDS controls. The three COVID-19 cohorts were identified according to ARDS status at enrollment, yielding an early hospitalization at-risk cohort, termed at risk; an intensive care unit (ICU) cohort with ARDS, termed ARDS; and a recovery cohort in early convalescence outside the ICU, termed recovery.
Cohort Descriptions
The at-risk cohort included 59 adult (aged >18 years) nonpregnant COVID-19 subjects admitted to the general wards of WCM with serum available and who did not meet ARDS criteria at study enrollment. The ARDS cohort included adult (aged >18 years) nonpregnant COVID-19 (N = 31) and historic non-COVID ARDS (N = 29) subjects admitted to the ICU at WCM. For the ARDS cohort, only study subjects meeting ARDS criteria and with blood sampling within 10 days of ICU admission were considered for analysis. The recovery cohort included 12 adult (aged >18 years) nonpregnant COVID-19 ARDS subjects with plasma samples available from both the time of ICU care and the subsequent recovery period to allow for longitudinal analyses.
Blood Sampling
In the at-risk cohort, between one and three consecutive daily samples were obtained from the central laboratory after routine processing to obtain serum. To obtain serum, blood collected in serum separator tubes was processed within 2 hours of venipuncture. Whole blood was centrifuged at 1500 × g for 7 minutes. The serum layer was aliquoted and stored at −80°C. These samples were obtained with a waiver of informed consent. In this cohort, samples collected after patient intubation were excluded from the analysis. In the ARDS and recovery cohorts, plasma was isolated from study subjects according to existing plasma isolation protocols.
To obtain plasma, blood collected in EDTA tubes was processed within 6 hours of venipuncture. Whole blood was centrifuged at 490 × g for 10 minutes. The plasma layer was removed in 200-μL aliquots and stored at −80°C. ARDS samples were obtained from patients in the intensive care unit, whereas recovery blood samples were obtained from patients convalescing in the hospital rehabilitation floors, as well as from the New York Presbyterian Weill Cornell Medicine Post-ICU recovery clinic.
Clinical Evaluation
Baseline clinical parameters and outcomes were extracted from the electronic medical record, as described previously.
Baseline comorbidities were manually extracted from the electronic medical record. Baseline clinical data (laboratory results, severity of illness, and ventilator data) were measured at time of blood sampling in both the at-risk cohort and ARDS cohort. In the recovery cohort, baseline clinical data were measured from the ICU time point to allow for direct comparison with the ARDS cohort. Severity of illness was defined by the Sequential Organ Failure Assessment score.
on behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure.
Two critical care investigators independently adjudicated the ARDS diagnosis. In all study subjects, COVID-19 was diagnosed if a subject had a syndrome compatible with COVID-19 and a nasopharyngeal swab positive for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) by reverse transcription PCR.
Recovery Evaluation
The EuroQol five-dimension, three-level (EQ-5D-3L) assessment was used to assess recovery at 12 months from ICU admission. The EQ-5D-3L is a self-assessment of the patient recovery, and considers five distinct domains (namely, mobility, self-care, usual activities, pain or discomfort, and anxiety or depression).
Each domain was scored as 0, 1, or 2, depending on whether the patient reported no, some, or extensive limitations in each respective domain. For each patient, a final score was defined as the sum of the scores across the five domains and treated as an ordinal variable in the statistical analysis. Maximal functional limitation would have a score of 2 × 5 = 10, whereas an optimal recovery would be scored 0.
Autopsy Studies
Twenty autopsies, performed between March 19 and June 30, 2020, with premortem nasopharyngeal swabs positive for SARS-CoV-2 were considered for lung tissue staining. Hematoxylin and eosin, ANGPT2, CD61, and phosphorylated mixed lineage kinase domain-like (pMLKL) staining were performed in all autopsy subjects. In addition, CD31 and ANGPT2/erythroblast transformation-specific-related gene (ERG) costaining were performed on the four autopsy subjects highlighted in the article.
All autopsies were performed in a negative pressure ventilation room with full personal protective equipment, including N-95 masks. No bone saw was used to prevent aerosolized dusts and, as such, brain examination was not performed. All tissues were immediately fixed in formalin for a minimum of 24 hours. To minimize exposure, only two individuals were allowed in the suite during the autopsy, and the room was disinfected before and after each case. Lung tissue specimens were fixed in 10% formalin for 48 to 72 hours. Hematoxylin and eosin staining was performed for all cases. Immunohistochemistry was performed for ANGPT2 (sc-74403; Santa Cruz Biotechnology, Dallas, TX; 1:100), CD61 (CD61 clone 2F2; Leica Biosystems, Deer Park, IL), ERG (ab92513; Abcam, Cambridge, UK; 1:100), CD31 (PA0250; Leica Biosystems; ready to use), and pMLKL (MAB91871; Novus Biologicals, Centennial, CO; 1:750 with casein for background reduction). Specimens were scanned by whole-slide image technique using an Aperio slide scanner (Leica Biosystems) with a resolution of 0.24 μm per pixel. Quantification of ANGPT2 and CD61 was performed on four random ×20 images selected using a random overlay of points and excluding fields with large vessels or airway. All 20 autopsies were analyzed using Immunohistochemistry profiler
as a plugin for ImageJ software version 1.52a (NIH, Bethesda, MD; https://imagej.nih.gov, last accessed March 10, 2022). After deconvolution of the ×20 images, both area of expression (eg, number of pixels) and intensity of expression (eg, intensity of pixel) were measured and combined into a single score, according to the equation score = (number of pixels in a zone × score of the zone)/total number of pixels in image. High, intermediate, low, and overall percentage positive were averaged over the four measurements. The median ANGPT2 quantification was used to define the high (greater than median) and low (less than median) ANGPT2 staining. The association between CD61 and ANGPT2 was then calculated on the basis of CD61 quantification in the low and high ANGPT2 groups using U-test.
