Plasma Proteomics of COVID-19 Associated Cardiovascular Complications: Implications for Pathophysiology and Therapeutics

Cardiovascular complications are common in COVID-19 and strongly associated with disease severity and mortality. However, the mechanisms driving cardiac injury and failure in COVID-19 are largely unknown. We performed plasma proteomics on 80 COVID-19 patients and controls, grouped according to disease severity and cardiac involvement. Findings were validated in 305 independent COVID-19 patients and investigated in an animal model. Here we show that senescence-associated secretory proteins, markers of biological aging, strongly associate with disease severity and cardiac involvement even in age-matched cohorts. FSTL3, an indicator of Activin/TGFβ signaling, was the most significantly upregulated protein associated with the heart failure biomarker, NTproBNP (β = 0.4;padj=4.6x10−7), while ADAMTS13, a vWF-cleaving protease whose loss-of-function causes microvascular thrombosis, was the most downregulated protein associated with myocardial injury (β=−0.4;padj=8x10−7). Mendelian randomization supported a causal role for ADAMTS13 in myocardial injury. These data provide important new insights into the pathophysiology of COVID-19 cardiovascular complications with therapeutic implications.


Introduction
Since its discovery in December 2019, the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has resulted in ~ 160 million cases of coronavirus disease 2019 (COVID-19) and nearly 3.3 million deaths globally. 1 Although viral pneumonia is the predominant clinical manifestation of severe COVID-19, cardiovascular complications occur frequently and are independently associated with morbidity and mortality. Among hospitalized COVID-19 patients, myocardial injury occurs in ~ 20-62%, 2-6 vascular thrombosis in ~ 16%, 7 and acute heart failure in ~ 2.5%. 8 Acute heart failure in COVID-19 often occurs in those without preexisting heart failure, and is associated with mortality rates approaching 47%. 8 Myocardial injury, the most common cardiac complication in COVID-19, also remains a strong predictor for adverse outcomes and is associated with ≥ 2-fold increase of in-hospital mortality. 6 There are multiple potential contributors to cardiac injury and dysfunction in COVID-19. 9-13 While myocardial injury, indicated by elevations in circulating cardiac troponins, is common in COVID-19, [2][3][4][5]14,15 both Type I myocardial infarction caused by epicardial coronary artery occlusion and myocarditis appear rare, 11,12 suggesting other mechanisms drive most COVID-19-related cardiac injury and failure. We systematically examined circulating proteins in controls and COVID-19 patients to identify potential drivers of cardiovascular complications. These studies support inferences into the pathogenesis of COVID-19-induced cardiac injury and implicate pathways amenable to intervention in this process.

Characteristics of discovery study participants
The cohort for the case control plasma proteomics discovery study had a mean age of 64 years and 41% women (Table 1). There were no signi cant differences in age or sex between COVID-19 patients and controls, though controls included more white subjects and there was a non-signi cant trend of higher age in moderate COVID-19 patients with cardiac involvement due to limited samples available for matching. Controls had higher body mass index, but less coronary artery disease, diabetes, and cancer.
There were no differences in baseline hypertension, hyperlipidemia, myocardial infarction, stroke, liver or pulmonary disease, except for a higher prevalence of obstructive sleep apnea in controls. Among moderate COVID-19 patients, those with cardiac involvement had more chronic kidney disease and baseline anticoagulation use. In severe COVID-19 patients, cancer was more prevalent in those with higher levels of cardiac involvement. Cardiac involvement in COVID-19 patients was generally associated with higher high-sensitivity cardiac troponin T (hsTnT), N-terminal pro-B natriuretic peptide (NTproBNP), C-reactive protein (CRP), and D-dimer levels (Supplementary Table 1). There was no signi cant difference in PE and DVT, although event rates were relatively low. Severe COVID-19 patients with high degrees of cardiac involvement had the highest in-hospital mortality (33%) (Supplementary Table 1).
