Unsupervised clustering of SARS-CoV-2 positive hospitalized patients identifies six endophenotypes of COVID-19 and points to FGFR and SHC4-signaling in acute respiratory distress syndrome

William Ma, Antoine Soulé, Katelyn Liu, Catherine Allard, Karine Tremblay, Simon Rousseau, Amin Emad
doi: https://doi.org/10.1101/2022.11.02.22281834

Abstract

Defining the molecular mechanisms of novel emerging diseases like COVID-19 is crucial to identify treatable traits to improve patient care. To circumvent a priori bias and the lack of in-depth knowledge of a new disease, we opted for an unsupervised approach, using the detailed circulating proteome, as measured by 4985 aptamers (SOMAmers), of 731 SARS-CoV-2 PCR-positive hospitalized participants to Biobanque québécoise de la COVID-19 (BQC19). The consensus clustering identified six endophenotypes (EPs) present in this cohort, with varying degrees of disease severity. One endophenotype, EP6, was associated with a greater proportion of ICU admission, mechanical ventilation, acute respiratory distress syndrome (ARDS) and death. Clinical features of this endophenotype, showed increased levels of C-reactive protein, D-dimers, elevated neutrophils, and depleted lymphocytes. Moreover, metabolomic analysis supported a role for immunothrombosis in severe COVID-19 ARDS. Furthermore, the approach enabled the identification of Fibroblast Growth Factor Receptor (FGFR) and SH2-containing transforming protein 4 (SHC4) signaling as features of the molecular pathways associated with severe COVID-19. Finally, this information was sufficient to train an accurate predictive model solely based on clinical laboratory measurements, suggesting the use of blood markers as surrogates for generalizing these EPs to new patients and automating identification of high-risk groups in the clinic.

Introduction

The coronavirus disease 2019 (COVID-19) is a new human disease caused by the coronavirus SARS-CoV-2 infection that has been assessed pandemic by the World Health Organization in March 2020. As SARS-CoV-2 infection spread, a breath of outcomes of the infection became apparent, from asymptomatic individuals to severely ill and dying, from complete recovery to long-lasting symptoms1,2. The emergence of novel diseases such as COVID-19 presents the medical and scientific community with numerous challenges. Among them, defining the molecular mechanisms of disease related to specific outcomes is important to identify treatable traits and improve the performance of healthcare systems facing the challenges brought by the pandemic.

Successfully reaching this precision medicine goal requires a more granular definition of the pathology. A symptom-based method to discover molecular mechanisms of the disease may result in a challenge emerging from the fact that the same higher-level phenomenon, such as COVID-19 severity, can be produced by several different molecular mechanisms, a phenomenon termed the « many-one» limitation3. Recent advances in computing strategies, such as machine learning, has enabled the development of methods that help overcome this limitation by starting from molecular profiles instead of symptoms to define endophenotypes, i.e. subgroups of individuals who are inapparent to traditional methods but share a common set of molecular factors that can lead to treatable traits4. Establishing successful treatment strategies requires a tailored approach to the underlying molecular mechanisms that can help predict and alter disease trajectories5. Endophenotypes can become apparent using extensive molecular phenotyping combined with machine learning algorithms6-8. Current investigations of endophenotypes in COVID-19 have mainly relied on supervised approaches using fixed outcomes (such as disease severity) and integrating clinical variables at the onset9. We hypothesize that using an unsupervised approach exploiting a rich molecular dataset can provide novel mechanistic insights into the pathobiology of severe COVID-19 that can help physicians improve diagnosis and prognosis.

Therefore, this study aims to identify endophenotypes linked to diverse clinical trajectories of COVID-19 using the extensive molecular phenotyping of a cohort of 731 SARS-CoV-2 positive hospitalized patients from the Biobanque québécoise de la COVID-19 (BQC19, www.quebeccovidbiobank.ca)10, a prospective observational cohort of SARS-CoV-2 positive and negative participants recruited in the province of Québec, Canada, to improve our understanding of COVID-19 pathobiology and our capacity to alter disease outcomes.

In this manuscript, we report the identification of six endophenotypes in hospitalized SARS-CoV-2 positive participants to BQC19, associated with different clinical trajectories. The molecular information underpinning these endophenotypes were used to increase our understanding of pathobiology and predict the likelihood of patients admitted to the hospital to belong to each endophenotype using clinical blood workups.

Results

Unsupervised clustering of SARS-CoV-2-positive hospitalized BQC19 participants reveal endophenotypes associated with varying disease severity

In this study, we aimed to identify endophenotypes of COVID-19 based on the circulating proteome of patients in our cohort of SARS-CoV-2 positive hospitalized participants to BQC19 (n = 1,362, Table 1), using an unsupervised approach. Figure S1 shows the distribution of the time of hospital admission of the patients and the corresponding waves as defined by National Institute of Public Health of Quebec (https://www.inspq.qc.ca/covid-19). For this purpose, we performed consensus agglomerative clustering of the subset of patients (n = 731, Table S1) for whom data corresponding to circulating proteome measured by a multiplex SOMAmer affinity array (Somalogic, ∼5,000 aptamers)11 was available in BQC19. The remaining samples were kept aside for follow-up analysis. First, the optimal number of clusters (k = 6) was identified using two criteria, Akaike’s Information Criteria (AIC) and Bayesian Information Criteria (BIC) (Figure 1A); then, consensus agglomerative clustering (Euclidean distance and Ward linkage)12,13 using bootstrap subsampling was performed to obtain six robust clusters (see Methods for details) (Figure 1, Figures S2 and S3).

View this table:
Table 1:

Clinical and pathological characteristics of the BQC19’s participants used in this study.

Figure 1:
Figure 1:

Unsupervised consensus clustering of SARS-CoV-2 positive patients.

A) The elbow points (circles in red) of Akaike’s Information Criteria (AIC) and Bayesian Information Criteria (BIC) curves versus number of clusters consistently corresponded to k=6 as the optimal number of clusters. B) The heatmap shows the expression of aptamers (rows) in each sample (columns). The dendogram shows the identified endophenotypes. C) Characterization of samples based on sex at birth, highest world health organization (WHO) severity level achieved, intensive care unit (ICU) admission, ventilatory support, and oxygen therapy. For the last three rows, a sample colored “black” reflects a label of “yes”.