O-link Plasma Proteomics
To quantify the circulating vascular proteome, plasma and serum samples from the at-risk, ARDS, and recovery cohorts were profiled using O-link through the Proteomics Core of Weill Cornell Medicine-Qatar. The O-link assays were performed using Inflammation version 3021, Cardiovascular II version 5005, and Cardiovascular III version 6113 panels (O-link, Uppsala, Sweden).
EDTA plasma and serum samples were heat inactivated at 56°C for 15 minutes. The protein measurements were performed with the Proximity Extension Assay technology, according to manufacturer's instructions. In summary, high-throughput real-time PCR of reporter DNA linked to protein-specific antibodies was performed on a 96-well integrated fluidic circuits chip (Fluidigm, San Francisco, CA). Signal quantification was performed on a Biomark HD system (Fluidigm, San Francisco, CA). Each sample was spiked with quality controls to monitor the incubation, extension, and detection steps of the assay. In addition, samples representing external, negative, and interplate controls were included in each analysis run. From raw data, real-time PCR CT values were extracted using the Fluidigm real-time PCR analysis software version 4.5.2 at a quality threshold of 0.5 and linear baseline correction. CT values were further processed using the O-link NPX manager software version 5.0 (O-link). Herein, log2-transformed CT values from each sample and analyte were normalized on the basis of spiked-in extension controls and scale inverted to obtain normalized log2 scaled protein expression values. Normalized log2 scaled protein expression values were further adjusted on the basis of the median of interplate controls for each protein and intensity median scaled between all samples and plates. As an external validation of the O-link platform, plasma protein levels of CD40 ligand (CD40LG) and ANGPT1 were measure by enzyme-linked immunosorbent assays (ELISAs; see section below) in high and low expression patient samples and correlation measured between O-link and ELISA protein values.
The at-risk cohort was profiled in two separate runs. The second run included a total of 11 samples, among which 5 bridge samples were used to scale this batch toward the first one, as recommended by O-link. First, for each bridge sample, the pairwise difference between the first and second batch was computed. An overall batch adjustment factor was then derived as the median of these pairwise differences and subtracted to the values in the second batch.
Subsequently, protein levels were exponentiated, normalized using probabilistic quotient normalization,
(k = 10). A total of 10 proteins were measured across multiple panels and, therefore, their duplicated values were averaged, leaving a total of 266 unique proteins. Protein values were standardized before statistical analysis. The preprocessed O-link and validation ELISA proteomics data for the three cohorts presented in this article are available (https://doi.org/10.6084/m9.figshare.19341536.v1, last accessed March 10, 2022).
ELISA Measurements
Plasma samples from the ARDS and recovery cohorts were used for ELISAs, according to manufacturer recommendations. Human ANGPT2 (R&D Systems, Minneapolis, MN; catalog number DANG20) and RIPK3 (Cusabio, Houston, TX; catalog number CSB-EL019737HU) kits were used to measure plasma protein levels. In addition, CD40LG (R&D Systems; catalog number DCDL40) and ANGPT1 (R&D Systems; catalog number DANG10) were measured in high- and low-expressing patient samples (N = 6 each, 12 total) to validate the O-link platform. Plasma samples were diluted (1:8 dilution for ANGPT2, 1:10 dilution for RIPK3, 1:8 dilution for ANGPT1, and 1:15 dilution for CD40L) before plating. Final sample absorbance was measured at 450 nm with wavelength correction performed at 570 nm. Sample concentrations were calculated from a four-parameter logistic curve generated from known standard concentrations. Dilution factors were accounted for to calculate the final sample concentration. Plasma protein values were log10 transformed before statistical analysis.
Statistical Analysis
In the at-risk cohort, proteomic analysis, protein associations with death (ie, whether the patient ended up dying), and platelet count (minimum value across the sampling days) were computed using a mixed linear effect model, which allows for proper accounting of the multiple samples collected per patient. The model was formulated as follows: protein ∼ outcome + replicate + batch + (1|patient), where outcome is either death or platelet count, replicate indicated the day of blood sample draw (first, second, or third since hospital admission), and batch indicated whether the sample was measured in the first or second run. Association P values were corrected for multiple testing using the Benjamini-Hochberg method for controlling the false discovery rate.
The protein vascular signature was selected from proteins significantly associated with both outcomes, and proteins significantly associated with either mortality or platelet count and with known, well-characterized links to vascular function. TIE2 was additionally included as it is the receptor for ANGPT2.
For all cohorts, patient hierarchical clustering based on the standardized proteomics value was performed using Ward linkage and euclidean distance. The differential analysis between patient clusters was performed using U-tests for continuous variables, Kendall rank correlation for ordinal variables, and log-rank tests for survival times. The correlation between ANGPT2 and RIPK3 was estimated using Pearson correlation. For these analyses, P < 0.05 was considered significant.
In the recovery cohort, patients were first divided into two groups based on unsupervised hierarchical clustering (Ward linkage, euclidean distance) performed on the recovery time point. Then, for each patient, a protein abundance difference (Δ) was calculated between the ICU and recovery time points. Finally, for each protein, it was determined whether the protein Δ was different across the two patient groups using the linear model Δ ∼ group. P values were corrected for multiple tests using the Benjamini-Hochberg method. Given the small sample size and validation of protein set in two prior cohorts, an adjusted P < 0.25 was considered significant.
All statistical analyses were performed in R 4.0.1 using the maplet package.
The study was approved by the institutional review board at WCM (20-05022072, 20-03021681, and 1811019771). 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 in the at-risk cohort).
