Dysregulated primary hemostasis in critically ill COVID-19 patients

Background: Microvascular thrombosis, as well as arterial and venous thrombotic events, have been largely described during severe Coronavirus disease 19 (COVID-19). Therapeutic anticoagulation has been proposed in critical patients, however mechanisms underlying hemostasis dysregulation remain unclear. Methods: We explored two independent cross-sectional cohorts to identify soluble markers and gene-expression signatures that discriminated COVID-19 severity and outcomes. Results: We found that elevated soluble (s) P-selectin at admission was associated with disease severity. Elevated sP-selectin was predictive of intubation and death (ROC AUC = 0.67, p = 0.028 and AUC = 0.74, p = 0.0047, respectively). An optimal cutoff value of 150 NC (normalized concentration) was predictive of intubation with 66% negative predictive value (NPV) and 61% positive predictive value (PPV), and of death with 90% NPV and 55% PPV. Next, an unbiased gene set enrichment analysis revealed that critically ill patients had increased expression of genes related to primary hemostasis. Hierarchical clustering identied ITG2AB, GP1BB, PPBP and SELPLG to be upregulated in a grade-dependent manner. ROC curve analysis for the prediction of mechanical ventilation was signicant for SELPLG and PPBP (AUC = 0.8, p = 0.046 for both markers). An optimal cutoff value for PBPP was predictive of mechanical ventilation with 100% NPV and 45% PPV, and for SELPLG was predictive of mechanical ventilation with 100% NPV and 50% PPV. Conclusion: We provide evidence that platelets contribute to disease severity with the identication of sP-selectin as a biomarker for poor outcome. Transcriptional analysis identied PPBP and SELPLG RNA count as biomarkers for mechanical ventilation. These ndings provide rationale for novel therapeutic approaches with antiplatelet agents.


Background
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is the causative agent of the coronavirus disease 2019 (COVID- 19) pandemic. Severe disease is characterized by an acute respiratory distress syndrome (ARDS), respiratory failure and ultimately death in about 1% of cases (1,2). Most important risk factors for severe disease include age, overweight, diabetes, hypertension and history of cardiovascular disease (3,4). Severe and critical patients were shown to develop arterial and venous thrombotic complications such as pulmonary embolism, stroke and myocardial infarction (5)(6)(7)(8). Markers of coagulation activation, in particular increased D-dimer, brinogen and von Willebrand factor (vWF) levels were found to be associated with critical illness, whereas only minor changes were noted in prothrombin time and platelet counts (9,10). Additionally, autopsy series described multiple thrombosis in deceased COVID-19 patients (11,12). These ndings suggest vascular microthrombotic disease as a primary factor for mortality in critically ill COVID-19 (6,13). Therefore, some authors supported the systematic use of curative anticoagulation upon admission to intensive care unit (ICU) (14), a strategy that has been reported to decrease mortality in severe COVID-19 (9).
Mechanisms underlying increased thrombotic events are still unclear but accumulating evidence point to a key role for endothelial and platelet activation (15,16). As viral inclusions were described in endothelial cells, it has been hypothesized that endothelial cell injury and activation could drive platelet activation and subsequent coagulopathy (12,17). Therefore, identifying the contribution of platelets to critical illness during SARS-CoV-2 infection is key to identify biomarkers predicting clinical outcome and discuss novel therapeutic strategies.
Platelet P-selectin is a key thromboin ammatory molecule involved in platelet activation and function. It has been demonstrated to play a crucial role in primary hemostasis by regulating platelet-leukocyte interactions, brin and tissue factor recruitment into platelet aggregates and thrombus formation (18). Its soluble form, sP-selectin, is released upon platelet and/or endothelial cell activation and measurement of sP-selectin has been proposed as a reliable marker of in vivo platelet activation (19).
The aim of our study was to assess, in hospitalized COVID-19 patients, the ability of sP-selectin to predict requirement for mechanical ventilation and death. Next, using whole-blood transcriptional data, we uncovered a platelet activation transcriptional signature associated with critical illness.