The clinical and pathological characteristics of patients in each endophenotype is provided in Table S1. To characterize the identified endophenotypes (EPs) with respect to disease severity, we performed two-sided Fisher’s exact test to assess their enrichment (or depletion) in “severe” or “dead” outcome. EP6 was significantly enriched in the severe/dead outcomes (Benjamini-Hochberg false discovery rate (FDR) = 1.72E-18) with these outcomes observed in 66.1% of EP6 patients. Meanwhile, EP1 was significantly depleted in severe/dead outcomes (FDR = 8.23E-11) (Figure 2A, Table S1) with these outcomes only observed in 11.6% of EP1 patients. In addition, EP6 was enriched in participants 1) receiving oxygen therapy (FDR = 6.24E-12), 2) receiving ventilatory support (FDR = 6.63E-12), and 3) being admitted to intensive care unit (ICU) (FDR= 1.26E-22) (Figure 2A, Table S1). Kaplan Meier analysis14 also confirmed that the identified EPs have a distinct temporal pattern of admission to ICU (multivariate logrank test P = 5.06E-36), with EP1 (EP6) having the highest (lowest) chance of not being admitted to ICU (or die prior to that) in a 40-day span since their admission to the hospital (Figure 2B). A similar pattern can be observed when patients that died before admission to ICU were excluded (Figure S4, multivariate logrank test P = 6.55E-36). A two-sided Mann-Whitney U (MWU) test showed that patients in EP5 were generally older than other EPs (FDR = 6.94E-5), while EP3 included younger patients (FDR = 1.82E-4). However, EP6 (which had the most severe patients) did not show enrichment in older patients or individuals with high BMI (two-sided MWU FDR>0.05) (Figure 2C, Table S1).

Figure 2:
Figure 2:

Characterization of endophenotypes (EPs).

A) Enrichment or depletion of each EP in clinical variables (one cluster versus rest). Two-sided Fisher’s exact tests are used to calculate the p-values, which are corrected for multiple tests using Benjamini-Hochberg false discovery rate (FDR). Gradients of blue show depletion, while gradients of red show enrichment. FDR values above 0.05 are depicted as white. B) The number of patients in each EP and the colors used to represent them in panels C, D, and E. C) Kaplan-Meier analysis of the time between patients’ admission to the hospital and their admission to intensive care unit (ICU) (or death if earlier) for each EP (Delta). D) Distribution of age in each EP. E) Distribution of BMI in each EP. F) COVID-19 severity in each EP.

These analyses revealed that the unsupervised approach was able to identify endophenotypes with distinct disease characteristics and outcomes using the circulating proteome of the patients. We identified EP6 as a group of participants with an increase in many measures of COVID-19 disease severity.

EP6 is enriched with BQC19’s participants having acute respiratory distress syndromes

In accordance with increase disease severity, EP6 was also enriched in COVID-19 medical complications (two-sided Fisher’s exact test): acute respiratory distress syndrome (ARDS) (FDR = 1.43E-10), acute kidney injury (FDR = 6.37E-7), bacterial pneumonia (FDR=2.13E-6), liver dysfunction (FDR = 9.32E-3), and hyperglycemia (FDR = 1.56E-2) (Figure 3, Table S2). The frequency of ARDS was just below 8% in EP1 compared to greater than 44% in EP6, making this complication a key feature of this cluster (Figure 3, Table S2).

Figure 3:
Figure 3:

Frequency and significance of complications in different EPs.

The value in each cell shows the percentage of patients of that EP (column) that suffered from the complication (row). The colors represent two-sided Fisher’s exact test false discovery rate (FDR, corrected for multiple tests). Red represents enrichment, while blue represents depletion. FDR values below 0.05 are shown as white.

EP6 is enriched in blood metabolites associated with severe COVID-19

To further characterize each EP and gain insight into mechanisms of disease, metabolomic profiling of plasma samples was done in parallel to the SOMAmer analysis. The results yielded data on 1,435 metabolites, of which 576 were found significantly altered in EP6 (two-sided MWU FDR < 0.01). Moreover, the metabolomic characterization of the plasma samples supported the distinction in blood composition at the levels of metabolomic sub-pathways and individual metabolites between the different EPs (Figure 4 and S3).

Figure 4:
Figure 4:

Metabolite characteristics of endophenotypes (EPs).

A) The heatmap shows the over-expression (red) and under-expression (blue) of metabolites in different EPs (two-sided Mann–Whitney U test). Row names show metabolites followed by the sub-pathway to which they belong in parentheses. Only top 15 metabolites (based on false discovery rate for EP6) for which a definite name and sub-pathway was available are shown. Full list is provided in Table S3. B) The heatmap shows the enrichment (one-sided Fisher’s exact test) of EPs in different metabolite sub-pathways. See Table S3 for the full list.

Pathway enrichment analysis identifies FGFR-signaling in severe COVID-19 acute respiratory distress syndrome

To gain insight into the molecular mechanisms underlying the severity in EP6, we performed pathway enrichment analysis using the Knowledge Engine for Genomics (KnowEnG)6 for the aptamers associated with EP6. Since aptamers were used to identify the EPs, it is expected that many of them would be significantly associated with the EPs. As a result, we selected a very strict threshold of FDR < 10E-20 (two-sided MWU test) to identify top aptamers associated with EP6 (Table S4). We then used the gene set characterization pipeline of KnowEnG with Reactome 15 pathway collection. This analysis showed that while EP6 is characterized by pathways associated with Interleukins and Cytokine Signaling in Immune system, multiple instances linked it with Fibroblast Growth Factor Receptor (FGFR) signaling, identifying this pathway as a potential driver of severe pathology that was not present in other EPs (Tables 2 and S4).

View this table:
Table 2:

Reactome pathways associated with EP6, based on expression of aptamers. KnowEnG analytical platform was used. The p-values were calculated using a one-sided Fisher’s exact test and were corrected for multiple tests using Benjamini-Hochberg method. Only signaling pathways with FDR<5E-4 are shown in this table (see Table S4 for the full list).

SHC4 genotype and protein expression levels are associated with higher odds of belonging to EP6

To further improve our understanding of the molecular mechanisms underlying EP6, we leveraged an additional dataset of Genome Wide association Study (GWAS) corresponding to these patients. We identified 25 single nucleotide variations (SNVs) distributed in 13 annotated genes, that were below a p-value threshold of 1E-4 differentiating EP6 versus the rest (Table 3). We then investigated each of the SNVs to which we could assign a gene and an aptamer, to assess whether their protein product in circulation was differentially regulated by the genotype (Table 4). We discovered two genes, SHC4 (encoding SHC adaptor protein 4) and CACNA2D3 (encoding calcium voltage-gated channel auxiliary subunit alpha2 delta3) for which there was a significant association between genotype and protein expression levels (p-values < 0.05). While CACNA2D3 may have mild impact on EP6 membership (odd ratio = 0.61, Table 4), SHC4 was one of the top-enriched aptamers (position 32 out of 4,985), with odds ratio of 11.98 and 2.00 for the protein product and SNV, respectively of belonging to EP6 for the alternative allele. Therefore, the GWAS analysis revealed that the signaling adaptor protein SHC4 may play an important mechanistic role in contributing to severe disease pathology.

View this table:
Table 3:

SNVs differentiating EP6 against all other endophenotype clusters.