Results
The Hospitalized At-Risk Cohort Blood Proteome Identifies a Signature of Vascular Limitation Preceding Critical Illness
A total of 1384 subjects were admitted to the medical floors of WCM during the study period. Of the 1384 hospitalized subjects with COVID-19, 59 were profiled (Figure 1). Profiled subjects were more likely to have cancer (15% versus 7.5%; P = 0.044) but otherwise were similar to unprofiled subjects (Supplemental Table S1). The median age of profiled subjects in the at-risk cohort was 69 years, and was majority male (64% male versus 36% female). Of the cohort, 53% had hypertension and 15% had cancer. Additional cohort characteristics are listed in Supplemental Table S2.
Figure 1Study design. A: Flowchart of study subjects selected for blood proteomic profiling in the at-risk, acute respiratory distress syndrome (ARDS), and recovery cohorts. B: Venn diagram showing the overlap among subjects with COVID-19 across the study cohorts. C: Conceptual schematic of study subject sampling in relation to hospital admission (at-risk cohort), intensive care unit (ICU) admission (ARDS cohort), and ICU discharge (recovery cohort). D: Characteristics of study cohorts, including total subjects (N), blood sample type, existence of longitudinal samples, sample timing, and whether proteomic profiling, angiopoietin 2 (ANGPT2), and receptor-interacting protein kinase 3 (RIPK3) measurements were performed.
Blood biomarkers in the at-risk cohort were measured using the O-link proximity extension assay. External validation of protein expression by ELISA showed excellent correlation between O-link and ELISA protein measurements: CD40LGOLink and CD40LGELISA was R = 0.94 (P < 0.001) (Supplemental Figure S1A), and ANGPT1OLink and ANGPT1ELISA was R = 0.94 (P < 0.001) (Supplemental Figure S1B). To first identify clinically relevant proteins associated with vascular injury and platelet thrombosis, a protein set was defined on the basis of the association of circulating proteins with death and platelet levels (Figure 2A ). Thirteen proteins were significantly associated with both parameters (false discovery rate, 0.1): platelet-derived growth factor subunit A (PDGFA), PDGFB, ANGPT1, sortilin 1 (SORT1), proheparin-binding EGF-like growth factor (HBEGF), latency-associated peptide transforming growth factor beta-1 (LAP TGFB1), CD84, C-X-C motif chemokine 5 (CXCL5), matrix metalloproteinase-9 (MMP9), plasminogen activator inhibitor 1 (PAI), interleukin-7 (IL7), interleukin-1 receptor antagonist (IL1RA), and C-X-C motif chemokine 1 (CXCL1). In addition, eight proteins were selected as they were associated with either death or platelet count (false discovery rate, 0.1) and have known vascular functions: a disintegrin and metalloproteinase with thrombospondin motifs 13 (ADAMTS13), CD40LG, epidermal growth factor receptor (EGFR), P-selectin (SELP), urokinase-type plasminogen activator (UPA), vascular endothelial growth factor A (VEGFA), platelet glycoprotein VI (GP6), and heme oxygenase 1 (HO1). TIE2 was additionally included because it is the receptor for ANGPT1 and ANGPT2.
The final set comprised of 22 proteins (Supplemental Figure S2), including proteins related to vascular junctional integrity (ANGPT1 and TIE2), angiogenesis (PDGFA and PDGFB), platelet degranulation (CD40LG and GP6), and coagulopathy (ADAMTS13 and PAI), highlighting the potential functional significance of the identified proteins. Notably, these representative vascular proteins had lower expression in at-risk subjects who died (Figure 2B), representing an early signal of vascular limitation in COVID-19 pathogenesis.
Figure 2The hospitalized at-risk cohort blood proteome identifies a signature of vascular limitation preceding critical illness. A: Overview of the associations of the protein set to death and platelets. All displayed proteins were included in the final protein set. Proteins in green associated with platelets, proteins in red associated with death, and proteins in black associated with both. TIE2 was additionally included as it is the receptor for angiopoietin (ANGPT) 1 and ANGPT2. B: Box plots demonstrating the association between proteins of vascular junctional integrity, angiogenesis, platelet degranulation, and coagulopathy to mortality in the at-risk cohort after adjusting for multiple comparisons. Boxes indicate the interquartile range (IQR) of the data distribution, the line in the box represents the median value, and the whiskers extend for 1.5 times the range of the IQR. Dots indicate the protein level in individual patients. C: Heat map of protein set abundance in the at-risk COVID-19 subjects. Hierarchical clustering was performed using Ward linkage and euclidean distance. Age, platelet count, and death are overlaid at the top. Mean abundance of the 22-protein set is displayed at the bottom. Mean protein abundance is progressively lower from cluster A to B to C. N = 59 (A–C).
Patient clustering based on this protein set identified three distinct patient groups (clusters A, B, and C) (Figure 2C), with mortality and low platelets progressively enriched. Interestingly, this mortality and low platelet enrichment was associated with lower mean abundance of the 22 proteins (P < 0.001) (Supplemental Figure S3A) and older age (P = 0.016) (Supplemental Figure S3B). Baseline cancer status (P = 0.043) (Supplemental Table S3) as well as creatinine (P = 0.011) (Supplemental Table S3) were additionally enriched across the patient clusters, identifying additional patient characteristics associated with low protein abundance.