Patient characteristics
Two independent cohorts were analyzed for this study. Cohort 1 consisted of 60 COVID-19 patients that were admitted to European Georges Pompidou Hospital and included upon admission. Clinical and biological characteristics are previously reported in (20) and described in Table 1. Brie y, median age was 58.5 years (IQR, 49 to 72) and 76.7% were male. Patients were analyzed after a median duration of 6 days (IQR, 4 to 7) after onset of rst symptoms. Fever was present in 97% of the patients, and other most common symptoms were cough (83.3%), dyspnea (71.7%), fatigue (70.0%), myalgia (40.0%) and diarrhea (26.7%). Degree of COVID-19 severity was categorized as mild-to-moderate in 35 patients (clinical symptoms associated with dyspnea and requiring a maximum of 3L/min), severe in 10 patients (respiratory distress requiring more than 3 L/min of oxygen and no other organ failure) and critical in 15 patients (respiratory failure requiring mechanical ventilation).
Eleven out of 35 patients with mild-to-moderate disease and 3/10 patients with severe disease experienced clinical worsening and required mechanical ventilation. Thrombotic events were noted in 8 patients (3 mild-to-moderate, 2 severe and 3 critical). Cohort 2 consisted of 50 patients with various degree of COVID-19 admitted to Cochin hospital and 18 healthy controls, and was described in (21). For the purpose of the present study, only patients with whole-blood transcriptional analysis were included, i.e. 32 COVID-19 patients and 13 healthy controls.
Characteristics of the patients are described in Table 2. Median age was 56 years (IQR, 51 to 65) and 78% were male, while median age of healthy controls was 59 years (IQR, 41 to 60) and 77% were male. Patients were analyzed after a median duration of 10 days (IQR, 9 to 11) after onset of rst symptoms.
Fever was present in 100%, and other most common symptoms were dyspnea (100%), fatigue (32%), cough (97%), myalgia (97%) and diarrhea (34%). Degree of COVID-19 severity was categorized as mild-tomoderate in 11 patients (median oxygen requirement 1.5 L/min), severe in 10 patients (median oxygen requirement 5 L/min) and critical in 11 patients. No patients with mild-to-moderate disease required admission to an ICU, while 5 out of 10 patients with severe disease were eventually admitted to the ICU.
Thrombotic events were noted in 3 patients (1 severe and 2 critical).
Increased plasma sP-selectin on admission to the hospital is associated with disease severity and is a predictor of mechanical ventilation and death Because a high proportion of critically ill patients present clinical and biological evidence of activated coagulation, we measured plasma sP-selectin in Cohort 1. Increased plasma sP-selectin levels on admission to the hospital was signi cantly associated with critical disease (Figure 1a). Moreover, sPselectin levels were positively correlated with in ammatory parameters, i.e. CRP levels (r = 0.30, p = 0.017, Figure 1b), but not with platelet count (r = 0.1, p = 0.43, Figure 1c), We next evaluated the ability of sP-selectin, normalized to platelet counts, to predict intubation ( Figure  1d) or death ( Figure 1e) using a receiver operating characteristic (ROC) curve. We found that outcomes were signi cantly associated with higher sP-selectin values and signi cant area under the curve (AUC = 0.67, p = 0.028 for intubation and AUC = 0.74, p = 0.0047 for death). An optimal cutoff value of 150 NC (normalized concentration: concentration in pg/mL normalized to 10 6 platelets/mL) was predictive of intubation with 66% sensitivity, 61% speci city, 66% negative predictive value (NPV) and 61% positive predictive value (PPV), and of death with 82% sensitivity, 60% speci city, 90% NPV and 55% PPV.
Together, these data suggested that sP-selectin, a reliable marker of platelet activation, was strongly associated with death and may identify patients at higher risk of admission to the ICU.