View this table:
Table 4:

Association between genotypes and aptamer expression levels

A predictive model based on blood markers predicts EPs in a separate validation cohort

To further characterize each EP, we assessed the clinical laboratory results obtained from blood draws and compared them between the groups. We focused on 21 markers that were measured in at least 50% of the patients used for consensus clustering (Figure 5A and Table S5) and used the summary value reported in the BQC19 database corresponding to the most extreme measurement among multiple blood draws (Table S5 provides this information for each blood marker). Figure 5A shows the elevation and depletion of these markers in the identified EPs. EP6 is characterized by abnormal values in markers of inflammation (lymphopenia, total white blood cells count, neutrophilia, C-reactive protein (CRP)), liver damage (alanine aminotransferase (ALT), albumin, lactate deshydrogenase (LDH)), blood clotting disorder (D-dimers, low hemoglobin, International Normalized Ratio (INR) and hyperglycemia (glucose).

Figure 5:
Figure 5:

Differential expression patterns of blood markers in the identified endophenotypes (EPs) and characterization of predicted endophenotypes (PEPs) based on the predictive model.

A) Heatmaps of false discovery rates (FDR) values for two-sided one-vs-rest Mann–Whitney U tests for 21 blood markers for each EP. Abbreviations used: WBC = white blood cells, LDH = lactate deshydrogenase, ALT = alanine aminotransferase, APTT = activated partial thromboplastin time, INR = International Normalized Ratio. FDR values below 0.05 are shown as white. B) World health organization COVID-19 severity assessed in each PEP. C) Distribution of age in each PEP. D) Distribution of BMI in each PEP. E) Kaplan-Meier analysis of the time between patients’ admission to the hospital and their admission to ICU (or death if earlier) for each PEP (Delta). The colormap in panel E shows the number of patients in each PEP group and the colors used to represent them in panels C, D, and E. PEPs were predicted based on blood markers and were not used to originally identify the EPs.

To identify relationships that may shed light on factors influencing clinical laboratory results defining EP6, we also performed Spearman’s rank correlation analyses between each blood test values and metabolites (Table S5).

Since EPs and particularly EP6, which we identified as the EP with worst outcome, showed a clear and distinct clinical laboratory result signature compared to other EPs, we sought to develop a predictive model based on these signatures. Due to the large number of missing values for these markers in our cohort, we developed a nearest-centroid classifier that is capable of dealing with missing values and can predict EPs based on blood markers (see Methods for details). To test the ability of this model on prediction of EPs on an independent yet similar dataset, we used data corresponding to 631 SARS-CoV-2 positive hospitalized BQC19’s participants that did not have circulating proteome data and hence were not used to identify the endophenotypes. The clinical and pathological characteristics of patients in each predicted endophenotype (PEP) is provided in Figure 5B-5E and Table S6.

Our predictive model identified 116 of these patients to belong to EP6. Fisher’s exact test showed a significant enrichment of the predicted EP6 (PEP6) in severity/dead (FDR = 1.77E-22), while PEP1 and PEP2 were significantly depleted in these outcomes (FDR = 1.61E-4 and FDR = 1.30E-8, respectively), as shown in Figure 5B and Table S6. Similar to EP6, PEP6 was also significantly enriched in participants 1) receiving oxygen therapy (FDR = 2.38E-8), 2) receiving ventilatory support (FDR = 4.40E-8), and 3) being admitted to ICU (FDR = 1.23E-24) (Table S6). Kaplan Meier analysis also confirmed that these PEPs have a distinct temporal pattern of admission to ICU (multivariate logrank test P = 1.05E-34), with PEP6 having the lowest chance of not being admitted to ICU (or die prior to that) in a 40-day span since their admission to the hospital (Figure 5E).

These results suggest that our predictive model can use these 21 blood markers to generalize the definition of endophenotypes to patients for whom the proteome data is unavailable.

Discussion

The results presented herein came from a large cohort of deeply phenotyped SARS-CoV-2 positive hospitalized participants, combined with unsupervised clustering that provided both expected and novel findings into the molecular mechanisms regulating COVID-19 outcomes.They led to the identification of an endophenotype (EP6) associated with worst clinical outcome of COVID-19 (enriched in acute respiratory distress syndrome) reflected by a greater proportion of ICU admission, mechanical ventilation, and severe/death outcomes (Figure 1). Clinical features of this endophenotype were consistent with published literature with increased levels of CRP, D-dimers, elevated neutrophils, and depleted lymphocytes (Figure 5A, Table S5). Our approach enabled the identification of interleukins, FGFR and SHC4 signaling as cardinal features of the molecular pathways associated with severe COVID-19. Importantly, this information was sufficient to train an accurate predictive model that could in the future support clinical care.

The approach: unsupervised clustering capacity at identifying clinically meaningful subpopulations

Our unsupervised clustering approach in conjunction with a rich molecular dataset enabled us to identify endophenotypes that could not be captured using traditional methods classifying the population in two bins solely based on severity. This is in part because of the « many-one»limitation: the same higher-level phenomenon (COVID-19 severity) can be produced by several different molecular mechanisms. Determining endophenotypes using an unsupervised method provides a higher granularity and increases the chance to identify distinct molecular mechanisms and pathways resulting in similar COVID-19 severity. Accordingly, we identified two endophenotypes with more favorable outcomes (EP1 and EP2), three endophenotypes with intermediate outcomes in terms of severity (EP3, EP4 and EP5) and one endophenotype which led to worst outcomes compared to all others (EP6).

The identification of endophenotypes was done systematically using robust consensus clustering of aptamer expression levels in which the optimum number of clusters was determined congruently using two well-established measures: AIC and BIC. The consensus clustering using bootstrap sampling (1000 times) ensured identification of robust clusters that are not sensitive to exclusion of some of the samples (20% randomly selected and excluded at each cycle). Moreover, identifying the best number of clusters using AIC/BIC (both of which agreed with each other) allowed us to reveal the patterns of the EPs directly from the data, instead of imposing a pattern onto it through human supervision. This is an important strength of the study that enabled us to identify distinct molecular patterns of patients that could have remained undetected using other traditional approaches.

Moreover, to improve the translational applicability of EPs, we developed a predictive model based only on laboratory measured blood markers to generalize the definition of these endophenotypes to unseen samples without measured aptamer expression levels. Characteristics of EPs predicted solely based on their blood markers were consistent with the original EPs, suggesting the use of blood markers as surrogates for generalizing these EPs to new patients and automating identification of high-risk groups in the clinic.

COVID-19 molecular pathology

The datasets used in this study carry rich molecular information on mechanisms of disease. SARS-CoV-2 infection has been shown to be associated with altered kynurenin levels associated with increase IL-6 and kidney injury16. Interestingly, we also observed Kynurenin to be enriched in EP6 but depleted in EP1, supporting the association of increase kynurenin with COVID-19 severity and the enrichment in acute kidney injury complication.

Spermidine/spermine N(1)-acetyltransferase (SSAT) contributes to polyamine synthesis. In addition, its extracellular metabolite, N1,N12-diacetylspermine, is one of the top 15 metabolites enriched in EP6 (Figure 4A), is positively correlated with Urea and creatinine, and is negatively correlated with lymphocyte numbers (Table S5). Deletion of SSAT in mice is protective against LPS-induced kidney injury17. SSAT activity is associated with white blood cell count in Acute Myeloid Leukemia and Chronic Myeloid Leukemia patients18. This suggest that the tryptophan and polyamine metabolisms are associated with acute kidney injury in COVID-19 and identifies potential pathways of disease progression.