Plasma proteins associated with senescence and microvascular in ammation increase with worsening

COVID-19 severity and cardiac involvement
To rst determine if the plasma proteomic pro les distinguish subjects according to COVID-19 disease severity and/or cardiac involvement, we performed unsupervised principal component analysis in the discovery cohort. Strikingly, the rst principal component (PC1), representing the maximum variance in the dataset, clustered most subjects into their pre-speci ed group (Fig. 1a). This suggests the plasma proteome not only distinguishes between subjects with and without COVID-19, but can also accurately discriminate among varying COVID-19 disease severity and cardiac involvement. Notably, cardiac TnT and NTproBNP, proteins used to assign subjects to the different groups, were not the dominant proteins driving PC1 ( Supplementary Fig. 2). Rather, enrichment analysis of PC1 proteins identi ed innate immune responses, senescence, RNA processing, and cell injury as the most signi cant biological processes that discriminate among these ve groups ( Fig. 1b; Supplementary Table 2). Interestingly, despite groups being age-matched, proteins in a senescence-associated secretory phenotype (SASP-1) set (Supplementary Table 3) 16 , markers of biological aging, emerged as the most upregulated process (normalized enrichment score [NES] = 2.2; p adj =0.01) (Supplementary Table 2). To con rm this nding, an additional SASP-2 set was generated from a literature search of all published SASP proteins (Supplementary Table 3) and was also highly enriched in PC1 (NES = 2.0; p adj =0.01). Subgroup analyses consistently showed SASP-2 to be the most signi cantly upregulated process associated with cardiac involvement in both moderate (NES = 1.6, p adj =0.03) and severe (NES = 1.7, p adj =0.02) COVID-19 (Supplementary Tables 4-7). In an independent validation cohort of 305 COVID-19 patients, the two SASP protein sets were the pathways most signi cantly associated with disease severity (Fig. 1b; Supplementary Table 8). Interestingly, in transcriptional pro les of lungs from hamsters infected with SARS-CoV2, we also saw marked enrichment of both SASP-1 (NES = 1.8; FDR q-value = 0.001) and SASP-2 (NES = 2.2; FDR q-value < 0.001) compared to age-matched non-infected controls (Fig. 1c, Supplementary Table 9), suggesting that this age-related cellular senescence pro le is induced by COVID-

ADAMTS13 de ciency in COVID-19 related myocardial injury
To gain insight into the drivers of myocardial injury in COVID-19, we regressed our datasets on cardiac troponin (TnT) SOMAmer levels, which strongly correlated with clinical hsTnT concentrations (r = 0.8, p = 3.1x10 − 13 , Supplementary Fig. 5). 1,143 proteins differed signi cantly after regression on TnT (Fig. 3a, Supplementary Table 12). The proteins with the most signi cant positive association with cardiac TnT were mostly intracellular myocyte proteins, likely re ecting myocardial necrosis. Interestingly, the protein with the most signi cant negative association with cardiac TnT was ADAMTS13 (β coe cient= -0.4; p adj =8x10 − 7 ). Given the marked and progressive decline in ADAMTS13 levels observed with increasing cardiac involvement along with its strong correlation with TnT ( Fig. 3b and 3c), we looked to see if a similar phenomenon was occurring in the COVID-19 patients from the MGH Emergency Department validation cohort. Indeed, in these 305 patients, plasma ADAMTS13 levels displayed a similarly robust inverse correlation with cardiac troponin I (r=-0.4, p = 2x10 − 14 , Supplementary Fig. 6) and decreased further as COVID-19 progressed (p = 2.2x10 − 15 ; Fig. 4a). Moreover, lower ADAMTS13 levels on presentation were associated with overall disease severity (p = 1.5x10 − 6 ) and 28-day mortality (p = 1.2x10 − 6 ) ( Fig. 4b and 4c).
To assess a potential causal role of decreased ADAMTS13 in myocardial injury, we performed Mendelian randomization (MR) analysis in 463,010 subjects in the UK BioBank using intronic cis-pQTLs, genetic determinants of ADAMTS13 protein levels that map to the gene itself. Since these genetic alleles are inherited independent of confounders, a positive association provides evidence that ADAMTS13 levels are in a causal pathway with cardiac injury, even outside the severe effects seen in TTP. Indeed, we found that lower genetically-determined levels of plasma ADAMTS13 were associated with higher likelihood of clinically diagnosed myocardial injury in a large, general population (IVW β= -2.34x10 − 4 ; p = 6.4x10 − 4 ) ( Fig. 5 and Supplementary Fig. 7), suggesting that reduced ADAMTS13 levels increase vulnerability to cardiac injury.
