This study was approved by the Conjoint Health Research Ethics Board (CHREB) at the University of Calgary (Ethics ID: REB23-0457_RENI). Additionally, ethics approval for the collection and use of plasma samples from the Biobanque Québécoise de la COVID-19 (BQC-19) was reviewed and approved by the Research Ethics Board of the Center Hospitalier de l’Université de Montréal (REB-CHUM). The BQC-19, a decentralized biobank managed by McGill University and the Research Institute of the McGill University Health Centre (RI-MUHC), contains data and biological specimens from approximately 3,000 COVID-19 patients in Quebec, Canada. Plasma samples from 109 hospitalized patients were included in this study, with 53 infected with the SARS-CoV-2 Delta variant and 56 with the Omicron variant, manually matched by age (± 5 years) and sex using an Excel spreadsheet.

Inclusion criteria required patients to (1): be hospitalized with confirmed COVID-19 infection via real-time polymerase chain reaction (RT-PCR) (2), be aged ≥18 years (3), provide consent to participate in the biobank and related studies, and (4) be categorized by severity groups using the NIH classification criteria, as well as vaccination status following the criteria established by the Government of Canada for complete and incomplete vaccination (Canada Go, 2023). Vaccine product/brand was not uniformly available across participants; therefore, vaccination was analyzed as completeness (≥2 vs. <2 doses). Among the subset with product information, most received mRNA vaccines—Pfizer-BioNTech (n=20) or Moderna (n=6)—with fewer adenoviral-vector (AstraZeneca/Covishield; n=2) or protein-subunit (Novavax; n=1) products. Exclusion criteria included (1): patients testing negative by RT-PCR (2), patients or surrogates unwilling to participate in the biobank and studies (3), patients <18 years old, and (4) patients who did not have a plasma sample from day 1 and at least one from either day 2 or day 7.

Blood samples were collected in acid citrate dextrose (ACD) tubes, then centrifuged at 850×g for 10 minutes at room temperature. Plasma was transferred into 500 µL and 250 µL aliquots and stored at −80°C. Samples were drawn on days 1, 2, and 7 of hospitalization, resulting in a total of 302 plasma samples (144 Delta, 158 Omicron) used in this study. Of these, 109 were collected on day 1 (53 Delta, 56 Omicron), 103 on day 2 (52 Delta, 51 Omicron), and 90 on day 7 (39 Delta, 51 Omicron). Aliquots were shipped on dry ice to the Critical Care Epidemiologic and Biologic Tissues Resource (CCEPTR, a tissue bank) at the University of Calgary for management and further processing.

All samples were stored at −80°C immediately upon arrival at the University of Calgary. At the University of Calgary, the samples were thawed on ice, and 100 µL aliquots were re-labeled, refrozen at -80°C, and then shipped on dry ice for metabolite quantification. All sample management followed biosafety level 2 protocols.

For organic acid, amino acid and lipid quantifications, procedures are provided in the Methods section of the Supplementary Material .

A total of 143 metabolites were measured using LC-MS/MS and DI-MS/MS ( Supplementary Table S1 ).

For mass spectrometry analysis and data processing details, see the Methods section of the Supplementary Material .

Initially, a univariate analysis for each metabolomic and demographic variable was conducted. For continuous variables, the data were summarized using the mean and standard deviation. Group comparisons between two categories were performed using the Wilcoxon rank-sum test. For categorical variables, frequencies and percentages were presented, and group differences were assessed using either the chi-square or Fisher’s exact test, selected based on cell counts. Statistical significance was determined at a p-value of less than 0.05. Furthermore, a two-way analysis of variance (ANOVA) was employed to understand the main and interaction effects of the two factors.

Subsequently, multivariate analyses were conducted. Principal component analysis (PCA) was used to identify potential outliers and inherent data structures. For exploratory data visualization, a heatmap was created, and hierarchical clustering was carried out using Euclidean distance as the similarity metric and Ward’s linkage for cluster aggregation. To clarify group discrimination and identify key metabolites, partial least squares discriminant analysis (PLS-DA) was employed. The R2 and Q2 statistics were applied to evaluate the model’s explanatory and predictive powers, respectively. Finally, pathway analysis was performed using over-representation analysis. All statistical procedures were carried out using MetaboAnalyst 6.0.