Critically ill COVID-19 patients display a dysregulated primary hemostasis transcriptional signature
We previously performed a whole-blood immunological transcriptional characterization on 32 patients with laboratory-con rmed COVID-19 displaying various disease severity (cohort 2) (21). We therefore next analyzed genes involved in hemostasis using an unbiased gene set enrichment analysis comparing severe to critical COVID-19 patients. We found that critically ill patients displayed an elevated expression of genes related to primary hemostasis (Figure 2a). Hierarchical clustering identi ed genes that were upregulated in a grade-dependent manner, thereby driving the hemostasis signature in critically ill patients (Figure 2b). These genes included ITG2AB, GP1BB, PPBP and SELPLG (Figure 3a), which recapitulated multiple steps of platelet activation (22,23). Importantly, and similarly to sP-selectin, RNA levels more strongly correlated with CRP level (Figure 3b) than with platelet counts (Figure 3c). We next evaluated the ability of these genes to predict clinical outcome after normalization with platelets.
Among patients that did not initially require critical care upon inclusion (n = 21), 5 (24%) presented respiratory failure afterwards requiring mechanical ventilation (Table 2). PPBP and SELPLG were found to be the best predictor of clinical worsening, with an AUC of 0.8 for both outcomes (p = 0.048) (Figure 4a, b and c). An optimal cutoff value for PBPP of 5 NRC (normalized RNA counts : counts normalized to 10 6 platelets/mL) was predictive of intubation with 100% sensitivity, 63% speci city, 100% NPV and 45% PPV, while a cutoff value of 2.65 NRC for SELPLG was predictive of intubation with 100% sensitivity, 69% speci city, 100% NPV and 50% PPV. Overall, these markers provide excellent negative predictive values, with no patient below the threshold value requiring future mechanical ventilation.

Discussion
Dysregulated hemostasis is emerging as a key factor in COVID-19 pathogenesis and severity. The present study provides new insights into the contribution of platelets to disease severity with the identi cation of a unique hemostasis signature in critically ill patients. We identi ed sP-selectin as a soluble marker associated with death, and elevated SELPLG and PPBP RNA levels as strong negative predictors of mechanical ventilation in hospitalized patients.
Recently, altered platelet gene expression and elevated sP-selectin were reported in COVID-19 patients (24,25). Our study supports these ndings and highlights the key role of primary hemostasis geneexpression signature in critical patients, especially the speci c platelet chemokine-encoding gene PPBP and the gene encoding for the P-selectin glycoprotein ligand-1 (PSGL-1), SELPLG. PPBP and SELPLG gene expression could also predict clinical outcome. Pro-platelet basic protein (PPBP), also known as CXCL7, is the most abundant platelet chemokine and (22,26) expressed within platelets as an inactive precursor and activated after cleavage during thrombus formation by enzymes released by neutrophils, while PSGL-1 is expressed by leukocytes, allowing the formation of platelet-leukocyte conjugates and adhesion of leukocytes to activated endothelium through its interaction with P-or E-selectin (22,27). PBPP has been shown to be essential for neutrophil migration into the thrombus (28) and the formation of plateletneutrophils aggregates (18), since its receptor, CXC chemokine receptor 2 (CXCR2), is expressed on the surface of neutrophil. Supporting thus, we previously observed increased CXCR2 expression in severe COVID-19 patients (21). Murine models of acute lung injury showed that deletion of PPBP protected the mice from lung disease (26), as well as blocking platelet activation (29,30). Overall, our data support a model whereby platelet activation induces PPBP release followed by neutrophil attraction into sites of injury, precipitating organ injury and failure. This hypothesis is also supported by increased sP-selectin in most severe patients, since P-selectin on the surface of platelets promotes the formation of plateletneutrophil or monocyte aggregates (27), via binding to its ligand PSGL-1 (encoded by SELPLG) expressed on the surface of leukocytes (31). Accordingly, increase in platelet-monocyte aggregates has been recently described in severe COVID-19 (32), a mechanism that was effectively blocked by platelet Pselectin neutralization.
