Introduction
COVID-19 has had an enormous global impact, affecting millions of people, causing many deaths, and still requiring a great effort to understand the mechanism of COVID-19 disease better. SARS-CoV-2 acts similarly to H1N1, the disease caused by Influenza A type virus [
1]. Although they utilize different receptors for viral entry, SARS-CoV-2 using the angiotensin-converting enzyme type 2 (ACE2) receptor and H1N1 using sialic acid receptors, they have both been implicated in affecting the renin–angiotensin–aldosterone system (RAAS) pathway [
2‐
4] and have been shown to cause ARDS [
4‐
7].
In severe forms, COVID-19 and H1N1 can result in acute respiratory distress syndrome (ARDS), leading to the development of multiorgan damage [
4‐
7]. Mortality rates for patients with ARDS are as high as 38%, with no specific ARDS pharmacologic therapy proven to date [
8]. Despite this, early non-specific therapy has improved outcomes, illustrating the importance of timely diagnosis [
9]. Few diagnostic biomarkers have been proposed, found, or validated for ARDS.
Metabolomics studies can help reveal altered metabolic pathways during COVID-19 infection (as well as other viral and bacterial infections) and the development of ARDS, providing insight into the disease processes. In addition, it provides an opportunity to investigate how SARS-CoV-2 affects the host's metabolism and immune response. Most metabolomics studies involving COVID-19 compare SARS-CoV-2-infected patients with normal controls and focus on differentiating severity. However, few studies have been done comparing ARDS caused by different etiologies [
10‐
12]. Overall, perturbed pathways currently observed in COVID-19 include pyruvate metabolism, kynurenine pathways, and amino acid metabolism—specifically tryptophan metabolism [
13‐
19].
We aimed first to compare the metabolomic profiles between different infectious etiologies of ARDS: COVID-19-associated ARDS (C19/A), bacterial pneumonia-associated ARDS (PNA/A), and H1N1-associated ARDS (H1N1/A). We additionally sought to compare the metabolomic profiles of patients with COVID-19 ARDS admitted to the ICU to those admitted to the hospital but not requiring ICU admission (i.e., those with COVID-19 pneumonia but not severe enough to require ICU admission). Subsequently, we sought to propose a bedside formula by identifying the minimal metabolites associated with the mechanisms differentiating these groups for early diagnosis of COVID-19 ARDS vs other infectious causes of ARDS and for COVID-19 severity assessment of patients.
Materials and methods
Data sources and measurements
We collected plasma samples from four different tissue banks in Canada. All samples were plasma collected, isolated, and managed in a similar fashion. Each study group consisted of 25 patients with plasma samples drawn within 24 h of ICU admission for ARDS patients and within 24 h of hospital admission for COVID-19 pneumonia (C19/P) patients not sick enough to be admitted to the ICU. C19/A is a group of COVID-19-infected (PCR-positive) ICU patients with ARDS who were ventilated on the first day of ICU admission. C19/P is a group of COVID-19-infected (PCR positive) non-ICU hospitalized pneumonia patients on the first day of admission to the hospital. PNA/A is a group of non-COVID-19, bacterial pneumonia-associated (culture-positive) ARDS patients ventilated on the first day of ICU admission. H1N1/A is a group of non-COVID-19, H1N1-associated (PCR positive) ARDS patients ventilated on the first day of ICU admission. Finally, the CTL group consisted of a group of patients not suspected of having pneumonia (viral or bacterial) mechanically ventilated ICU controls who were either postoperative patients included in the study with samples taken while ventilated in the ICU 6–24 h following major cardiovascular surgery, such as coronary artery bypass graft (CABG), or patients with severe neurological diseases, such as stroke, subarachnoid hemorrhage, or meningitis without pneumonia, with samples taken within 24 h of ICU admission while intubated and ventilated.
C19/A and C19/P samples were collected as part of the ARBs CORONA I multicenter study [
20,
21]. CTL and PNA/A samples were collected at Foothills Medical Center and Peter Lougheed Center (Calgary, AB, Canada) during the period 2009–2014 and processed similarly (as published in the Canadian Critical Care Translational Biology Group website protocols) and stored at −80 °C at the University of Calgary as part of the CCEPTR ICU tissue bank. H1N1/A samples were collected during the H1N1 pandemic in 2009 and processed identically to the CCCTBG website protocol. They were stored at −80 °C and made available from Winnipeg, Manitoba. In addition, C19/AV, a validation group of COVID-19-associated ARDS, consisted of 25 patients with plasma samples collected under identical conditions as the C19/A samples from the ARBs Corona I study (i.e., patients with COVID-19-associated ARDS (PCR positive) admitted to the ICU with samples taken within 24 h of ICU admission) but are from the University of Toronto and the University of Calgary. These samples were processed identically to the CCCTBG and CCEPTR protocols and stored in aliquots at −80 °C.