A total of 109 patients were enrolled in this study, comprising 53 individuals infected with the Delta variant and 56 with the Omicron variant ( Table 1 ). The median age was 61.3 years for Delta and 59.3 years for Omicron, with no statistically significant difference (p=0.794) in age. Similarly, the sex distribution was alike between the two groups because they were matched by sex (66% male in Delta vs. 67.9% in Omicron; p=1.00). This match was purposefully designed to reduce confounding metabolomic variability, given the influence of age and sex on metabolites (Costanzo et al., 2022).

Hospital stay duration differed significantly: the Omicron group had a longer median stay (20 days; IQR = 13.25-31) than the Delta group (15 days; IQR = 9.00–22.00, p = 0.013). Although ICU admission rates were similar (47.2% Delta vs 50% Omicron, p = 0.921), NIH-based severity classification revealed that Delta patients had a greater proportion of severe and critical cases combined (71.7%) compared to Omicron (50.0%, p=0.133). Conversely, Omicron exhibited a higher rate of mild and moderate cases (50.0%) compared to Delta patients (28.3%, p=0.033). Complete vaccination (≥2 doses) was more common in Omicron cases (86.7%) than in Delta (58.1%, p = 0.145), whereas fewer than two doses were observed in 41.9% of Delta versus 13.3% of Omicron patients (p = 0.026). Corticosteroid usage rates were high overall (79.2% Delta vs. 66.1% Omicron, p=0.326), while tocilizumab administration differed significantly (37.7% in Delta vs. 7.1% in Omicron, p=0.001) ( Table 1 ).

Twenty-two metabolites were significantly altered (p<0.05) in the plasma of Delta-infected patients compared to those infected with Omicron. Among these, metabolites with increased median concentrations in Delta relative to Omicron included asparagine, asymmetric dimethylarginine, decanoylcarnitine (C10), propionylcarnitine (C3), butyrylcarnitine (C4), hydroxybutyrylcarnitine (C4OH), valerylcarnitine (C5), glutamine, isoleucine, leucine, lysophosphatidylcholine acyl (LysoPC a) C18:2, LysoPC a C20:3, LysoPC a C20:4, phosphatidylcholine diacyl (PC aa) C38:6, PC aa C40:6, threonine, and tyrosine. Conversely, aspartic acid, octadecenoylcarnitine (C18:1), homocysteine, and taurine exhibited decreased median concentrations in Delta relative to Omicron (see Supplementary Table S1.1 for day-specific fold changes). For an overview of all the metabolites significantly altered between variants, refer to Supplementary Figure S1A .

Figure 1 provides a visual representation of these metabolomic differences between Delta and Omicron-infected patient plasma metabolites. Panel A shows the boxplots of the most significantly altered metabolites between the Delta and Omicron variants on days 1, 2 and 7 of hospitalization, emphasizing the variant differences. Panel B is a principal component analysis (PCA) showing the metabolic clustering between the groups, in Delta and Omicron patient plasma. Overall, as expected, there is a great deal of overlap between the cohorts, but there are visible differences between them. There are also inherent differences within the cohorts, which are likely due to the inherent heterogeneity within the patient cohorts, including factors such as disease severity, comorbidities, and metabolic responses.

Pathway analysis incorporated all metabolites that showed a significant difference between the Delta and Omicron variants on days 1, 2 and 7 ( Supplementary Figure S2A ; Supplementary Table S2 ). The top pathways with the greatest impact scores related to these variant differences, in descending order, were phenylalanine, tyrosine, and tryptophan biosynthesis; taurine and hypotaurine metabolism; alanine, aspartate, and glutamate metabolism; tyrosine metabolism; and cysteine and methionine metabolism. Whereas, the most statistically significant pathways, listed in descending order, were valine, leucine, and isoleucine biosynthesis; arginine biosynthesis; alanine, aspartate, and glutamate metabolism; valine, leucine, and isoleucine degradation; and phenylalanine, tyrosine, and tryptophan biosynthesis pathways.

Thirty-two metabolites were significantly altered (p<0.05) in the plasma of the severe and critical group (SC) compared to the mild and moderate (mM) group. Metabolites with increased median concentrations in SC group included alanine, alpha-aminoadipic acid, asparagine, carnitine (C0), C3, C4, C5, creatine, fumaric acid, glutamine, homovallinic acid, kynurenine, lactic acid, leucine, lysine, LysoPC a C20:4, methionine, methionine-sulfoxide, ornithine, PC aa C32:2, phenylalanine, pyruvic acid, succinic acid, threonine, tryptophan, tyrosine, and valine. Conversely, acetyl-ornithine, homocysteine, and trans-hydroxyproline exhibited decreased median concentrations in the SC group compared to the mM group. For an overview of all metabolites significantly altered between severity groups, refer to Supplementary Figure S1B .