COVID-19 ARDS

EP6 is characterized by its enrichment in ARDS (44% vs 8% in EP1, Figure 3). Low levels of Sphingosine 1-phosphate (S1P) are associated with ARDS and were shown to be associated with greater ICU admission and decrease survival in COVID-1919. Accordingly, we found that S1P levels are depleted in EP6 while they are enriched in EP1 (Figure 4A). Interestingly, the aptamers detecting neutral ceramidase, an enzyme converting ceramides into sphingosine, is enriched in EP1 (although changes in protein abundance as detected by aptamers may not necessarily reflect changes in enzymatic activity). Accordingly, dihydroceramides and ceramides are depleted in cluster EP1. Conversely, dihydroceramides and ceramides are significantly enriched in EP6, suggesting that there is a shunting of the pathway away from sphingosine towards more pro-inflammatory ceramides in EP6 associated with metabolic disorders20. Moreover, one of the top aptamers found enriched in EP6, is the enzyme Serine palmitoyltransferase 2 (SPTLC2) (Table S4)21,22. These results suggest a counterbalance between ceramides and sphingosine, where the former is associated with poorer outcomes during critical illness, whereas higher levels of the latter has more favorable outcomes in ARDS.

Metabolomic profile of EP6 supports a role for immuno-thrombosis-mediated organ damage in COVID-19

To obtain a more comprehensive understanding of the alteration in metabolic profiles in EP6, we investigated which metabolic subpathways were significantly enriched in EP6 (Figure 4, Table S3). The two top pathways (FDR < 1E-2) are “Methionine, Cysteine, SAM and Taurine Metabolism” and “phosphatidinylethanolamine (PE)”. Interestingly, these two pathways are known to interact23, with PE methylation a major consumer of S-Adenosylmethionine (SAM) leading to the synthesis of S-Adenosylhomocysteine (SAH) and cystathionine24, which itself has been found upstream of 2-hydroxybutyrate and 2-aminobutyrate in a model of hepatoxicity25. SAH, cysthathionine, 2-hydroxybutyrate and 2-aminobutyrate are significantly enriched in EP6. Congruently, EP6 is enriched (FDR < 1E-2) in liver dysfunction (Figure 3, Table S2), with markers of liver dysfunction all significantly altered in clinical blood works: ALT, albumin, bilirubin and LDH (Figure 5A, Table S5).

PEs become exposed at the surface of cell membranes upon exposure to stress, inflammation, and cell death26,27. In a Syrian hamster model, infection with SARS-CoV-2 had markedly increased PE expression in the animals that were fed a high salt, high fat diet, demonstrating the interaction between infection and metabolic disorder with the abundance of circulating PE28. Phospholipids-containing microparticles from platelet activation contribute to Tissue Factor activation and pro-thrombinase activity29. Platelets-derived microparticles have a much greater procoagulant activity than activated platelets30. Exposure of glycerophospholipids in conjunction with phosphatidinylserine (PS) enhances factor X activation and increases pro-thrombinase activities31,32. Interestingly, red blood cells exposed to paclitaxel, PS were exposed to the surface by protein kinase C (PKC) zeta activation of scramblase33. The aptamer detecting PKC zeta is one of the top aptamers associated with EP6 (Table S4). PKC zeta can be activated by testosterone, Dehydroepiandrosterone(DHEA)34 and dexamethasone35, signals of relevance to EP6. The membranes of azurophilic granules, which contains Cathepsin G representing the most enriched aptamer in EP6, are enriched in PE relative to PC36. Accordingly, EP6 in addition to significantly enhanced PE abundance, showed decrease platelets count, increased D-Dimers, INR and activated partial thromboplastin time (APTT) (Figure 5A, Table S5), hallmarks of Disseminated Intravascular Coagulation (DIC), a serious and often lethal complication of sepsis37. Liver lesions are frequently observed in DIC38, where liver damage can cause DIC, or exacerbate its manifestation due to its function in clearing activated products of the coagulation cascade. Taken together, the metabolomic profile of EP6 supports pro-coagulation activity in circulation that can be linked to organ damage.

Many early reports suggested a role for immunothrombosis involving neutrophil-mediated release of NETs contributing to endothelial dysfunction as a mechanism of microthrombosis in COVID-19 associated ARDS39-41. These findings have been supported by several studies carried out in humans42-50. All of these studies were performed with 7 to 77 participants. Our study supports these findings in two important ways: 1) we used a much greater sample size (n = 731) and, 2) the identification of molecular factors associated with immuno-thrombosis emerged from an unsupervised analysis of deep phenotyping of the participant population. The strength of the extensive characterization performed in this study has enabled the finer definition of molecular mechanisms of disease by providing associations between the circulating proteome, metabolome and clinical laboratories results.

FGFR and SHC4 intracellular signaling in COVID-19 ARDS

Two of the outstanding novel molecular factors identified by our study associated with COVID-19 ARDS are FGFR and SHC4. Circulating levels of the pro-angiogenic FGF-2 has been associated with COVID-19 severity and creatine levels in a study of 208 SARS-CoV-2 positive participants51. It is noteworthy that the use of Nintedanib, an inhibitor of FGFR, Vascular Endothelial Growth Factor Receptor (VEGFR) and Platelet-Derived Growth Factor Receptor (PDGF-R) approved for use in interstitial lung disease, improved pulmonary inflammation and helped wean off mechanical ventilation of three middle-aged obese COVID-19 patients where lung function restoration has been challenging52. While the adaptor protein SHC4 has not been experimentally demonstrated to modulate FGFR signaling, a 12-gene biomarker signature associated with melanoma contains FGFR2, FGFR3 and SHC453. In view of the limited knowledge of this understudied member of the SHC family, it is attractive to speculate that it may act downstream of FGFR or other associated growth factor receptors linked to angiogenesis, favoring immunothrombosis associated with COVID-19 ARDS. Additional experimentation is required to establish a sound scientific basis for these hypotheses. Moreover, the identity of the cells expressing SHC4, leading to its presence in the circulation is not known, also the focus of ongoing investigations. Taken together, our identification of FGFR and SHC4 signaling pathways distinguishing EP6 from other endophenotypes, supports further investigation of antagonists of those pathways to treat severe lung manifestations of COVID-19 and their potential use as biomarkers of severe disease activity.

Limitations and considerations

The data presented in this study come from individuals participating to BQC19, a prospective observational cohort built to study COVID-19 in Québec (Canada) with its specific population profile as reported previously10. While the number of participants was sufficient to establish the endophenotypes using the extensive proteomic profile available in BQC19, it was insufficient for traditional genome-wide association study (GWAS) to identify relations between SNVs and the identified endophenotypes54. Instead, we exploited the top SNVs that distinguished EP6 from other EPs in a pQTL analysis. Because these results show a potential genetic functional causality, it gives us the confidence that these associations are likely not due to random chance; however, the robustness of this approach needs to be further tested in other studies. A chronological bias may also be present, as most of the participants used for endophenotyping in this study were recruited during the first two waves of the pandemic (Figure S1), prior to widespread vaccination in Québec and prior to the appearance of the Omicron variant and sub-variants. Therefore, some of the features of the identified endophenotypes may change over the course of the pandemic. It will be essential to continue to assess the molecular profiles longitudinally to better understand the dynamic nature of host-pathogen interactions. It will also be interesting to compare the profiles of COVID-19 ARDS to other viral-induced ARDS, to identify commonalities as well as distinguishing features.