The Berlin definition was used for ARDS diagnosis. Two investigators verified the diagnosis for the bacterial pneumonia-associated ARDS in particular—Dr. Brent Winsrton and an MD, Ph.D. student, Dr. Sayed Metwaly, from a previously published study (Metwaly, S. et al. "ARDS Metabolic Fingerprints: Characterization, Benchmarking, and Potential Mechanistic Interpretation." Am J Physiol Lung Cell Mol Physiol. 2021 May 5, 321: L79–L90. doi,
https://doi.org/10.1152/ajplung.00077.2021.) The diagnosis of ARDS in the H1N1 cohort was made by Dr. Anand Kumar and Dr. Brent Winston, and Dr. Brent Winston verified the ARDS diagnosis in the COVID-19 cohort. We collected clinical information such as age, sex, PaO
2/FiO
2, type of COVID-19 medication used (e.g., steroids or Remdesivir), ventilation support, COVID-19 test result, H1N1 test result, bacterial culture result, survival status at 28 days from hospital admission and ICU and hospital discharge.
Study design
All study groups are matched by age and sex. Patients were chosen randomly from each cohort if they matched age and sex, and plasma samples were available. Age matching was done ± 5 years. Four groups (C19/A, PNA/A, H1N1/A, and CTL) underwent quantitative metabolomics analysis (as described above) followed by multiple and pairwise comparisons of the metabolite findings. We started with characterizing the metabolomic profile of each group using all metabolites included in the study and ran simultaneous comparisons of their profiles. We then ran six pairwise comparisons as follows. The three pairwise comparisons, each with CTL as a reference group and the other ARDS groups for direct comparison, allowed us to see how the specific ARDS subgroups deviated from the ICU-ventilated control group (CTL) regarding metabolomic profile. In addition, the other three pairwise comparisons informed us of how different the infectious-mediated ARDS groups are. To assess the severity of COVID-19 patients, we compare C19/A to C19/P (the C19/P patients had COVID-19 pneumonia but not ARDS and were not severe enough to be admitted to the ICU) with plasma samples taken on day one after hospital admission for C19/P or day one after ICU admission for C19/A. Finally, plasma metabolomics of a validation cohort for COVID-19 ARDS (C19/AV) patient samples was compared to C19/A patient samples to validate our COVID-19 ARDS findings. The Conjoint Health Research Ethics Board, University of Calgary, has reviewed and approved this study (Ethics ID: REB20-0654). We used 25 patients per cohort based on a previous study [
22].
Sample preparation
For organic acid quantification, 50 µl plasma samples were thawed on ice, followed by adding 150 µl of ice-cold methanol and 10 µl of isotope-labeled standards. The mixtures were kept overnight at −20 °C to precipitate proteins, followed by centrifugation at 13,000 × g for 20 min. A total of 50 µl of supernatant of the extracts were added to the center of a 96-well plate, followed by adding a 3-nitrophenylhydrazine reagent to the extract and incubated for two hours. Butylated hydroxytoluene (2 mg/ml) stabilizer and water were added to the extract before LC–MS/MS injection.
For amino acid and lipid quantifications, samples were vortexed and centrifuged, adding 10 µl of samples to a 96-well plate and a stream of nitrogen-dried samples. Phenyl-isothiocyanate reagent was added to the samples in the plate. Samples were incubated and then dried using an evaporator. Three hundred microliters of extraction solvent was added to the analytes. Extracts were centrifuged to the lower part of the 96-well plate; a dilution step was performed using 0.2% formic acid in the water and 0.2% formic acid in acetonitrile.