To further investigate why circulating ADAMTS13 levels are lower in severe COVID-19 patients, we looked to see if ADAMTS13 gene expression changed in the SARS-CoV2 infected hamsters. Similar to humans, hamsters have severe endothelialitis associated with SARS-CoV2 infection 9,22 . We found evidence of vascular thrombus formation in the lungs and small vessel brin deposition and cardiomyocyte degeneration consistent with microvascular compromise in infected animals ( Fig. 6a-f). Compared to naïve controls, ADAMTS13 expression was 67% lower in SARS-CoV2 infected animals (p adj =1x10 − 5 ; Fig. 6c), suggesting that at least part of the decrease in circulating ADAMTS13 seen in those with severe COVID-19 is through decreased synthesis of this anti-thrombotic protein. This was associated with a 3.8fold increase in vWF expression (p adj =1.2x10 − 18 ; Fig. 6c).
Increased TGFβ superfamily signaling associated with the heart failure biomarker, NTproBNP, in COVID-19 Although cardiac microvascular thrombosis is likely a major contributor to myocardial injury in COVID-19, it is only identi ed in a subset of patients. 11,12 Non-ischemic etiologies, such as right ventricular strain, myocarditis, and stress cardiomyopathy, are other sources of cardiac injury and dysfunction in COVID-19, which often result in high circulating levels of both cardiac troponins and the heart failure biomarker, NTproBNP. 11,23,24 While ADAMTS13 showed an inverse relationship with NTproBNP, the correlation was modest (r=-0.3; p = 0.03, Fig. 3c), suggesting other processes are likely more substantial contributors to myocardial stress in COVID-19. After regression on NTproBNP Somamer levels, which strongly correlated with the clinical assay (r = 0.8; p = 3.8x10 − 12 ; Supplementary Table 13). FSTL3 levels correlated strongly with both NTproBNP and cardiac TnT (r = 0.7, p < 0.001 for both), and were markedly higher in both moderate and severe COVID-19 patients with cardiac involvement compared to their respective controls (Figs. 7b and 7c, Supplementary Tables 4 and  6).
MR analysis was unable to be performed for FSTL3 because only a single pQTL (rs12986335) has been identi ed for this protein and this pQTL did not reach genome-wide signi cance. Because FSTL3 is an indirect marker of increased TGFβ and Activin receptor signaling rather than part of the causal pathway, we would not anticipate a causal relationship between FSTL3 and heart failure. Since expression of FSTL3 is induced by activation of TGFβ and Activin receptors, which can be initiated by multiple ligands of the TGFβ superfamily, including TGFβ, Activins, bone morphogenic proteins (BMPs), and growth differentiation factors (GDFs), 25,26 we looked to see which of these ligands might be contributing to the increased circulating FSTL3 observed in COVID-19. Fourteen of the relevant ligands are measured on the SOMAscan platform, and twelve of these positively correlated with FSTL3 ( Supplementary Fig. 10). To determine which of these are likely responsible for the association of FSTL3 with increased myocardial stress in COVID-19, we exposed isolated cardiomyocytes to the various ligands, of which only Activins and TGFβ robustly increased cardiomyocyte FSTL3 expression ( Supplementary Fig. 11). Of these proteins, TGFβ1 was the most signi cantly increased in COVID-19 patients (74% higher than controls, p adj =7.2x10 − 11 ), suggesting it may be the primary circulating ligand responsible for the increase in FSTL3 observed in COVID-19 patients with elevated NTproBNP. Although we were unable to determine if TGFβ1 expression is directly increased in the hearts of those with COVID-19, RNAseq pro les of lungs from hamsters infected with SARS-CoV2 indicated that TGFβ1 signaling is indeed upregulated in animals with severe infection (NES = 1.5; FDR q-value = 0.03, Supplementary Fig. 12).