In this study, we used circulating proteome as determined by aptamers to identify endophenotypes, a task for which it is well suited as it can capture a dynamic landscape. However, several important considerations need to be mentioned. First, raw (unnormalized) expression values of the same aptamer can be used to compare different samples, but these values cannot be used to compare different aptamers against each other in the same sample, since they only show the relative abundance of expression and not absolute expression. As such, one needs to first normalize these values across samples (one aptamer at a time) and then subject them to follow-up analysis such as clustering, an approach that we adopted in this study. Second, since information only shows relative abundance, focusing on an individual aptamer and analyzing it will require additional measurements to establish absolute abundance. Moreover, if one wants to analyze aptamers individually (instead of the collective approach that we used in this study), they need to consider the effect of complexes and non-specific binding that may result in noisy data. As such, we suggest that the aptamer expression data be used collectively and after proper normalization, which enabled us to identify various EPs and important molecular mechanisms discussed in this study.

In this work, we developed a predictive model based on blood markers that enables us to generalize the definition of EPs to scenarios where data on circulating proteome is not available. This significantly increases the applicability of this approach and may enable automating identification of high-risk individuals in the clinic. However, time and follow-up studies are required to move this predictive model and the definitions of EPs from research to clinic and develop it as an acceptable clinical practice. Nonetheless, our approach can be used to study COVID-19 in different cohorts and identify characteristics that can guide the treatment of the disease.

Future directions

There is enormous amount of data within this study and BQC19 that can and should be exploited by the scientific and medical community to improve our understating of COVID-19 and novel emerging acute respiratory illnesses. Analyzing in more details the molecular profiles of the other EPs, in particular EP3-5, which lead to similar outcomes from distinct molecular pathways should further yield important insights into mechanisms of the disease. For example, EP5 is enriched in males and cardiovascular disease complications, while EP4 is enriched in female (like EP1) but with different distinct clinical trajectories pointing to sex-dependent and independent molecular mechanisms of disease.

Conclusion

The major strength of this study is its starting point: unsupervised analysis of a large and deeply phenotyped cohort. We showed that this approach can address both fundamental scientific questions pertaining to mechanisms of disease and help the medical community improve patient outcomes through early identification of patient that may follow a severe clinical course during COVID-19.

Methods

Datasets and preprocessing

The Biobanque québécoise de la COVID-19 (BQC19; www.quebeccovidbiobank.ca) is aimed at coordinating the collection of patients’ data and samples for COVID-19 related research. Data and samples were collected from ten sites across the province of Québec (Canada)10. BQC19 organizes the collected data, including clinical information and multi-omics experimental data, before making it available in successive releases. For this study, we used release #5 of the clinical data published in December 2021, the circulating proteome determined using SOMAmers54 and Metabolomics data55. BQC19 GWAS imputation data was generated by Tomoko Nakanishi at Brent Richards lab, Jewish General Hospital and McGill University. Detailed codes used for generating the data can be found in: https://github.com/richardslab/BQC19_genotype_pipeline

Our main corpus of analysis consisted of n = 1,362 hospitalized and SARS-CoV-2 positive patients (based on qRT-PCR) of BQC19. This included n = 731 patients for which both clinical and proteomic data were available as well as n = 631 patients for which proteomic data was not available, but their clinical data contained measurements for more than half (at least 11 out of 21) of the blood markers that we used as a validation set for the predictive model developed in this study.

We also obtained data (n = 731) corresponding to the circulating proteome measured between April 2, 2020 and April 20, 2021 by a multiplex SOMAmer affinity array (Somalogic, 4,985 aptamers) from BQC19 (release #3 Sep. 2021, associated patients’ data updated in release #5 Dec. 2021). When measurements of the same patients but at different time points were available, we used the one corresponding to the first time point. SomaScan is a biotechnological protocol commercialized by the Somalogic company. It relies on a set of artificial aptamers linked to a fluorophore and each designed to bind a single protein. Once added to the sample, the activity of each aptamer is measured through fluorescence and used to approximate the expression level of the targeted protein. SomaScan protocol comprises several levels of calibration and normalization to correct technical biases11. Log2 and Z-score normalization were performed on each aptamer separately in addition to the manufacturer’s provided normalized data (hybridization control normalization, intraplate median signal normalization, and median signal normalization). Since the data was analyzed by Somalogic in two separate batches, we applied the z-score transformation separately to each batch, to reduce batch effects. These additional transformations ensure that the measured values of different aptamers are comparable and can be used in cluster analysis.

We obtained metabolomic data (1,435 metabolites) from BQC19. We used the batch-normalized, missing values imputed and log-transformed version of the data.

Analyses of the GWAS dataset

For the GWAS analyses, annotation of SNVs were done using the biomaRt package56 from R57 and all analyses were done using R version 4.1.3. Quality control steps were derived in majority from a 2017 QC tutorial article58. At the beginning, we had 867,450 markers and 2,429 samples. We import Plink format data into R using the “read_plink” function from genio package59 from R. We removed 103,592 non ACGT bi-allelic markers. We calculated the predicted sex by looking at the rate of homozygote markers on chromosome 23. We removed 3,588 markers with call rates < 98%, 448,932 monomorphic markers and markers with MAF <0.05, and 28,092 markers with Hardy Weinberg equilibrium < 1E-6 (calculated by the “HWE.exact” function from the genetics package60 from R. For the EP6 cluster group analyses, we removed in addition 1,747 samples that were not in the cluster analysis, 8 samples with a sex discrepancy (based on predicted sex calculated earlier and reported sex), and 3 samples with a heterozygosity rate > 3 standard deviations. We finally removed a pair of samples who had approximately the same genome probably due to an error of manipulation. We couldn’t know which one was the right sample, so we removed both of them. We also had 2 pairs of individuals who had a pi-hat of ∼0.5 (meaning first degree relatives), we decided to keep one sample per pair, the one with the higher call rate. All other pairs of individuals had a pi-hat < 0.21 that is judge acceptable considering our population. Pi-hat were calculated with the “snpgdsIBDMoM” function from SNPRelate package61 from R. At the end of quality controls, 283,246 markers on 655 samples have been used to perform association analyses.

To perform the principal component analyses (PCA), we took a subsample of independent markers (pruning) with a maximum sliding window of 500,000 base pairs and a linkage disequilibrium (LD) threshold of 0.2 using the “snpgdsLDpruning” function from the SNPRelate package from R. We ran the PCA with the “snpgdsPCA” function from SNPRelate package from R. The first 2 principal components (PCs) were considered significant.