Plasma-based targeted metabolomics was performed to quantify the concentration of 143 metabolites developed by The Metabolomics Innovation Center (TMIC) at the University of Alberta, Edmonton (see list of metabolites in the supplement) [
16,
23] and as we have previously done [
24]. Reverse-phase liquid chromatography-tandem mass spectrometry (LC–MS/MS) was applied to analyze amino acids, biogenic amines, and organic acids. Direct infusion tandem mass spectrometry (DI-MS/MS) was applied to quantify glycerophospholipids, lysophosphatidylcholines (lysoPCs), and phosphatidylcholines (PCs), acylcarnitines (Cs), and sphingomyelins (SMs). Mass spectrometry was analyzed using an ABSciex 4000 Qtrap tandem MS instrument (Applied Biosystems/MDS Analytical Technologies, Foster City, CA, USA). An Agilent 1260 series UHPLC system (Agilent Technologies, Palo Alto, CA) was combined with MS for LC–MS/MS [
16,
23].
LC–MS/MS analyses
For chromatography, an Agilent reversed-phase Zorbax Eclipse XDB C18 column (3.0 mm × 100 mm, 3.5 μm particle size, 80 A pore size) with a Phenomenex (Torrance, CA, USA) Security Guard C18 pre-column (4.0 mm × 3.0 mm) was used for analyzing amino acids and biogenic amines. The parameters for LC–MS/MS analysis were as follows: Mobile phase A was 0.2% (v/v) formic acid in the water, and mobile phase B was 0.2% (v/v) formic acid in acetonitrile. The gradient parameters were t = 0 min, 0% B; t = 0.5 min, 0% B; t = 5.5 min, 95% B; t = 6.5 min, 95% B; t = 7.0 min, 0% B; and t = 9.5 min, 0% B. The chromatography column was set as 50 ºC. Ten microliters of samples was injected into the column with a flow rate of 300 µl/min.
For chromatography of organic acids, mobile phase A was 0.01% (v/v) formic acid in the water, and mobile phase B was 0.01% (v/v) formic acid in methanol. The gradient parameters were t = 0 min, 30% B; t = 2.0 min, 50% B; t = 12.5 min, 95% B; t = 12.5 min, 100% B; t = 13.5 min, 100% B; and t = 13.6 min, and finally 30% B for 4.4 min. The chromatography column was set as 40 ºC. Ten microliters of samples was injected into the column with a flow rate of 300 µl/min.
DI-MS/MS analysis
The direct infusion was performed using the connection of the LC autosampler to the MS ion source using red PEEK tubing. The mobile phase was set by mixing 60 µl of formic acid, 10 ml of water, and 290 ml of methanol. The flow rate was t = 0 min, 30 µl/min; t = 1.6 min, 30 µl/min; t = 2.4 min, 200 µl/min; t = 2.8 min, 200 ul/min' and t = 3.0 min, 30 µl/min. Twenty microliters of samples was injected into the MS.
A seven-point standard calibration curve was obtained for each metabolite to quantify organic acids, amino acids, and biogenic amines. The signal intensity of each metabolite was corrected to the corresponding isotope-labeled internal standard, and the known concentrations were calculated based on the quadric regression with a 1/ × 2 weighting. The concentrations of lipids and glucose were calculated semi-quantitatively using a single-point calibration of representative metabolites built based on the same class of compound with the same core structure, assuming a linear regression through zero. Analyst 1.6.2 and MultiQuant 3.0.3 were used to analyze all metabolites in the assay.
Both LC–MS/MS and DI-MS/MS analytical platforms were applied in a targeted approach to quantify 143 metabolites, including different metabolite classes. Mass spectrometry analysis was performed using an ABSciex 4000 Qtrap tandem MS instrument (Applied Biosystems/MDS Analytical Technologies, Foster City, CA, USA). An Agilent 1260 series UHPLC system (Agilent Technologies, Palo Alto, CA) was combined with MS for LC–MS/MS.[
16,
23].
Statistical analysis and validation
As previously done [
24], we processed the raw metabolite concentration data with median-fold normalization, logarithm transformation, and z-score standardization to identify outliers, stabilize variability, and give metabolites an equal contribution weight for model determination (i.e., we normalized the raw data as a standard processing procedure for metabolomics data). Partial least-squares discriminant analysis (PLS-DA) was used as a major analytical model because of the multicollinearity in high-dimensional metabolomic data. A model fitted to the data was assessed by three metrics—R
2Y (the amount of variance explained by a model of fit), Q
2Y (cross-validated R
2, a measure of goodness of prediction of the model), and response permutation test (for validity and to prevent overfitting). The performance of a model was discussed using sensitivity (Se), specificity (Sp), and area under the receiver operator characteristics curve (AUROC). We defined a metabolite selection rule by considering (1) the variable importance of a projection (VIP) score > 1.0 from PLS-DA, (2) the absolute value of logarithm with base 2 of fold change > 1, (3) nonzero coefficients from a penalized logistic regression. The number of latent variables was identified by threefold cross-validation when the PLS-DA model was run. This selection was based on 1000 resamples. All analyses were carried out using a standard statistical computing language and environment, R-4.0.0. The Kyoto Encyclopedia of Genes and Genomes (KEGG) was employed to understand chemical classes and biological pathways. The metabolites with VIP > 1 were projected onto their corresponding KEGG pathways using MetaboAnalyst.