Loss of Circulating Vascular Proteins Is Associated with Low Platelets, Mortality, and Plasma ANGPT2 in ARDS
Next, the vascular protein signature was evaluated in the ARDS cohort. A total of 439 subjects with COVID-19 ARDS were admitted to the WCM ICUs during the study period. Thirty-one subjects with COVID-19 ARDS were profiled, together with 29 non–COVID-19 ARDS controls selected from 57 total non–COVID-19 ARDS subjects in the ICU biobank (Figure 1). Profiled subjects with COVID-19 ARDS were younger (62 versus 67 years; P = 0.032), but otherwise were similar to unprofiled subjects with COVID-19 ARDS (Supplemental Table S4). There were no baseline differences between profiled and unprofiled non–COVID-19 ARDS control subjects (Supplemental Table S4). Among all profiled subjects with ARDS (N = 60), there were no significant age, sex, or race differences between COVID-19 ARDS (N = 31) and non–COVID-19 ARDS subjects (N = 29) in the cohort (Supplemental Table S2). Cancer was overrepresented in the non–COVID-19 ARDS control subjects (48.0% versus 3.2% in COVID-19 ARDS). There were also notable differences in respiratory physiology. COVID-19 ARDS was associated with more severe hypoxemia [partial pressure of arterial oxygen:fraction of inspired oxygen (PaO2:FiO2) ratio, 84 versus 193 in non–COVID-19 ARDS] but lower ventilator ratio (1.65 versus 2.89 in non–COVID-19 ARDS).
First, the vascular protein set was investigated only in subjects with COVID-19 ARDS (Supplemental Figure S4A). Protein results from the COVID-19 at-risk cohort were confirmed and low mean protein abundance of the protein set was associated with worse survival (P = 0.026) (Supplemental Figure S4B), low platelet count (P < 0.001) (Supplemental Figure S4C), and older patient age (P = 0.035) (Supplemental Figure S4D). The addition of non–COVID-19 ARDS patients (bacterial sepsis and influenza ARDS) led to a similar trend (Figure 3A ), with survival (P = 0.020) (Figure 3B) and low platelets (P < 0.001) (Supplemental Figure S5A) associated with the low mean vascular protein abundance cluster (P < 0.001) (Supplemental Figure S5B). Age was not associated with low mean vascular protein abundance in this ARDS cohort (P = 0.180) (Supplemental Figure S5C). Similar to the at-risk cohort, baseline cancer status (P < 0.001) (Supplemental Table S3) and elevated baseline creatinine (P = 0.022) (Supplemental Table S3) were more frequent in ARDS subjects in the low mean vascular protein abundance cluster. More importantly, this low abundance signature did not reflect the relative expression of these proteins compared with healthy controls. As an example, CD40LG and ANGPT1 had opposite expression patterns in ARDS compared with healthy controls. Whereas CD40LG increased in ARDS compared with healthy control (Supplemental Figure S1C), ANGPT1 expression was similar to healthy control (Supplemental Figure S1D). Despite the opposite expression patterns of the ARDS biomarkers compared with healthy controls, lower expression of both CD40LG and ANGPT1 was associated with the low platelet, worse clinical outcomes cluster.
Figure 3Loss of circulating vascular proteins is associated with low platelets, mortality, and plasma angiopoietin 2 (ANGPT2) in the acute respiratory distress syndrome (ARDS) cohort. A: Heat map of 22-protein set abundance in diverse ARDS subjects, divided into two clusters. Hierarchical clustering was performed using Ward linkage and euclidean distance. Age, log10(ANGPT2), platelet count, mortality, and ARDS etiology are overlayed at the top. Mean protein abundance of the 22-protein set is overlayed at the bottom. B: Kaplan-Meier survival analysis for the two heat map clusters showing worse survival for the high ANGPT2 ARDS cluster B. The x axis was capped at 60 days. The table at the bottom indicates the number of patients at risk at each time point in the two clusters. C: Log10(ANGPT2) values in the two clusters, demonstrating higher ANGPT2 expression in low vascular protein abundance cluster B. Differential statistic was assessed with a two-sided U-test. The boxes indicate the interquartile range (IQR) of the data distribution, the line in the box represents the median value, and the whiskers extend for 1.5 times the range of the IQR. Dots indicate the protein level in individual patients across the different ARDS categories: COVID-19 (orange), bacterial sepsis (brown), and influenza (yellow). N = 60 (A–C).
Similar to the at-risk cohort, the junctional integrity proteins ANGPT1 and TIE2 had lower expression in the low platelet, high mortality cluster B (Figure 3A). As ANGPT2 is known to negatively regulate ANGPT1 and TIE2, plasma ANGPT2 was measured in the ARDS cohort. Notably, plasma ANGPT2 was higher in the low mean protein abundance cluster (P = 0.001) (Figure 3C and Supplemental Figure S4E), linking low vascular protein abundance and plasma ANGPT2 in diverse ARDS subjects.
Interestingly, when COVID-19 ARDS was considered alone (Supplemental Figure S4), this higher vascular injury signature was present in 39% (12 of 31) of subjects with COVID-19 ARDS. However, when all three infection types were considered (Figure 3), only 13% (4 of 31) of subjects with COVID-19 ARDS were in the higher vascular injury cluster compared with 58% (14 of 24) of subjects with bacterial sepsis ARDS and 80% (3 of 4) of subjects with influenza ARDS. This demonstrated that COVID-19 ARDS may be associated with less ANGPT2-associated vascular injury than bacterial sepsis and influenza-related ARDS. This finding is supported by a lower ventilator ratio in subjects with COVID-19 ARDS compared with non–COVID-19 ARDS subjects (Supplemental Table S2), a physiological surrogate for vascular injury in ARDS.
This is also consistent with previous investigations showing higher platelet counts and less platelet consumption in COVID-19 compared with bacterial sepsis ARDS.
CRICS TRIGGERSEP Group (Clinical Research in Intensive Care and Sepsis Trial Group for Global Evaluation and Research in Sepsis) High risk of thrombosis in patients with severe SARS-CoV-2 infection: a multicenter prospective cohort study.