For the GWAS analyses of the 283,246 remaining markers between EP6 cluster compared to all others, we modeled a logistic regression with the dichotomous variable indicating if the participants belong to the EP6 cluster as the outcome variable. We used the additive model for markers as an independent variable and we adjusted the models with the first two PCs. Odds ratio and p-values were calculated on each model. QQ-plots have been performed as quality control of the models; p-values were plotted using “qqplot.pvalues” function from gaston package62 from R (data not shown). We compared the aptamers’ normalized level of expression (based on normalization described earlier) between the three groups of genotypes for each studied genes by performing standard ANOVA analyses followed by Tukey post hoc tests (referred to in this study as pQTL analysis). Since aptamers tested are limited compared to the SNVs, we have fixed significance p-values threshold below 1E-4 to report more SNVs instead of the more common 1E-5 suggestive threshold. Finally, to identify if the cluster EP6 may be explained by the aptamers and the SNVs, we performed multiple logistic regression analyses models include aptamers expression values, SNV genotypes (additive model) and the two principal components. OR are reported with 95% confidence intervals.

Consensus agglomerative clustering

Patients were clustered using agglomerative clustering, with Euclidean distance and Ward’s linkage12,13. To identify number of clusters k, we used the elbow method based on the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). More specifically, we calculated the AIC and BIC for clustering using k = 2, 3, …, 20 and used the Kneedle algorithm63 to identify the value of k corresponding to the “elbow”, where increasing the value of k does not provide much better modeling of the data. Kneedle identified k = 6 as the number of clusters based on both AIC and BIC (Figure 1A).

Given the number of clusters in the data, we then used consensus clustering with sub-sampling to obtain robust endophenotypes. We randomly sampled 80% of the patients 1000 times. Each time, we used agglomerative clustering above with k = 6 to identify clusters. Given these 1000 clusterings, we calculated the frequency of two patients appearing in the same cluster, when both were present in the randomly formed dataset. We then performed one final agglomerative clustering of these frequency scores to identify the six endophenotypes (Figure S2A and Figure 1B). The distribution of Rand-Index, showing the concordance between each one of the 1000 clusterings and the final consensus clustering is provided in Figure S2B (mean Rand-Index = 0.823), reflecting a high degree of consistency.

Metabolomic pathway characterization of EP6

The 1435 metabolites measured were organized into 122 sub-pathways in the original dataset (denoted as “SUB_PATHWAYS”). We first identified metabolites whose values were significantly higher or lower in EP6 compared to other EPs (two-sided MWU test, FDR<0.01). Then, we used these metabolites to perform pathway enrichment analysis (one-sided Fisher’s exact test) based on 122 pathways. The resulting p-values (Table S3) were then corrected for multiple tests using Benjamini-Hochberg FDR.

Nearest-centroid predictor based on blood markers

In order to predict endophenotypes from blood tests, we developed a missing-value resilient nearest-centroid classifier. We used the set of patients that were used to form the original EPs (n = 731) as the training set and the set of patients that did not have proteome data as the validation set (n = 631). First, we z-score normalized each blood marker across all the patients in the training set, one marker at a time. We then formed a blood marker signature (a vector of length 21) for each EP. Each element of an EP’s signature corresponds to the mean of the corresponding marker across all patients of that EP.

To predict the EP label of each patient in the test set, we first z-score normalized their blood marker measurements using the mean and standard deviation of the blood markers calculated from the training set. Then, we calculated the cosine distance between each test patient’s blood marker profile and the centroids (excluding missing values) and identified the nearest EP as the predicted EP (PEP) label of the patient.

Data Availability

Input data corresponding to the cohort can be obtained form BQC19 (www.quebeccovidbiobank.ca). The data generated as a result of the analyses are provided as tables and supplementary tables.

https://www.quebeccovidbiobank.ca

Permissions

The study was approved by the Institutional Ethics Review Board of the “Centre intégré universitaire de santé et de services sociaux du Saguenay-Lac-Saint-Jean” (CIUSSS-SLSJ) affiliated to Université de Sherbrooke [protocol #2021-369, 2021-014 CMDO – COVID19].

Acknowledgements

This work was made possible through open sharing of data and samples from the Biobanque québécoise de la COVID-19, funded by the Fonds de recherche du Québec - Santé, Génome Québec, the Public Health Agency of Canada and, as of March 2022, the Ministère de la Santé et des Services Sociaux du Québec. We thank all participants to BQC19 for their contribution. This study was supported by the Fonds de recherche du Québec - Santé (FRQS)-Cardiometabolic Health, Diabetes and Obesity Research Network (CMDO)-Initiative. This work was also supported by Natural Sciences and Engineering Research Council of Canada (NSERC) grant RGPIN-2019-04460 (AE).