Discussion
It has been proposed that ARDS due to COVID-19 has different clinical features compared with ARDS by other causes [
25], including both viral-associated and bacterial pneumonia-associated ARDS [
26]. Thus, we investigated whether different infectious causes of ARDS, both viral and bacterial, altered plasma metabolites reflecting different mechanisms of injury. Indeed, our study does show metabolomic differences between viral causes of ARDS (specifically, C19/A and H1N1/A) that were significant for taurine and hypotaurine metabolism, pyruvate metabolism, citrate cycle (TCA cycle), lysine degradation, and glycerophospholipid metabolism. We also found distinct differences between bacterial pneumonia-associated ARDS (PNA/A) and viral-associated ARDS (both C19/A and H1N1/A) in taurine and hypotaurine, arginine and proline, and histidine metabolisms. Finally, we found distinct metabolite differences in COVID-19 severity as reflected by those COVID-19 patients requiring ICU admissions (C19/A) vs those that do not (C19/P) in phenylalanine, tyrosine and tryptophan biosynthesis, lysine degradation, and tyrosine metabolism.
Our main finding is significant metabolomic differences between COVID-19 and other viral causes of ARDS, specifically H1N1. Metabolic pathways also differ between COVID-19 and bacterial pneumonia-associated ARDS and non-ARDS ICU-ventilated control patients. Our data also reveal a high level of similarity between the C19/A and C19/P involved metabolites when compared to H1N1/A, PNA/A, and CTL groups. COVID-19 ARDS is more severe than COVID-19 pneumonia without ARDS who are not admitted to the ICU and is characterized by increased branched-chain amino acids (BCAAs), glucose, some short- and long-chain acylcarnitines, and decreased acetyl-ornithine, propionic acid, and long phosphatidylcholines (PC 40:1 and 40:2).
We found that lipid metabolism is important in viral-mediated ARDS, as seen in the heatmap of COVID-19 ARDS and as noted by others for H1N1 [
27] and COVID-19 [
28]. This is an interesting and potentially important finding as there is early evidence that using a PCSK9 inhibitor alters COVID-19 inflammation and outcome [
29]. Just how lipid metabolism affects inflammation in COVID-19 is not known, but there is an association between lipid disorders and COVID-19 severity [
30], and several studies involving statin use in COVID-19 have been undertaken [
30].
We also note that aromatic amino acid and lysine metabolism were highlighted in the differentiation of COVID-19 ARDS and COVID-19 pneumonia without ARDS. Our findings agree with others, where arginine metabolism, glycolytic pathway, and one-carbon metabolism were highlighted as the most perturbed metabolic phenotype in COVID-19 [
31‐
33]. Importantly, this may be used to potentially predict those individuals with COVID-19 pneumonia that will develop more severe disease (i.e., ARDS) and, therefore, may need close attention for the need to transfer to the ICU. This may be a marker of COVID-19 severity; however, this will need to be validated in future investigations.