Angiopoietin 2 Correlates with CD61 Staining Microthrombi in Subjects with COVID-19 ARDS
The observed link between platelet activation and ANGPT2 was then explored in COVID-19 ARDS lung tissue. Twenty COVID-19 ARDS lung autopsy specimens were stained for the lung injury marker ANGPT2 and the activated platelet stain CD61. Representative sections from a high and low ANGPT2 subject are shown in Figure 4A . ANGPT2 staining was pronounced in the microvasculature and was mirrored by CD61-positive microthrombi in a similar distribution, linking vascular injury and platelet-rich microvascular microthrombi in COVID-19 ARDS. ANGPT2 and endothelial nuclear stain ERG costaining confirmed endothelial origin of the ANGPT2 staining (Supplemental Figure S6A). More importantly, staining for CD31 as a constitutive endothelial marker (Supplemental Figure S6B) showed that differences in ANGPT2 and CD61 staining in these subjects were not due to differences in preservation of endothelium between autopsy subjects. Quantification of ANGPT2 and CD61 staining showed that high ANGPT2 protein was associated with increased CD61 (P = 0.005) (Figure 4B). Blood proteomics was performed in three of the autopsy subjects (Figure 4, A and C) (subjects labeled as P1, P2, and P3). Consistent with the ARDS cohort, low protein abundance of the protein set, including lower expression of angiopoietin axis proteins ANGPT1 and TIE2 (Figure 4C), was associated with increased vascular ANGPT2 staining (Figure 4A).
Figure 4Angiopoietin 2 (ANGPT2) is correlated with CD61 staining microthrombi in subjects with COVID-19 acute respiratory distress syndrome (ARDS). A: Representative hematoxylin and eosin (H&E), ANGPT2, and CD61 staining in subjects with COVID-19 ARDS. H&E demonstrates alveolar septal wall thickening across displayed autopsy subjects (arrows). Increased ANGPT2 (open arrowheads) and CD61 (closed arrowheads) immunostaining is seen in subjects NA and P3 in a vascular distribution. B: Lung autopsy specimens from 20 subjects with COVID-19 ARDS were stained for ANGPT2 and platelet activation stain CD61. High ANGPT2 corresponds to autopsy subjects with ANGPT2 quantification above the median of the autopsy cohort, whereas low ANGPT2 represents the low ANGPT2 cohort. High ANGPT2 was associated with increased CD61 staining (P = 0.005). C: Blood proteomic data from autopsy subjects P1 and P2 (both low ANGPT2/low CD61 staining) and subject P3 (high ANGPT2 and high CD61) demonstrate that low expression of the vascular protein set is associated with high ANGPT2 and high CD61 staining. Angiopoietin axis proteins ANGPT1 and TIE2 highlighted in red. N = 10 for high ANGPT2 and N = 10 for low ANGPT2 (B). Scale bars = 50 μm (A). NA, no blood proteomic data available for the autopsy subject.
Induction of Vascular Cell Death in ANGPT-2–Associated Vascular Injury
Having validated the vascular injury signature in diverse ARDS populations and COVID-19 ARDS autopsy tissue, whether ANGPT2-associated vascular injury could be associated with genetically regulated necrotic cell death, known as necroptosis, was investigated next. Increased expression of plasma RIPK3, a vital necroptosis protein,
was first demonstrated in subjects with high ANGPT2 ARDS (P = 0.020) (Figure 5A ). Plasma RIPK3 was also correlated with plasma ANGPT2 (r = 0.40; P = 0.003) (Figure 5B), supporting the existence of a link between circulating necroptosis proteins and ANGPT2-associated vascular injury. Autopsy subjects with high ANGPT2 staining/low vascular protein abundance (Figure 4A) (NA and P3) demonstrated diffuse microvascular staining for pMLKL, a terminal protein in necrotic cell death execution downstream of RIPK3 (Figure 5C). Conversely, autopsy subjects with low ANGPT2 staining/high vascular protein abundance (Figure 4A) (P1 and P2) were negative for pMLKL. Taken together, these data link ANGPT2-associated vascular injury to necroptosis induction in COVID-19 ARDS.
Figure 5Induction of vascular cell death in angiopoietin 2 (ANGPT2)–associated vascular injury. A: Plasma receptor-interacting protein kinase 3 (RIPK3) in acute respiratory distress syndrome (ARDS) by heat map cluster (Figure 3A). RIPK3 is associated with high ANGPT2 (P = 0.020). COVID-19 (orange), bacterial sepsis (brown), and influenza (mustard) data are shown. B: Correlation of plasma RIPK3 and plasma ANGPT2 in the ARDS cohort (Figure 3). r indicates the Pearson correlation coefficient of the two variables, and P indicates its corresponding P value. The black line represents the linear regression line, and the gray area indicates the 95% CI of the fit. Dots indicate the protein level in individual patients across the different ARDS categories. C: Phosphorylated mixed lineage kinase domain-like (pMLKL) staining in COVID-19 ARDS autopsy subjects P1 and P2 (both low ANGPT2/low CD61 staining) and subjects NA and P3 (high ANGPT2 and high CD61), demonstrating increased expression of pMLKL (open arrowheads) in autopsy subjects with high ANGPT2 staining. N = 60 (A and B). Scale bars = 50 μm (C). NA, no blood proteomic data available for the autopsy subject.