Figure 2 provides a comprehensive visualization of these severity-based metabolomic differences. Panel A displays boxplots of the most significantly altered metabolites between mild/moderate (mM) and severe/critical (SC) cases on days 1,2 and 7 of hospitalization, emphasizing their potential role in disease progression. Panel B presents a heatmap displaying the metabolic profiles across COVID-19 severity groups (critical, severe, moderate, mild), highlighting significant variations in metabolite levels associated with disease severity. Critical cases (marked in red) exhibit a unique metabolic signature compared to less severe cases, reflecting profound metabolic disruptions linked to critical illness. The hierarchical clustering analysis groups metabolites with similar expression patterns, revealing clusters of lipid metabolites (e.g., PC aa C36:0, LysoPC C18:0) that are predominantly altered in severe and critical cases. Additionally, amino acids and intermediates related to energy metabolism (e.g., kynurenine, methionine, tyrosine) are distinctly elevated in critically ill patients, indicating metabolic stress and dysregulation of amino acid metabolism. The clustering patterns emphasize co-regulated metabolites that may play critical roles in the pathophysiology of severe COVID-19 and serve as potential biomarkers for disease severity.

Figure 3 illustrates distinct metabolic signatures associated with SARS-CoV-2 variants (Delta and Omicron) and disease severity (mild to severe) over three time points (Days 1, 2 and 7). Critical cases exhibit substantial upregulation of lipids (e.g., LysoPC and SM species) and amino acids (e.g., lysine, methionine, tyrosine) compared to milder cases. Metabolite clustering reveals coordinated regulation patterns, highlighting potential biomarkers that differentiate variant types and disease severity, reflecting the dynamic metabolic alterations over time.

Furthermore, the metabolic interaction effects of severity (m+M and S+C) and SARS-CoV-2 variants (Delta and Omicron) were assessed. A comprehensive summary of the two-way ANOVA results for days 1, 2 and 7 are provided in Supplementary Table 3 . Fifteen metabolites demonstrated significant interaction effects, Pr (>F) 3, p <0.05, including asymmetric dimethylarginine, betaine, creatine, creatinine, fumaric acid, isoleucine, leucine, lysine, LysoPC C18:0, methylhistidine, PC aa C38:0, spermidine, spermine, total dimethylarginine, and valine ( Supplementary Figure S3 ).

Finally, pathway analysis included all significantly altered metabolites associated with COVID-19 severity between the Delta and Omicron variants on days 1, 2 and 7 ( Supplementary Figure S2B ; Supplementary Table S2 ). The pathways with the greatest impact on severity, in descending order, were phenylalanine, tyrosine, and tryptophan biosynthesis; alanine, aspartate, and glutamate metabolism; phenylalanine metabolism; cysteine and methionine metabolism; and tryptophan metabolism. In contrast, the most statistically significant pathways, also listed in descending order, were alanine, aspartate, and glutamate metabolism; arginine biosynthesis; valine, leucine, and isoleucine biosynthesis; citrate cycle (TCA cycle); and phenylalanine, tyrosine, and tryptophan biosynthesis.

A comprehensive summary of the metabolite profiles affected by corticosteroid treatment, categorized as non-corticosteroid use and corticosteroid use on days 1, 2 and 7 of hospitalization, is provided in Supplementary Figure 1C . Supplementary Figure S1C illustrates the significant median changes in metabolomic concentrations associated with corticosteroid use.

Twenty-six metabolites were significantly altered (p<0.05) in the plasma of the corticosteroid use group compared to the non-corticosteroid use group. Metabolites with increased median concentrations in the treated group included alanine, alpha-aminoadipidic acid, asparagine, beta-hydroxybutyric acid, C3, C4, C5, fumaric acid, glutamine, leucine, lysine, methionine, PC aa C32:2, phenylalanine, SM (OH)C14:1, SM(OH)C22:1, threonine, tryptophan, tyrosine, and valine. In contrast, acetyl-ornithine, asymmetric dimethylarginine, indole acetic acid, LysoPC a C18:0, and total dimethylarginine showed decreased median concentrations in the corticosteroid use group compared to the non-corticosteroid use group.