COVID-19 metabolomics studies have generally been compared to normal controls. There have been many small COVID-19 metabolomics studies; a few will be highlighted here. López-Hernández et al. [
16] found differences in serum metabolites between COVID-19-negative and non-hospitalized COVID-19-positive individuals, including increased kynurenine/tryptophan ratio, lysoPC(aC26:0), and pyruvic acid. Examining COVID-19-positive non-hospitalized and hospitalized individuals, they found increased decanoylcarnitine (C10:2), butyric acid, and pyruvic acid. Finally, when COVID-19-positive hospitalized patients were compared to COVID-19-positive intubated patients, they found increased lysophosphatidylcholine (lysoPC aC28:0) [
17]. When they compared severe COVID-19 patients to normal controls, they found increased glutamate, aspartic acid, kynurenine, and lysoPCs and decreased glutamine, citrulline, tryptophan, serotonin, and nicotinamide mononucleotide in the severe COVID-19 patients [
18]. Examining plasma using targeted DI-MS/MS from three small groups: 10 patients with COVID-19, ten patients who were COVID-19 negative, and ten normal controls, they found kynurenine was the most significantly increased metabolite between COVID-19-positive and healthy controls and decreased metabolites included: arginine, sarcosine, lysoPC. They also found increased kynurenine and arginine/kynurenine ratio in COVID-19-positive vs. COVID-19-negative patients [
34]. When serum and plasma profiles of COVID-19 patients were compared to healthy controls, they found involvement of tryptophan metabolism via the kynurenine pathway and elevated tryptophan, kynurenine, and 3-hydroxykynurenine. Blasco et al
. [
35] examined plasma in 55 COVID-19-positive patients and 45 healthy controls, and they found involvement of the cytosine and tryptophan‐nicotinamide pathways that were linked to the tryptophan-kynurenine pathway and increased cytosine levels in COVID-19 patients. Despite this wide range of metabolomics findings, the overall perturbed metabolic pathways currently observed in COVID-19 include pyruvate metabolism, kynurenine pathways, and amino acid metabolism; this was linked specifically to tryptophan metabolism [
13‐
19]. Our findings, in general, agree with this summary. Our data revealed the same level of increased kynurenine in C19/A, H1N1/A, and PNA/A compared to CTL, suggesting immune dysregulation [
36] due to viral and bacterial ARDS. Although it has been shown that the kynurenine/tryptophan ratio may be correlated with COVID-19 severity, our data showed non-significant kynurenine and tryptophan concentrations when C19/A and C19/P were compared. It has been shown that altered unsaturated lysophosphatidylcholines are associated with COVID-19 infection, with some lipid types showing decreased and others showing increased levels. Nonetheless, LysoPCs 16:0, 18:0, 18:1, and 18:2 were reduced in COVID-19-positive individuals [
37,
38]. However, our findings showed these lysoPCs were significantly increased in C19/A and C19/P compared to H1N1/A, PNA/A, and CTL. Our data also demonstrated that LysoPCs C16:0, C18:1, and C18:2 were specifically elevated in PNA/A. The association of reduced LysoPC compounds with mortality and severity among bacterial CAP patients has been previously shown [
39]. In a previous study, the metabolomic investigation of COVID-19 and H1N1 patients with ARDS showed distinct metabolic phenotypes between these two viral causes (with model characteristics showing
Q2 = 0.89 and AUC = 1.0) [
40]. Our data agreed with this study to show significantly increased glucose, lactate, glutamate, and fatty acid levels in COVID-19 ARDS vs. H1N1 ARDS. Data in the present study revealed specific metabolites involved with PNA/A that are significantly different from C19/A and H1N1/A; however, these were not significantly different between C19/A and H1N/A, suggesting that these metabolites could be specific to viral infections with ARDS compared with bacterial pneumonia. This suggests increased sarcosine, lysoPC 16:0, C18:1, C18:2, and decreased levels of homovanillic acid, isobutyrate, glucose, histidine, and methionine sulfoxide were associated with viral infections.
Just as we show the importance of different pathways between COVID-19 and other causes of ARDS, others have shown that aromatic amino acids and one-carbon metabolism differ between ARDS patients compared to healthy controls[
8]. We extend these findings by showing that specific metabolomic pathways characterize different infectious causes of ARDS. COVID-19 ARDS had prominent arginine metabolism, H1N1 ARDS had increased taurine and hypotaurine metabolism, while bacterial pneumonia ARDS had increased alanine, aspartate, and glutamate metabolism.
COVID-19 ARDS is further differentiated by pyruvate metabolism and glutamine/glutamate metabolism compared to H1N1 ARDS and bacterial pneumonia ARDS. Notably, taurine/hypotaurine, histidine, and one-carbon metabolism were more specific to H1N1 ARDS.
Similarly, a previous
1H-NMR plasma metabolomics study examining H1N1 pneumonia vs. controls [
27] shows many similar elevated metabolites (including beta-alanine, phenylalanine, and ornithine) and decreased (citrate, taurine, glycine, glutamine, and serine) metabolites as we show here using DI/LC–MS/MS. As previously shown, the data here reveal that aminoacyl-tRNA biosynthesis is the most impactful metabolomic pathway comparing H1N1 ARDS patients vs. ICU-ventilated controls.