Among Subjects with COVID-19 ARDS Recovery, Longitudinal Plasma Proteomics Identifies a Stable Protein Trajectory Associated with Good Functional Recovery
The 22-protein set was further investigated with respect to its predictive ability during COVID-19 ARDS recovery. Among 276 recovery subjects during the study period, 12 subjects with COVID-19 ARDS had plasma available from both their ICU and recovery time point and were profiled longitudinally (Figure 1). Profiled recovery subjects were younger (47 versus 62 years old; P = 0.002) than unprofiled recovery subjects (Supplemental Table S1). The median age of profiled recovery cohort was 47 years, and was majority male (67% versus 33% female) (Supplemental Table S2).
Patient clustering based on the recovery plasma protein set revealed two distinct clusters (Figure 6A ). Again, the low protein abundance cluster was associated with platelet level (P = 0.048) (Supplemental Figure S7A) and older age (P = 0.048) (Supplemental Figure S7B) but not ANGPT2 level (P = 0.68) (Supplemental Figure S7D). One-year follow-up functional recovery data based on the EQ-5D-3L questionnaire were available on 11 of these 12 recovery individuals (Figure 6A, see Materials and Methods for details). Notably, the cluster of patients with lower abundance of the protein set (P = 0.004) (Supplemental Figure S7C) displayed worse functional recovery 12 months after admission from the ICU, whereas higher vascular protein abundance was associated with better functional recovery (P = 0.027) (Figure 6B). To test whether the protein trajectory from ICU to recovery was different between good and poor functional recovery subjects, the differences in protein abundances was compared between the two time points in the two patient clusters (see Materials and Methods). For proteins representative of junctional barrier integrity (TIE2; P = 0.20), angiogenesis (PDGFA; P = 0.20), platelet degranulation (GP6; P = 0.20), and coagulopathy (PAI; P = 0.20), good functional recovery was associated with stable protein trajectory (Figure 6C), as opposed to the large protein changes among the poor recovery subjects. This stable trajectory among good functional recovery subjects was similar for platelet levels (P = 0.086) (Supplemental Figure S7E) and ANGPT2 (P = 0.083) (Supplemental Figure S7F).
Figure 6Among subjects with COVID-19 acute respiratory distress syndrome recovery, longitudinal plasma proteomics identifies a stable protein trajectory associated with good functional recovery. A: Heat map of COVID-19 recovery subjects. Functional recovery, age, platelet count, and 12-month recovery scores are overlaid at the top. Hierarchical clustering was performed with Ward linkage and euclidean distance. B: Follow-up recovery scores at 12 months after intensive care unit (ICU) admission in the two heat map clusters. Differential statistic was assessed with a two-sided U-test. The boxes indicate the interquartile range (IQR) of the data distribution, the line in the box represents the median value, and the whiskers extend for 1.5 times the range of the IQR. Dots indicate the protein level in individual patients. High scores indicate worse functional recovery. C: Trajectory of vascular proteins from ICU to recovery time points by functional recovery group. The boxes indicate the IQR of the data distribution, the line in the box represents the median value, and the whiskers extend for 1.5 times the range of the IQR. Dots indicate the protein level in individual patients in the two time points. Values from the same patient are linked by a line and colored according to the corresponding heat map cluster: A (cream) or B (red). Differential statistic of the protein trajectories between the two patient clusters was computed with a linear model. All displayed trajectory differences were significant to an adjusted P < 0.25. N = 12 (A–C).
This study traced a vascular protein signature through the natural history of COVID-19 ARDS from hospital admission to either recovery or death. As reflected in both the blood proteome and lung tissue, the clinical relevance of low abundance of circulating vascular proteins was demonstrated with known vascular functions and a link between ANGPT2 and vascular cell death, and in particular specialized necroptotic cell death, was revealed.
This vascular phenotype is notably present in certain subjects with COVID-19 before ICU admission. Although vascular injury spans the COVID-19 disease continuum from asymptomatic blue toes to catastrophic thromboembolic disease and ARDS-associated microangiopathy, the identification of broad loss of vascular protein expression in early severe disease generalizes this vascular dysfunction to the large population of hospitalized subjects with COVID-19. The loss of vascular proteins could result from SARS-CoV-2 endothelial infection.
was able to induce vascular cell death. Indeed, induction of genetically regulated necrotic cell death mediator (pMLKL) was demonstrated in the microvasculature of autopsy subjects with high ANGPT2-associated vascular injury COVID-19. Diverse upstream mediators previously linked to COVID-19 (eg, tumor necrosis factor-α
providing a crucial link between SARS-CoV-2 infection and both direct (virus) or indirect (tumor necrosis factor and interferons) induction of vascular cell death in subjects with COVID-19. Vascular cell death in COVID-19 ARDS is also supported by lung imaging mass cytometry studies that show a reduction in the absolute number of endothelial cells in late COVID-19 autopsy tissue, which could reflect disease-related vascular cell death.
The role of activated platelets in vascular injury and repair is also apparent in the data. Activated platelets amplify immune responses in early ARDS but also play an essential role in vascular repair. The consistently low platelet levels across the cohorts and the extensive microthrombi observed in the autopsy subjects imply a circulating milieu of platelet consumption. This milieu is supported by a blood signature of ongoing thrombolysis (high UPA and low PAI) and low levels of platelet-derived proteins (low SELP and GP6) in subjects with high ANGPT2-associated vascular injury. Relative loss of ADAMTS13, linked to secondary microangiopathy in COVID-19,
is similarly deficient in the high ANGPT2 subjects, linking platelet consumption with microangiopathy in severe COVID-19. Low platelets have previously been linked to ARDS mortality,
Consistently low circulating angiogenic (low PDGFA and PDGFB) and barrier protein (low ANGPT1) in the higher ANGPT2 and low platelet subjects imply limitations in these essential reparative processes.