In Figure 4B , a heatmap illustrates the hierarchical clustering of metabolites based on their abundance in COVID-19 patients receiving corticosteroids and those who did not receive corticosteroids at three time points (Days 1, 2 and 7) of the study for both Delta and Omicron variants. The color gradient ranges from blue to red, indicating low and high metabolite abundance, respectively. Clustering patterns reveal metabolic differences linked to corticosteroid treatment and variant type. Patients who received corticosteroids show a marked increase in certain metabolites, such as lipids (e.g., LysoPC and SM species) and amino acids (e.g., lysine and methionine), compared to patients who did not receive corticosteroids, implying a significant metabolic effect of corticosteroid therapy. Furthermore, alterations in metabolic pathways influenced by treatment and variant type result in clusters of coordinated metabolites. These findings suggest that corticosteroid administration may elicit different metabolic responses in SARS-CoV-2 variants and could serve as potential therapeutic and prognostic biomarkers.

Moreover, the metabolic interaction effects of corticosteroid use (non-corticosteroid use and corticosteroid use) and SARS-CoV-2 variant (Delta and Omicron) were evaluated. A comprehensive summary of the two-way ANOVA results from days 1, 2 and 7 is provided in Supplementary Table 4 . Fourteen metabolites demonstrated significant interaction effects, Pr (>F) 3, p-value <0.05 which included: C0, C4, glycine, homocysteine, indole acetic acid, kynurenine, LysoPC a C14:0, LysoPC a C18:1, PC aaC36:0, PC aa C40:1, phenylalanine, hydroxyhippuric acid, hydroxysphingomyeline C24:1, and taurine ( Supplementary Figure S4 ).

Pathway analysis incorporates all significant metabolites associated with corticosteroid treatment, comparing Delta and Omicron variants on days 1, 2 and 7 ( Supplementary Figure S2C ; Supplementary Table S2 ). The pathways exhibiting the greatest impact, in descending order, included phenylalanine, tyrosine, and tryptophan biosynthesis; phenylalanine metabolism; tyrosine metabolism; alanine, aspartate, and glutamate metabolism; and tryptophan metabolism. In contrast, the most statistically significant pathways, also listed in descending order, were alanine, aspartate, and glutamate metabolism; arginine biosynthesis; valine, leucine, and isoleucine biosynthesis; phenylalanine, tyrosine, and tryptophan biosynthesis; and phenylalanine metabolism.

A comprehensive summary of the metabolite profiles affected by vaccination, categorized as incomplete and complete vaccination, on days 1, 2 and 7 of hospitalization, is provided in Supplementary Figure 1D . Supplementary Figure S1D illustrates the significant median changes in metabolomic concentrations associated with vaccination.

Twenty-three metabolites were significantly altered (p<0.05) in the plasma of the completely vaccinated group compared to the incompletely vaccinated group. Metabolites with increased median concentrations in the completely vaccinated group included acetyl-ornithine, C10:1, C4OH, citrulline, glycine, isobutyric acid, propionic acid, taurine, hydroxyproline and uric acid. Conversely, alpha-aminoadipic acid, ketoglutaric acid, asparagine, C3, C4, lysine, LysoPC a C26:1, methionine, PC aa C36:6, proline, pyruvic acid, SM C20:2, and threonine showed decreased median concentrations in the completely vaccinated group compared to the incompletely vaccinated group.

Figure 5 provides a comprehensive visualization of the metabolomic differences between vaccination groups. Figure 5A displays boxplots of the most significantly altered metabolites between completely vaccinated and incompletely vaccinated cases on days 1, 2 and 7 of hospitalization. Figure 5B depicts a heatmap of the top 40 metabolites, illustrating clustering patterns based on vaccination.

Furthermore, the metabolic interaction effects of vaccination status (complete vaccination and incomplete vaccination) and SARS-CoV-2 variant (Delta and Omicron) were evaluated. A comprehensive summary of the two-way ANOVA results from days 1, 2 and 7 are provided in Supplementary Table 5 . Three metabolites demonstrated significant interaction effects, Pr (>F) 3, p-value <0.05, which included: taurine, C7DC, and C5DC ( Supplementary Figure S5 ).

Finally, pathway analysis incorporates all significant metabolites associated with vaccination status, comparing Delta and Omicron variants on days 1, 2 and 7 ( Supplementary Figure S2D , see Supplementary Table S2 ). The pathways exhibiting the greatest impact, in descending order, included glycine, serine, and threonine metabolism; arginine biosynthesis; alanine, aspartate and glutamate metabolism; arginine and proline metabolism; and lipoic acid metabolism. In contrast, the most statistically significant pathways, also listed in descending order, were arginine biosynthesis; alanine, aspartate, and glutamate metabolism; lipoic acid metabolism; glycine, serine and threonine metabolism; and arginine and proline metabolism.