ARDS is clinically heterogeneous [
41‐
43], and our study and others [
8] highlight several potential metabolomic sub-phenotypes of COVID-19 and other viral causes of ARDS. We add new insights regarding metabolomic sub-phenotypes within COVID-19 that mark the severity of illness such that COVID-19 pneumonia non-ICU patient metabolites differ from metabolites found in COVID-19 ARDS ICU patients. ARDS has been previously subphenotypes into hyper- and hypo-inflammatory using cytokine analyses [
41‐
43]; however, we did not examine cytokines in this study. Others have begun exploring some metabolomics differences between hyper- and hypo-inflammatory ARDS phenotypes [
22,
44].
We believe C19/AV was useful to validate the findings from C19/A externally. For this conclusion, we discussed how similar C19/A and C19/AV are using adjusted p-values from multiple tests. Subsequently, we showed from a parallel analysis that approximately 80% of metabolites have the same conclusion between C19/A and C19/AV compared to C19/P, H1N1, and PNA/A, respectively, with different measures.
One must consider limitations of this study, such as, a relatively small sample size in each cohort and the use of targeted quantitative metabolomics that captured only 143 metabolites. Finally, our cohorts were drawn from sample collections at three different periods or dates. Although all samples were prepared similarly and frozen at −80 °C and management of ARDS over this period has not changed significantly, these factors may have affected the results. However, we believe our findings are robust and we show this by applying 1,000 resampling to our analysis.
This study is unique in that it compares three infectious causes of ARDS (COVID-19, H1N1, and bacterial pneumonia-associated ARDS) as well as comparing COVID-19 pneumonia patients not sick enough to be admitted to the ICU vs COVID-19 ARDS admitted to the ICU. We found distinct differences in metabolites between bacterial pneumonia-associated ARDS (PNA/A) and viral-associated ARDS (caused by COVID-19 (C19/A) and H1N1 influenza (H1N1/A)). Importantly, we also see differences between viral causes of ARDS, namely COVID-19 ARDS (C19/A) and H1N1 ARDS (H1N1/A). Finally, we found metabolomics differences between COVID-19 pneumonia non-ICU patients and COVID-19 ARDS ICU patients with metabolite changes reflecting the severity of the disease (which may be used to help define which COVID-19 pneumonia patients may require ICU care early in the progression to ARDS).
Acknowledgements
We would like to thank all of the ARBs CORONA I participating centers, researchers, and research coordinators for collecting patient samples and patient information. We thank Josee Wong and the Critical Care Epidemiologic and Biologic Tissue Resource (CCEPTR), a tissue bank at the University of Calgary, for managing samples and patient data. We thank Drs. Anand Kumar, Ma Lou, and their team in Winnipeg for the H1N1 patient samples and information, as well as Drs. Uriel Trahtemberg, Claudia dos Santos, and Andrew Baker and their team in Toronto for collecting COVID-19 patient samples and information for validation. We thank Dr. David Wishart, The Metabolomics Innovation Centre (TMIC), and the University of Alberta for helping us perform metabolomics measurements on patient samples. We also thank all the patients, nurses, and physicians for participating in and assisting with this study.
*ARBs CORONA I investigators: J.A. Russell, K.R. Walley, J. Boyd, T. Lee, J. Singer (St. Paul's Hospital [Coordinating Centre], Vancouver, British Columbia [BC], Canada); D. Sweet, K. Tran (Vancouver General Hospital, Vancouver, B.C., Canada); S. Reynolds (Royal Columbian Hospital, Vancouver, B.C., Canada); G. Haljan (Surrey Memorial Hospital, Surrey, B.C., Canada); M. Cheng, D. Vinh (McGill University Centre Hospital, Montreal, Quebec, Canada);T. Lee (Jewish General Hospital, Montreal, Quebec, Canada); F. Lamontagne (University de Sherbrooke, Sherbrooke, Quebec, Canada); B. Winston (Foothills Medical Centre, Calgary, Alberta, Canada); O. Rewa (University of Alberta, Edmonton, Alberta, Canada); J. Marshall, A. Slutsky (St. Michael's Hospital, Toronto, Ontario, Canada); A. McGeer, V. Sivanantham (Mount Sinai Hospital, Toronto, Ontario, Canada); R. Fowler (Sunnybrook and Women's College Health Science Centre, Toronto, Ontario, Canada); D. Maslove, S. Perez Patrigeon (Kingston General Hospital, Kingston, Ontario, Canada); RAS assays: KD. Burns (Ottawa Hospital Research Institute, University of Ottawa, Ottawa, Ontario, Canada)
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.