Discussion
Cardiovascular complications are common in COVID-19 and are strongly associated with disease severity and mortality. Despite this, the principal drivers of cardiac injury and dysfunction remain unclear and no effective therapies for these complications of COVID-19 have been identi ed. To gain mechanistic insights into the underlying pathophysiology, we systematically compared circulating plasma proteins in controls and COVID-19 patients across a range of disease severity and cardiac involvement. Key ndings from our 80 subject discovery cohort were not only validated in a distinct, longitudinal cohort of 305 COVID-19 patients, but also investigated in an experimental animal model of severe COVID-19. Our analyses identi ed alterations in two biological pathways that strongly associated with cardiac injury in COVID-19: a reduction in the anti-thrombotic protein, ADAMTS13, and an increase in TGFβ superfamily signaling, previously linked to cardiac dysfunction, injury, and brosis. 25,27 These changes, along with an increase in senescence-associated secretory proteins, were also associated with disease severity and mortality. Notably, these pathways are potentially amenable to therapeutic intervention with agents already in clinical trials or FDA approved for other indications. 28,29 Particularly striking was the marked ADAMTS13 reduction that progressed with COVID-19 duration and severity. Of the 4996 protein analytes assessed in our discovery study, ADAMTS13 was not only the most signi cantly decreased protein in severe COVID-19, but also displayed the strongest inverse association with myocardial injury. Autopsies have found evidence of microvascular thrombosis in multiple organs, including the heart, in COVID-19 patients. [9][10][11][12]30 Given its role in thrombotic microangiopathies, 21 the possibility that an acquired ADAMTS13 de ciency or local imbalance with its substrate, vWF, could be driving thrombotic complications in COVID-19 has been raised. [30][31][32] Our data provide new evidence supporting an important role for reduced ADAMTS13 in COVID-19 related myocardial injury. Mendelian randomization analysis of genetic variants within the ADAMTS13 gene strongly support a causal relationship between plasma ADAMTS13 protein levels and myocardial injury. Furthermore, the decreased mRNA expression of ADAMTS13 seen in the hamster model after SARS-CoV2 infection suggests that the decline in circulating ADAMTS13 in severe COVID-19 is at least partially due to a decrease in its synthesis. Given recent safety concerns of therapeutic systemic anticoagulation in critically ill COVID-19 patients 33 , speci cally replenishing ADAMTS13 in patients with low levels could provide a safer, targeted, and potentially more effective strategy for preventing vascular thrombosis in patients with severe COVID- 19. Notably, recombinant ADAMTS13 has demonstrated safety in patients with congenital TTP, 28 and recent pilot testing of ex vivo plasma samples from patients with severe COVID-19 suggests it can effectively reduce the heightened vWF activity and increased ultra-high molecular weight vWF multimers observed in COVID-19 34 . Based on our collective ndings in humans and animals with COVID-19, we propose that targeted ADAMTS13 repletion warrants evaluation as a tailored approach to preventing myocardial injury and possibly other thromboembolic complications in patients with severe COVID-19 and reduced ADAMTS13 levels.
In addition to microvascular thrombosis, other non-ischemic etiologies, including stress cardiomyopathy and right ventricular strain, also contribute to myocardial necrosis in COVID-19, often resulting in substantial increases in not only cardiac troponins but also the heart failure biomarker, NTproBNP. 11,23,24 Indeed in our analysis of the plasma proteome, we found the dominant proteomic signal associated with NTproBNP was not ADAMTS13, but rather FSTL3, a marker of TGFβ superfamily signaling. 25 FSTL3 has been identi ed as one of only four circulating proteins (out of 1310 assayed) that potentially distinguishes between patients with acute stress cardiomyopathy in comparison to acute myocardial infarction. 35 It is possible that cardiac dysfunction driven by TGFβ superfamily signaling is acutely reversible as seen in stress cardiomyopathy, consistent with some clinical observations in COVID-19. 36 However, the close association of TGFβ superfamily signaling with brosis raises the possibility that these pathways may also contribute to some of the long-term sequelae beginning to be appreciated in survivors of severe COVID-19. Importantly, multiple TGFβ superfamily inhibitors, including those currently being investigated in clinical trials, have shown bene ts in preclinical models of heart failure, as well as acute lung injury. 25,37−39 While these data suggest that this class of reagents could attenuate cardiac and pulmonary injury in COVID-19, given the immunomodulatory roles of TGFβ, 40 careful patient selection and optimal timing of such interventions will be critical to maximizing therapeutic bene t while minimizing adverse effects on immune responses to the virus.