References

  1. ↵
    Rubin, R. As their numbers grow, COVID-19 “long haulers” stump experts. Jama 324, 1381–1383 (2020).
    OpenUrl
  2. ↵
    Marshall, J. C. et al. A minimal common outcome measure set for COVID-19 clinical research. The Lancet Infectious Diseases 20, e192–e197 (2020).
    OpenUrlCrossRefPubMed
  3. ↵
    Hull, D. L. in PSA: Proceedings of the biennial meeting of the philosophy of science association. 653–670 (Cambridge University Press).
  4. ↵
    Berrettini, W. H. Genetic bases for endophenotypes in psychiatric disorders. Dialogues Clin Neurosci 7, 95–101 (2005). https://doi.org/10.31887/DCNS.2005.7.2/wberrettini
    OpenUrlPubMed
  5. ↵
    Russell, C. D. & Baillie, J. K. Treatable traits and therapeutic targets: goals for systems biology in infectious disease. Current opinion in systems biology 2, 140–146 (2017).
    OpenUrl
  6. ↵
    Blatti III, C. et al. Knowledge-guided analysis of “omics” data using the KnowEnG cloud platform. PLoS biology 18, e3000583 (2020). https://doi.org/10.1371/journal.pbio.3000583
    OpenUrlCrossRefPubMed
  7. Emad, A. et al. Superior breast cancer metastasis risk stratification using an epithelial-mesenchymal-amoeboid transition gene signature. Breast Cancer Research 22, 74 (2020). https://doi.org/10.1186/s13058-020-01304-8
    OpenUrl
  8. ↵
    Te Pas, M. F. W., Madsen, O., Calus, M. P. L. & Smits, M. A. The Importance of Endophenotypes to Evaluate the Relationship between Genotype and External Phenotype. International Journal of Molecular Sciences 18, 472 (2017).
    OpenUrl
  9. ↵
    Al-Hadrawi, D. S., Al-Rubaye, H. T., Almulla, A. F., Al-Hakeim, H. K. & Maes, M. Lowered oxygen saturation and increased body temperature in acute COVID-19 largely predict chronic fatigue syndrome and affective symptoms due to Long COVID: A precision nomothetic approach. Acta Neuropsychiatr, 1–12 (2022). https://doi.org/10.1017/neu.2022.21
  10. ↵
    Tremblay, K. et al. The Biobanque québécoise de la COVID-19 (BQC19)—A cohort to prospectively study the clinical and biological determinants of COVID-19 clinical trajectories. PloS one 16, e0245031 (2021). https://doi.org/10.1371/journal.pone.0245031
    OpenUrl
  11. ↵
    Gold, L. et al. Aptamer-based multiplexed proteomic technology for biomarker discovery. Nature Precedings, 1–1 (2010).
  12. ↵
    Ward Jr, J. H. Hierarchical grouping to optimize an objective function. Journal of the American statistical association 58, 236–244 (1963).
    OpenUrlCrossRefWeb of Science
  13. ↵
    Murtagh, F. & Legendre, P. Ward’s Hierarchical Agglomerative Clustering Method: Which Algorithms Implement Ward’s Criterion? Journal of Classification 31, 274–295 (2014). https://doi.org/10.1007/s00357-014-9161-z
    OpenUrl
  14. ↵
    Kaplan, E. L. & Meier, P. Nonparametric Estimation from Incomplete Observations. Journal of the American Statistical Association 53, 457–481 (1958). https://doi.org/10.1080/01621459.1958.10501452
    OpenUrlCrossRefWeb of Science
  15. ↵
    Gillespie, M. et al. The reactome pathway knowledgebase 2022. Nucleic Acids Research 50, D687–D692 (2021). https://doi.org/10.1093/nar/gkab1028
    OpenUrlCrossRef
  16. ↵
    Thomas, T. et al. COVID-19 infection alters kynurenine and fatty acid metabolism, correlating with IL-6 levels and renal status. JCI insight 5 (2020).
  17. ↵
    Zahedi, K. et al. The role of spermidine/spermine N 1-acetyltransferase in endotoxin-induced acute kidney injury. American Journal of Physiology-Cell Physiology 299, C164–C174 (2010).
    OpenUrlCrossRefPubMedWeb of Science
  18. ↵
    Pirnes-Karhu, S. et al. Spermidine/spermine N1-acetyltransferase activity associates with white blood cell count in myeloid leukemias. Experimental Hematology 42, 574–580 (2014).
    OpenUrl
  19. ↵
    Marfia, G. et al. Decreased serum level of sphingosine-1-phosphate: a novel predictor of clinical severity in COVID-19. EMBO molecular medicine 13, e13424 (2021).
    OpenUrl
  20. ↵
    Lachkar, F., Ferré, P., Foufelle, F. & Papaioannou, A. Dihydroceramides: their emerging physiological roles and functions in cancer and metabolic diseases. American Journal of Physiology-Endocrinology and Metabolism 320, E122–E130 (2021).
    OpenUrl
  21. ↵
    Lone, M. A. et al. Subunit composition of the mammalian serine-palmitoyltransferase defines the spectrum of straight and methyl-branched long-chain bases. Proceedings of the National Academy of Sciences 117, 15591–15598 (2020).
    OpenUrlAbstract/FREE Full Text
  22. ↵
    Han, G. et al. Identification of small subunits of mammalian serine palmitoyltransferase that confer distinct acyl-CoA substrate specificities. Proceedings of the National Academy of Sciences 106, 8186–8191 (2009).
    OpenUrlAbstract/FREE Full Text
  23. ↵
    Blachier, F., Andriamihaja, M. & Blais, A. Sulfur-Containing Amino Acids and Lipid Metabolism. The Journal of Nutrition 150, 2524S–2531S (2020). https://doi.org/10.1093/jn/nxaa243
    OpenUrl
  24. ↵
    Ye, C., Sutter, B. M., Wang, Y., Kuang, Z. & Tu, B. P. A metabolic function for phospholipid and histone methylation. Molecular cell 66, 180-193. e188 (2017).
    OpenUrlCrossRefPubMed
  25. ↵
    Parman, T. et al. Toxicogenomics and metabolomics of pentamethylchromanol (PMCol)-induced hepatotoxicity. Toxicological Sciences 124, 487–501 (2011).
    OpenUrlCrossRefPubMedWeb of Science
  26. ↵
    Stafford, J. H. & Thorpe, P. E. Increased exposure of phosphatidylethanolamine on the surface of tumor vascular endothelium. Neoplasia 13, 299–IN292 (2011).
    OpenUrlCrossRefPubMedWeb of Science
  27. ↵
    Ran, S., Downes, A. & Thorpe, P. E. Increased exposure of anionic phospholipids on the surface of tumor blood vessels. Cancer research 62, 6132–6140 (2002).
    OpenUrlAbstract/FREE Full Text
  28. ↵
    Port, J. R. et al. High-fat high-sugar diet-induced changes in the lipid metabolism are associated with mildly increased COVID-19 severity and delayed recovery in the Syrian hamster. Viruses 13, 2506 (2021).
    OpenUrl
  29. ↵
    Ataga, K. I. Hypercoagulability and thrombotic complications in hemolytic anemias. Haematologica 94, 1481–1484 (2009). https://doi.org/10.3324/haematol.2009.013672
    OpenUrlFREE Full Text
  30. ↵
    Sinauridze, E. I. et al. Platelet microparticle membranes have 50-to 100-fold higher specific procoagulant activity than activated platelets. Thrombosis and haemostasis 97, 425–434 (2007).
    OpenUrlCrossRefPubMed
  31. ↵
    Tavoosi, N. et al. Molecular Determinants of Phospholipid Synergy in Blood Clotting *. Journal of Biological Chemistry 286, 23247–23253 (2011). https://doi.org/10.1074/jbc.M111.251769
    OpenUrlAbstract/FREE Full Text
  32. ↵
    Majumder, R., Liang, X., Quinn-Allen, M. A., Kane, W. H. & Lentz, B. R. Modulation of prothrombinase assembly and activity by phosphatidylethanolamine. Journal of Biological Chemistry 286, 35535–35542 (2011).
    OpenUrlAbstract/FREE Full Text
  33. ↵