The identi ed hemostasis gene-signature also included ITG2AB and GP1BB. ITG2AB encodes integrin alpha-IIb subunit, a member of the receptor complex integrin alpha-IIb/beta-3 (αIIbβ3), which is the receptor for brinogen, plasminogen, prothrombin and thrombospondin (23). Importantly, binding of integrin αIIbβ3 to brinogen forms bounds between adjacent platelets, an essential step for clot formation. Finally, GP1BB encodes glycoprotein Ib, the receptor for vWF, and mediates platelet adhesion in the arterial circulation (23,33). Altogether, these upregulated genes recapitulate multiple steps of platelet activation (22,23).
In addition to their role in primary hemostasis, platelets are an integral part of the immune response to pathogens. They possess a wide array of microbial sensors and secrete in ammatory mediators upon activation, hence amplifying innate immune responses. The functional interdependence and the coordinated activation of both processes, designated as thrombo-in ammation, may drive adverse effects such as thrombosis, multiple organ failure and death (34). We and others have shown speci c in ammatory signatures that characterized severe and critical from mild-to-moderate COVID-19 patients (e.g. type I interferon pathways, NF-κB pathways). Intriguingly, we found that the hemostasis gene signature was able to further distinguish critical from severe COVID-19 patients, underlying the hypothesis that deregulated hemostasis, in a continuum with in ammation, could drive COVID-19 progression into critical disease. Viral infection and sepsis models have been associated with platelet gene expression and functional alterations (35,36). However, some unique features may characterize SARS-CoV-2-mediated platelet dysregulation. Manne et al. demonstrated that platelet gene-expression pro le was unique in COVID-19 patients in comparison to H1N1 pandemic patients (25). Moreover, thrombocytopenia, another feature of sepsis during viral infection (37,38), was not observed in COVID-19 patients. Conversely, we and others found that platelets counts tended to increase with disease severity, supporting the unique aspects of platelet dysregulation in COVID-19 critical patients.
Identi cation of such biomarkers of clinical outcome can have strong impact on clinical practice and help provide novel therapeutic rationale. Prediction of illness trajectories after the onset of symptoms is di cult but remains critical to more accurately identify patients that will require ICU. Our ndings also suggest that a subset of patient may bene t from drugs preventing platelet activation such as antiplatelet agents (e.g. NCT04365309 ; NCT04363840) or P-selectin inhibitors such as crizanlizumab (e.g. NCT04435184) (39). This strategy, in combination with anti-in ammatory drugs (glucocorticoids, anti-interleukin-6) could have substantial impact on thrombo-in ammation, which seems to drive COVID-19 severity.
Our study has some limitations. We used two independent cohorts with a small sample size performed each in a single center. The two cohorts were not homogenous in regards to timing of measurements and severity of patients. Also, the study was not designed as a longitudinal study, so no sequential measurement was available. Serial measurement of identi ed biomarkers over time will be important to more precisely de ne the optimal time window for their measurement, and hence improve their prediction performance. Moreover, our transcriptomic analysis was performed on whole-blood RNA, so we cannot evaluate the contribution of each cell population to hemostasis dysregulation. Separate transcriptional pro ling of platelets versus other circulating populations may provide further insights into the contribution of each population to thrombo-in ammation and dissect the mechanisms of plateletneutrophil or platelet-monocyte aggregates.
This exploratory study sheds light on dysregulated hemostasis and the role of thrombo-in ammation in critical patients. We identi ed platelet activation marker sP-selectin, SELPLG and PPBP as potential biomarkers of critical worsening and provide rationale for evaluating sP-selectin blockade or anti-platelet drugs in most severe COVID-19 patients. Additional studies designed to test the performance of these biomarkers are required to both validate our ndings and optimize their ability to predict progression to critical disease.