Lastly, our proteomics analysis detected a novel association between senescence associated secretory proteins, markers of biological aging, with disease severity and cardiac involvement in COVID-19. While age is a recognized risk factor for adverse COVID-19 outcomes, since study subjects in our discovery study were mostly age-matched, this nding suggests that biological aging may represent a more accurate marker of COVID-19 severity than chronological aging alone. Indeed, in the validation cohort, a subject's SASP pro le on presentation was the most signi cant pathway associated with disease severity over the subsequent 28 days. Furthermore, the marked SASP enrichment seen in hamsters with severe COVID-19, suggests that SARS-CoV2 infection induces an increase in SASP production. This is consistent with the increase in circulating SASP proteins observed in patients with more severe COVID-19, which interestingly has also been linked to an increase in vascular thrombosis. 41 Of course, these possibilities are not mutually exclusive. It is possible that pre-existing biological aging, re ected in circulating SASP proteins, puts patients at higher risk for severe COVID-19 and that SARS-CoV2 infection also induces cellular senescence, further increasing circulating SASP proteins in a positive feedback loop. While targeting senescence has been proposed as a potential therapeutic strategy for severe COVID-19 42,43 , these are the rst data to report an increase in SASP pro les in COVID-19 patients that tracks with disease severity. Understanding the role of senescence in COVID-19 pathophysiology could be important in the context of therapeutic development and as vaccine development and rollout continue over the coming year. 44 Several limitations of our study are worth noting. First, observational data cannot de nitively establish causality. In the case of ADAMTS13, however, additional inferences from genetics strongly support a pathogenic role. Further studies of SARS-CoV2 infection in Syrian hamsters also mechanistically link viral infection to reduced ADAMTS13 expression. Second, the discovery study sample sizes were relatively small. In this context, the robust validation of our key ndings in a larger dataset generated using an independent proteomic technology is reassuring. Finally, while promising candidates were identi ed that might be targeted to mitigate COVID-19 cardiac complications, these hypotheses require rigorous additional testing. Importantly, our clinical observations strongly resonate with ndings in the hamster model of severe COVID-19, further supporting an important role of these processes in COVID-19 pathophysiology. These ndings also suggest this model could provide a preclinical tool for testing the safety and therapeutic e cacy of currently available reagents targeting these candidates as a prelude to clinical trials.
In summary, systematic analysis of circulating plasma proteins revealed important insights in COVID-19 cardiovascular pathophysiology and provides a foundation for novel therapeutic approaches to prevent or treat cardiac complications of COVID-19. Further investigation is warranted to determine if restoring ADAMTS13 levels, inhibiting TGFβ superfamily signaling, or modulating senescence can mitigate cardiovascular complications and improve outcomes in COVID-19 patients.

Methods
Discovery Study: To identify proteins that change with cardiac involvement in COVID-19, a case-control plasma proteomics study was performed in 80 subjects whose plasma samples were available from the multi-institutional Massachusetts Consortium on Pathogen Readiness and the Mass General Brigham (MGB) Biobank. COVID-19 subjects were recruited and plasma samples were collected with informed consent and IRB approval between March and June 2020 at Massachusetts General Hospital and Brigham and Women's Hospital in Boston, MA. Subjects were chosen based on data from their electronic medical records, to ll ve pre-speci ed groups representing a range of disease severity and cardiac involvement. To reduce the confounding in uence of factors previously linked to COVID-19 severity, groups were matched as much as possible for age, sex, and race. Patients enrolled in immunomodulator clinical trials or receiving such agents off-label were excluded. Study groups included non-COVID-19 controls; COVID-19 patients with moderate disease (hospitalized, but not requiring intensive care unit [ICU] level of care) with or without cardiac involvement; and COVID-19 patients with severe disease (requiring ICU care) with low or high degrees of cardiac involvement (Table 1). In patients with moderate COVID-19, cardiac involvement was de ned as at least one of the following near time of plasma collection: a high-sensitivity cardiac troponin T (hsTnT) > 14ng/L, N-terminal pro-B natriuretic peptide (NTproBNP) > 1,800pg/ml, and/or cardiac dysfunction documented by transthoracic echocardiography (left ventricular ejection fraction [LVEF] < 50%, regional wall motion abnormality, right ventricular systolic dysfunction, or > 25% relative decrease from prior LVEF). The hsTnT threshold of 14 ng/L is consistent with the Fourth Universal De nition of Myocardial Infarction, 45 which de nes myocardial injury as a troponin concentration greater than the 99th percentile upper reference limit. For severe COVID-19 patients in the ICU, who have a high prevalence of mild troponin elevation, we used a hsTnT threshold of 20 ng/L to distinguish between low and higher levels of myocardial injury, while the remainder of the criteria were the same. Clinical outcomes included intubation, vasopressor requirement, neuromuscular blockade use, sedation, renal replacement therapy (RRT), pulmonary embolism (PE), deep venous thrombosis (DVT), and in-hospital mortality. Plasma Proteomic Pro ling All blood samples obtained from subjects in the discovery study were anticoagulated with EDTA and collected plasma was frozen at -80C and underwent two freeze-thaw cycles before proteomic analysis. An aptamer-based proteomics platform (version 4; Somalogic) was used to measure relative levels of 4996 analytes, corresponding to 4730 distinct human proteins. Clinical Validation Study: Key ndings in the discovery proteomics study were validated in a publicly available, independent plasma proteomics dataset from the MGH Emergency Department COVID-19 Cohort (Filbin, Goldberg, Hacohen) analyzed using Olink, an antibody-based proteomics platform 46 47 . A multivariant instrument was constructed to genetically estimate protein levels of ADAMTS13 using locus-wide signi cant pQTLs acquired from PhenoScanner (www.phenoscanner.com).
All locus wide intronic signi cant pQTLs (p < 3.88x10 − 4 ) were ranked by effect size and pruned for independence (R 2 < 0.1). Rare variants were excluded (MAF > 0.01). For each direction of in uence, we combined the MR effect estimates from each instrument SNP using inverse variance weighted (IVW) meta-analysis to obtain a meta-analytic estimate. Sensitivity analyses were performed through the standard and additional evaluation of a weighted median and an MR-Egger estimate. Given an experimentally derived hypothesis that COVID-19 results in cardiac injury without infarction, we sought the most proximate ICD10 code for acute myocardial injury (ICD: I24.8, Other forms of acute ischemic heart disease). Summary statistics were acquired from UKB-b:11309 (n = 463,010 with the dataset containing 269 cases and 462,741 controls). Outcome summary statistics were accessed 9/20/2020 in MR-Base. See Supplementary Fig. 1 for a detailed ow diagram of the MR procedure used.

Golden Syrian Hamster COVID-19 Model
Cardiac and pulmonary tissue from six golden Syrian hamsters (Envigo), ages 10-12 week old, from our prior work 22 were used to further investigate candidate targets from the human discovery study. All animals were housed at Bioqual and studies were approved by their Institutional Animal Care and Use Committee. Animals were divided into two groups, naïve (uninfected) controls and SARS-CoV2 infected (n = 3/group). SARS-CoV2 infection was performed using previously published methods 22 . Brie y, animals in the infected group were each challenged with 5.0x10 5 TCID 50 (6x10 8 vp, 5.5 × 10 4 plaqueforming units (PFU)) SARS-CoV2, derived from USA-WA1/2020 (NR-52281, BEI Resources). Each animal in the infected group received 100 ml virus via intranasal administration. On day 4 post-infection, all animals were euthanized. Heart and lung samples were either xed in 4% paraformaldehyde for histologic staining or homogenized in QIAzol (Qiagen) and stored at -80C for RNA sequencing. Histological tissue processing, Carstair's method, and hematoxylin and eosin staining protocols were performed using previously published methods 22,48 . RNA was extracted from frozen lung tissue using the miRNeasy kit (Qiagen) with on-column DNase digestion, and quality was assessed using an Agilent Bioanalyzer. Library preparation was done using the SMARTer Stranded Total RNA-Seq V2 Pico Input Mammalian kit (Taara Bio). Libraries were validated with an Agilent 4200 TapeStation, and sequenced on an Illumina NovaSeq6000 at 100SR, targeting 25-30 million reads per sample.