    Kim, K. et al. Pro-coagulant and pro-thrombotic effects of paclitaxel mediated by red blood cells. Thrombosis and Haemostasis 118, 1765–1775 (2018).
    OpenUrl
  34. ↵
    Sato, K., Iemitsu, M., Aizawa, K. & Ajisaka, R. Testosterone and DHEA activate the glucose metabolism-related signaling pathway in skeletal muscle. American Journal of Physiology-Endocrinology and Metabolism 294, E961–E968 (2008).
    OpenUrlCrossRefPubMedWeb of Science
  35. ↵
    Kajita, K. et al. Glucocorticoid-induced insulin resistance associates with activation of protein kinase C isoforms. Cellular signalling 13, 169–175 (2001).
    OpenUrlCrossRefPubMedWeb of Science
  36. ↵
    MacDonald, J. & Sprecher, H. Distribution of arachidonic acid in choline-and ethanolamine-containing phosphoglycerides in subfractionated human neutrophils. Journal of Biological Chemistry 264, 17718–17726 (1989).
    OpenUrlAbstract/FREE Full Text
  37. ↵
    Papageorgiou, C. et al. Disseminated intravascular coagulation: an update on pathogenesis, diagnosis, and therapeutic strategies. Clinical and Applied Thrombosis/Hemostasis 24, 8S–28S (2018).
    OpenUrlCrossRef
  38. ↵
    Esaki, Y., Hirokawa, K., Fukazawa, T. & Matsuda, T. Immunohistochemical study on the liver in autopsy cases with disseminated intravascular coagulation (DIC) with reference to clinicopathological analysis. Virchows Archiv A 404, 229–241 (1984).
    OpenUrl
  39. ↵
    Barnes, B. J. et al. Targeting potential drivers of COVID-19: Neutrophil extracellular traps. Journal of Experimental Medicine 217 (2020).
  40. Bonaventura, A. et al. Endothelial dysfunction and immunothrombosis as key pathogenic mechanisms in COVID-19. Nature Reviews Immunology 21, 319–329 (2021).
    OpenUrlCrossRefPubMed
  41. ↵
    Ding, J. et al. A network-informed analysis of SARS-CoV-2 and hemophagocytic lymphohistiocytosis genes’ interactions points to Neutrophil extracellular traps as mediators of thrombosis in COVID-19. PLoS Computational Biology 17, e1008810 (2021). https://doi.org/10.1371/journal.pcbi.1008810
    OpenUrl
  42. ↵
    Blasco, A. et al. Assessment of neutrophil extracellular traps in coronary thrombus of a case series of patients with COVID-19 and myocardial infarction. JAMA cardiology 6, 469–474 (2021).
    OpenUrl
  43. Desilles, J. P. et al. Impact of COVID-19 on thrombus composition and response to thrombolysis: Insights from a monocentric cohort population of COVID-19 patients with acute ischemic stroke. Journal of Thrombosis and Haemostasis 20, 919–928 (2022).
    OpenUrl
  44. Englert, H. et al. Defective NET clearance contributes to sustained FXII activation in COVID-19-associated pulmonary thrombo-inflammation. EBioMedicine 67, 103382 (2021).
    OpenUrl
  45. Leppkes, M. et al. Vascular occlusion by neutrophil extracellular traps in COVID-19. EBioMedicine 58, 102925 (2020).
    OpenUrlPubMed
  46. Middleton, E. A. et al. Neutrophil extracellular traps contribute to immunothrombosis in COVID-19 acute respiratory distress syndrome. Blood 136, 1169–1179 (2020).
    OpenUrlCrossRefPubMed
  47. Obermayer, A. et al. Neutrophil extracellular traps in fatal COVID-19-associated lung injury. Disease markers 2021 (2021).
  48. Ouwendijk, W. J. et al. High levels of neutrophil extracellular traps persist in the lower respiratory tract of critically ill patients with coronavirus disease 2019. The Journal of infectious diseases 223, 1512–1521 (2021).
    OpenUrl
  49. Petito, E. et al. Association of neutrophil activation, more than platelet activation, with thrombotic complications in coronavirus disease 2019. The Journal of infectious diseases 223, 933–944 (2021).
    OpenUrl
  50. ↵
    Skendros, P. et al. Complement and tissue factor–enriched neutrophil extracellular traps are key drivers in COVID-19 immunothrombosis. The Journal of clinical investigation 130, 6151–6157 (2020).
    OpenUrl
  51. ↵
    Smadja, D. M. et al. Placental growth factor level in plasma predicts COVID-19 severity and in-hospital mortality. Journal of Thrombosis and Haemostasis 19, 1823–1830 (2021).
    OpenUrl
  52. ↵
    Bussolari, C. et al. Case Report: Nintedaninb May Accelerate Lung Recovery in Critical Coronavirus Disease 2019. Frontiers in Medicine 8, 766486 (2021).
    OpenUrl
  53. ↵
    Liu, W., Peng, Y. & Tobin, D. J. A new 12-gene diagnostic biomarker signature of melanoma revealed by integrated microarray analysis. PeerJ 1, e49 (2013). https://doi.org/10.7717/peerj.49
    OpenUrl
  54. ↵
    Zhou, S. et al. A Neanderthal OAS1 isoform protects individuals of European ancestry against COVID-19 susceptibility and severity. Nature Medicine 27, 659–667 (2021). https://doi.org/10.1038/s41591-021-01281-1
    OpenUrl
  55. ↵
    Ford, L. et al. Precision of a Clinical Metabolomics Profiling Platform for Use in the Identification of Inborn Errors of Metabolism. The Journal of Applied Laboratory Medicine 5, 342–356 (2020). https://doi.org/10.1093/jalm/jfz026
    OpenUrl
  56. ↵
    Durinck, S., Spellman, P. T., Birney, E. & Huber, W. Mapping identifiers for the integration of genomic datasets with the R/Bioconductor package biomaRt. Nature Protocols 4, 1184–1191 (2009). https://doi.org/10.1038/nprot.2009.97
    OpenUrl
  57. ↵
    R Core Team. R: A Language and Environment for Statistical Computing. https://www.R-project.org/. (R Foundation for Statistical Computing, Vienna, Austria, 2022).
  58. ↵
    Marees, A. T. et al. A tutorial on conducting genome-wide association studies: Quality control and statistical analysis. Int J Methods Psychiatr Res 27, e1608 (2018). https://doi.org/10.1002/mpr.1608
    OpenUrlCrossRefPubMed
  59. ↵
    Ochoa, A. genio: Genetics Input/Output Functions. R package version 1.1.1. https://CRAN.R-project.org/package=genio (2022).
  60. ↵
    Warnes, G., Gorjanc, G., Leisch, F. & Man, M. genetics: Population Genetics. R package version 1.3.8.1.3. https://CRAN.R-project.org/package=genetics (2021).
  61. ↵
    Zheng, X. et al. A high-performance computing toolset for relatedness and principal component analysis of SNP data. Bioinformatics 28, 3326–3328 (2012). https://doi.org/10.1093/bioinformatics/bts606
    OpenUrlCrossRefPubMedWeb of Science
  62. ↵
    Perdry, H. & Dandine-Roulland, C. gaston: Genetic Data Handling (QC, GRM, LD, P.A) & Linear Mixed Models. R package version 1.5.7. https://CRAN.R-project.org/package=gaston (2020).
  63. ↵
    Satopaa, V., Albrecht, J., Irwin, D. & Raghavan, B. in 2011 31st international conference on distributed computing systems workshops. 166–171 (IEEE).
Back to top
Posted November 04, 2022.
Download PDF

Supplementary Material

Data/Code
Email
Unsupervised clustering of SARS-CoV-2 positive hospitalized patients identifies six endophenotypes of COVID-19 and points to FGFR and SHC4-signaling in acute respiratory distress syndrome
William Ma, Antoine Soulé, Katelyn Liu, Catherine Allard, Karine Tremblay, Simon Rousseau, Amin Emad
medRxiv 2022.11.02.22281834; doi: https://doi.org/10.1101/2022.11.02.22281834
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area