Cohorts
Two independent cohorts were analyzed for this study. Data used for the analysis of soluble P-selectin's ability to predict admission to ICU were extracted from a non-interventional study that was conducted at European Georges Pompidou Hospital (Paris, France) and partially described in (20) (Cohort 1). Brie y, Cohort 1 included consecutive patients with SARS-CoV-2 infection. Inclusion criteria were patients over 18 years of age, with a proven SARS-CoV-2 infection, which presented to the emergency department with hospitalization criteria. Patients were then hospitalized into conventional wards or directly to the ICU. For all patients, baseline characteristics (demographic, treatment, clinical, cardiovascular risk factors and body mass index) and biological data were retrieved from the medical records using a standardized data collection.
Cohort 2 (21) was conducted between March 19, 2020 and April 3, 2020 in Cochin Hospital (Paris, France), in the setting of the local RADIPEM biological samples collection derived from samples collected in routine care. Inclusion criteria for COVID-19 inpatients were: age between 18 and 80 years old, diagnosis of COVID-19 according to WHO interim guidance, and positive SARS-CoV-2 RT-PCR testing on a respiratory sample (nasopharyngeal swab or invasive respiratory sample). Detailed clinical and immunological characterization of the cohort was previously described in (21). Epidemiological, demographic, clinical, laboratory, treatment, and outcome data were extracted from electronic medical records using a standardized data collection form.
The severity of COVID-19 was classi ed at the time of inclusion based on the adaptation of the Sixth Revised Trial Version of the Novel Coronavirus Pneumonia Diagnosis and Treatment Guidance and described in (21) . Soluble P-selectin measurement Soluble P-selectin quanti cation was performed on Cohort 1. Plasmas were collected on 0.129 M trisodium citrate tubes (9NC BD Vacutainer, Plymouth, UK). Plasma poor platelet was obtained after centrifugation twice at 2500 g for 15 minutes. PPP was frozen after a second centrifugation at 2500 g for 15 minutes and stored at -80°C until analysis of vascular markers. Soluble P-selectin were quanti ed in PPP with a Human Magnetic Luminex Assay from R&D systems (Lille, France). Data were assessed with the Bio-Plex 200 using the Bio-Plex Manager 5.0 software (Bio-Rad, Marnes-la-Coquette, France). Normalized concentration (NC) used to calculate ROC AUC p-values represents sP-selectin concentration in pg/ml normalized to platelet numbers (10 6 /mL).

Gene expression analysis
Detailed methods was previously reported in Hadjadj et al. (21). Brie y, we analyzed 100 ng (5 μl) of total RNA from each sample using the Nanostring Human Immunology kit v2 according to manufacturer's instructions. Raw RNA counts were adjusted using ve housekeeping genes selected from the 15 candidate control genes provided by Nanostring, following the geNorm method. For gene set enrichment analysis (GSEA), genes were ordered by t-statistic from unpaired t-test comparing normalized RNA levels of severe vs critical patients, and then fed to the gene set enrichment algorithm (GSEA version 4.0.3, Broad Institute), along with a pathway data set built from the Nanostring Immunology panel version 2 annotation le. Parameters were set as follows: method, pre ranked gene list; number of permutations, 2,000; enrichment statistic, classic; min set size, 5; max set size, 200; and all other parameters as default. Hierarchical clustering of genes belonging to the hemostasis gene set was performed using hclust with default distance matrix. Heatmap displaying genes that are upregulated in a grade-dependent manner was obtained using pheatmap (package pheatmap), with data centered to 0 and scaled to unit variance for each gene. Normalized RNA count (NRC) used to calculate ROC AUC p-values represents adjusted RNA count (to house keeping genes) normalized to platelet numbers (10 9 /L).

Statistical analyses
Quantitative variables were compared among groups using Kurskal-Wallis test followed by Dunn's post test, while quantitative variables were compared using the χ 2 test of independence. Correlations coe cients and p-values were assessed using Spearman's method. ROC AUC p-values were determined using Hanley's method. Optimal thresholds were determined by maximization of Youden's index. All analyses were two-sided and a p-value smaller than 0.05 was considered statistically signi cant. Statistical analyses were performed using R v. 3.4.3 (CRAN).