FSTL3 Induction in Isolated Cardiomyocytes
Neonatal rat cardiomyocyte isolation, culture, and gene expression studies were performed using previously published methods 25 . Brie y, isolated cardiomyocytes were cultured in serum-free medium and incubated for 18 hours with 10ng/ml of various recombinant ligand proteins (Peprotech or R&D Systems) known to bind TGFβ receptors or Activin type II receptors (ActRII). RNA was extracted using Trizol, and polymerase chain reaction was done using SYBR green and standard ampli cation protocols. Relative changes in gene expression were measured using the ΔΔCT method.

Bioinformatic and Statistical Analyses
For analysis of the SOMAscan proteomics discovery study, median normalized relative abundances of the 4996 analytes were imported into R (v3.6.1) using the SomaDataIO package. Principal component analysis (PCA) was performed on log-transformed values using R's prcomp function, differential protein abundance using limma (version 3.40.6), and protein enrichment using clusterPro ler (version 3.12.0). Pathway analysis was done using Gene Ontology data (MSigDB), appended with two curated senescence associated secretory phenotype (SASP) pathways. For gene-set enrichment analyses (GSEA), proteins were ranked by t-statistic and 50,000 permutations used to compute the nominal enrichment p-values.
For all analyses, p-values were adjusted for multiple comparisons using the Benjamini & Hochberg method.
For validation using the data released by Filbin and colleagues 46 , we obtained Olink Normalized Protein eXpression (NPX) data and associated patient metadata from www.olink.com/mgh-covid-study. Of the 383 individuals included in this study, we removed patients without documented COVID-19 positivity, leaving 305 patients. In assessing the relationships of ADAMTS13 and FSTL3 with days post diagnosis, WHO severity, and 28-day survival we applied linear regression models and provide the overall model t. For the regressions on WHO severity and 28-day survival we used only protein abundances in NPX units from D0 patient samples.
Between-group comparisons in patient demographics and outcomes were examined using analysis of variance (ANOVA), followed by pairwise comparisons using chi-squared, t-tests, or Fisher's exact tests as appropriate between those with and without COVID-19, as well as those with and without cardiac involvement within each disease severity stratum. Pearson correlations were used to measure correlation of candidate proteins with cardiac biomarkers.
Analysis of the golden Syrian hamster lung RNA sequence data was done using STAR version 2.7.3a with the MesAur1.0 (GCF_000349665.1) assembly and annotation of the hamster downloaded from NCBI.
Transcript abundance estimates were calculated internal to the STAR aligner using the algorithm of htseq-count. DESeq2 was used for normalization, producing both a raw and normalized read count table. Differential expression at the gene level were performed by DESeq2 implemented in the DESeq2 R package. An adjusted p-value < 0.05 was used to determine genes that were signi cantly upregulated or downregulated by SARS-CoV2 at day 4 post challenge compared to naïve controls using the Benajmini & Hochberg method. Declarations DATA AVAILABILITY: All data that support the ndings of this study are included in the published article and its supplementary information les. The datasets during the current study are available from the corresponding author on reasonable request. Values are mean (SD) or n (%). P ANOVA tested differences between groups. * indicates P<0.05 on Fisher exact pairwise testing between COVID-19 patients and COVID-19 negative controls -groupings or between CV (+/-) or CV (low/high) within moderate and severe COVID-19 groups, respectively. CV = cardiac involvement. Figure 1 Senescence associated secretory phenotype is associated with COVID-19 severity and cardiac involvement. a) Principal component analysis of the discovery proteomics study with boxplots displaying unsupervised clustering of the 80 patient plasma samples based on PC1 proteins. b) Venn diagram displaying overlapping pathways associated with increasing COVID-19 severity in the discovery and validation cohorts. In the discovery cohort, enrichment analysis was done with PC1 proteins (Supplementary Table 2). In the validation cohort, pathway analysis was performed using Day 0 proteomic pro les regressed on maximum WHO disease severity per patient (Supplementary Table 8). c) Heat maps displaying differential expression of the 50 most signi cantly regulated genes in the SASP-1 or SASP-2 gene sets in lungs from hamsters infected with SARS-CoV2 versus naïve controls. Tissue samples were collected four days after infection. n=3/group. SASP = senescence associated secretory phenotype.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.