The validation of the vascular phenotype across diverse causes of ARDS broadens the relevance of the reported findings. In linking low platelets, vascular function, and mortality in COVID-19, bacterial sepsis, and influenza ARDS, the study hints at a common final pathway of vascular injury that is more disease- (ARDS) than cause- (COVID-19) specific. Matching COVID-19 and non–COVID-19 ARDS subjects would also have strengthened any comparison between these groups. However, the fact that there were similarities in protein expression despite stark differences in clinical parameters, including a marked difference in baseline cancer prevalence [14/29 (48%) in non–COVID-19 ARDS versus 1/31 (3%) in COVID-19 ARDS], remains a strength of the analysis as it suggests that different causes of platelet depletion (eg, failure of production in cancer patients with non–COVID-19 ARDS and consumption in COVID-19 ARDS patients) leads to the same protein expression pattern and worse clinical outcomes. Of note, this vascular injury pattern may be related to a reduced baseline vascular resilience in subjects with high ANGPT2-associated vascular injury. Consistently, subjects with high vascular injury are older,
all variables known to be associated with vascular disease.
The identification of this severe vascular phenotype across infectious causes of ARDS also presents an opportunity for targeted vascular therapies in ARDS, including those that have shown promise in COVID-19,
NHLBI ARDS Network Latent class analysis of ARDS subphenotypes: a secondary analysis of the statins for acutely injured lungs from sepsis (SAILS) study.
and early human experimental therapies, including a ANGPT1 supplementation trial in subjects with COVID-19 (https://clinicaltrials.gov/ct2/show/NCT04737486, last accessed March 5, 2022). Although an ANGPT2 neutralizing antibody study in hospitalized patients with COVID-19 was stopped for futility in October 2020 (https://clinicaltrials.gov/ct2/show/NCT04342897, last accessed March 5, 2022), the present data could improve patient selection for similar trials in the future, including the use of platelet levels to identify subjects with vascular limitation.
Finally, the identification of a vascular recovery proteome is novel. Nearly 4.5 million patients have been hospitalized in the United States since the start of the COVID-19 pandemic (https://www.cdc.gov/coronavirus/2019-ncov/covid-data/covidview/index.html, last accessed March 5, 2022), with countless more hospitalized worldwide. However, even in recovery, patients remain at risk for disease-related morbidity and mortality.
This study demonstrated that a stable circulating vascular proteome is important for functional recovery. This association between vascular stability, platelet levels, and functional recovery could also support platelet levels as a novel biomarker in ARDS recovery. Larger studies will be needed to validate this observation.
This study has some limitations. Although a milieu of platelet consumption in subjects with ARDS and increased vascular injury was described here, alternative mechanisms of platelet depletion in these subjects, including decreased bone marrow production, particularly in the ARDS subjects with malignancies, cannot be ruled out. Platelet depletion also tracks with the vascular protein signature in the three cohorts, yet the study failed to establish cohort-specific platelet value that clinicians can use to identify these subjects. Although endothelial cells, pericytes, and platelets represent the likely cellular origin of the vascular signature, the lack of spatial proteomic or transcriptomic data allows for the possibility that some identified proteins are from nonvascular sources, including immune cells.
Future murine cell-specific knockout studies will be needed to resolve the cellular origin of the signature. Finally, although an association between ANGPT2, vascular cell death proteins RIPK3, and pMLKL was observed, additional in vivo and in vitro studies will be needed to determine the critical cross talk between vascular injury and vascular cell death proteins.
In summary, this study identified a vascular injury signal in COVID-19 ARDS that has predictive value in early disease through recovery as well as in bacterial sepsis and influenza ARDS. This signal was able to improve patient selection and timing of vascular targeted therapies in ARDS.
Acknowledgment
We thank Ilias I. Siempos for critically reviewing the article.
Author Contributions
D.R.P. and E.B. share first authorship; D.R.P. is listed first based on higher total effort to the project. D.R.P., E.B., J.K., A.M.K.C., and S.R. designed the study; D.R.P., A.C.R., and A.C.B. performed the autopsy staining analyses; L.G.-E., S.A.-M., A.C., C.N.P., A.R., J.G.C., and S.Z.J. processed samples and organized the patient clinical data; E.B., H.S., R.B., M.B., K.C., F.S., and J.K. analyzed the proteomic data; K.L.H. and I.E. provided statistical support for patient clinical data; E.L., K.W., C.N.P., and L.L. performed functional assessment of recovery subjects; D.R.P., E.B., J.K., A.M.K.C., R.B., F.S., J.G.C., E.J.S., A.C.R., H.O.R., J.L., M.E.C., and S.R. critically appraised the final data set; D.R.P. and E.B. wrote the article. All authors approved the final article.
O-link validation and healthy control versus acute respiratory distress syndrome (ARDS) CD40L and angiopoietin (ANGPT) 1 plasma levels. A and B: CD40L and ANGPT1 levels from 12 ARDS subjects were measured first by O-link and then by enzyme-linked immunosorbent assay (ELISA). Correlation plots show excellent correlation: CD40LGOLink and CD40LGELISA was R = 0.94 (P < 0.001), and ANGPT1OLink and ANGPT1ELISA was R = 0.94 (P < 0.001). C and D: Plasma levels of CD40L and ANGPT1 were measured by ELISA in healthy control and matched subjects with ARDS. C: CD40L levels were significantly higher in ARDS compared with healthy control but similar (blue circle) among subjects with ARDS in the low protein abundance/high ANGPT2 group (Figure 3A). D: ANGPT1 levels were similar among ARDS and healthy controls but lower (blue circles) among subjects with ARDS in the low protein abundance/high ANGPT2 group (Figure 3A). N = 12 ARDS subjects (A and B); N = 12 healthy control subjects (C and D); N = 9 matched ARDS subjects (C and D).
Protein subset selection. Proteins in bold represent the final protein subset. To identify these proteins, proteins associated with both death (adjusted P < 0.1) and platelet level (adjusted P < 0.1) were considered first, as indicated in the overlap region. In addition, proteins associated with either death or platelet level with known essential vascular functions were included (ADAMTS13, CD40LG, EGFR, SELP, UPA, and VEGFA). These proteins are annotated with an asterisk in the figure. Finally, TIE2 was included in the final subset based on its role as the receptor for angiopoietin 1 (ANGPT1) and ANGPT2.
Mean protein abundance and age by patient cluster in the at-risk cohort. A: Successive patient clusters based on the protein set abundance demonstrate significant loss of protein set expression (P < 0.001). B: Patient clusters are characterized by significantly different age distributions (P = 0.016). The boxes indicate the interquartile range (IQR) of the data distribution, the line inside the box represents the median value, and the whiskers extend for 1.5 times the range of the IQR. Dots indicate values in individual patients. Differential statistic was assessed with a two-sided Wilcoxon rank test. N = 59 (A and B).
Loss of essential vascular proteins is associated with low platelets, mortality, and angiopoietin 2 (ANGPT2) levels in COVID-19 acute respiratory distress syndrome (ARDS). A: Heat map of protein set abundance in COVID-19 ARDS patients. Age, log10(ANGPT2), platelet count, and mortality are overlayed at the top. B: Kaplan-Meier plot for the two heat map clusters. The test statistic was estimated with a log-rank test. The x axis was capped at 60 days. The table at the bottom indicates the number of patients at risk at each time point in the two clusters. C–E: Platelet count, age, and log10(ANGPT2) values in the two heat map clusters, respectively. The boxes indicate the interquartile range (IQR) of the data distribution, the line inside the box represent the median value, and the whiskers extend for 1.5 times the range of the IQR. Dots indicate values in individual patients. Differential statistic was assessed with a two-sided Wilcoxon rank test. N = 31 (A–E).
Platelet count (A), mean protein abundance (B), and age (C) distribution in the two patient clusters defined in the acute respiratory distress syndrome (ARDS) cohort. Both platelet count and mean protein abundance were significantly lower in cluster B (P < 0.001). There were no significant age differences between the two patient clusters (P = 0.18). The boxes indicate the interquartile range (IQR) of the data distribution, the line inside the box represents the median value, and the whiskers extend for 1.5 times the range of the IQR. Dots indicate values in individual patients across the different ARDS categories: COVID-19 (orange), bacterial sepsis (brown), and influenza (mustard). Differential statistic was assessed with a two-sided Wilcoxon rank test. N = 60 (A–C). ANGPT2, angiopoietin 2.
Angiopoietin 2 (ANGPT2)/erythroblast transformation-specific-related gene (ERG) and CD31 staining of acute respiratory distress syndrome (ARDS) lung tissue. A: ANGPT2/ERG costaining in COVID-19 ARDS subjects shows ERG-positive, ANGPT2-negative endothelial cells (closed arrowheads) in autopsy subjects with low ANGPT2 staining and ERG-positive, ANGPT2-positive endothelial cells (open arrowheads) in autopsy subjects with high staining ANGPT2. B: CD31 staining shows similar expression (arrows) among subjects with COVID-19 ARDS previously shown to have high and low ANGPT2 staining. Scale bars = 50 μm (A and B). NA, no blood proteomic data available for the autopsy subject.
Differential analysis of clinical parameters in the recovery cohort. A–D: Platelet count, age, mean protein abundance, and log10[angiopoietin 2 (ANGPT2)] level in the two patient clusters defined in the recovery cohort. The boxes indicate the interquartile range (IQR) of the data distribution, the line inside the box represents the median value, and the whiskers extend for 1.5 times the range of the IQR. Dots indicate values in individual patients. Differential statistic was assessed with a two-sided Wilcoxon rank test. E and F: Platelet and log10(ANGPT2) trajectories from the intensive care unit (ICU) time point to the recovery time point. The boxes indicate the IQR of the data distribution, the line inside the box represents the median value, and the whiskers extend for 1.5 times the range of the IQR. Dots indicate values in individual patients. Values from the same patient are linked by a line, where the color indicates the corresponding heat map group: cluster A (cream) or cluster B (red). Differential statistic of the protein trajectories between the two patient clusters (good versus poor functional recovery) across the two time points was computed with a linear model. N = 12 (A–F).
Supported by the Biomedical Research Program at Weill Cornell Medicine in Qatar, a program funded by the Qatar Foundation ; the NIH National Institute of Aging award 1U19AG063744 (J.K.); the NIH National Heart, Lung, and Blood Institute grant T32HL134629-01A1 (D.R.P.); and Stony Wold-Herbert Fellowship Fund (United States; D.R.P.).
D.R.P. and E.B. share first authorship.
J.K. and S.R. share last authorship.
Disclosures: A.M.K.C. is a cofounder, is a stock holder, and serves on the Scientific Advisory Board for Proterris, which develops therapeutic uses for carbon monoxide. A.M.K.C. also has a use patent on carbon monoxide. The spouse of M.E.C. is a cofounder and shareholder and serves on the Scientific Advisory Board of Proterris, Inc. E.J.S. reports consulting fees from Axle Informatics for work assisting National Institute of Allergy and Infectious Diseases with coronavirus disease 2019–related vaccine clinical trials. All other authors declare no competing interest.
00114-6&id=gr1.jpg){kind=link}
00114-6&id=gr2.jpg){kind=link}
00114-6&id=gr3.jpg){kind=link}
00114-6&id=gr4.jpg){kind=link}
00114-6&id=gr5.jpg){kind=link}
00114-6&id=gr6.jpg){kind=link}