Nucleotide, Phospholipid, and Kynurenine Metabolites Are Robustly Associated with COVID-19 Severity and Time of Plasma Sample Collection in a Prospective Cohort Study

Abstract

Understanding the molecular underpinnings of disease severity and progression in human studies is necessary to develop metabolism-related preventative strategies for severe COVID-19. Metabolites and metabolic pathways that predispose individuals to severe disease are not well understood. In this study, we generated comprehensive plasma metabolomic profiles in >550 patients from the Longitudinal EMR and Omics COVID-19 Cohort. Samples were collected before (n = 441), during (n = 86), and after (n = 82) COVID-19 diagnosis, representing 555 distinct patients, most of which had single timepoints. Regression models adjusted for demographics, risk factors, and comorbidities, were used to determine metabolites associated with predisposition to and/or persistent effects of COVID-19 severity, and metabolite changes that were transient/lingering over the disease course. Sphingolipids/phospholipids were negatively associated with severity and exhibited lingering elevations after disease, while modified nucleotides were positively associated with severity and had lingering decreases after disease. Cytidine and uridine metabolites, which were positively and negatively associated with COVID-19 severity, respectively, were acutely elevated, reflecting the particular importance of pyrimidine metabolism in active COVID-19. This is the first large metabolomics study using COVID-19 plasma samples before, during, and/or after disease. Our results lay the groundwork for identifying putative biomarkers and preventive strategies for severe COVID-19.

Keywords: COVID-19 severity; biomarkers; longitudinal cohort; phospholipid metabolism; prospective sampling; pyrimidine metabolism; regression analysis; tryptophan metabolism; untargeted metabolomics.

Figure 1

Study design schematic. Metabolomic profiles of plasma samples drawn from 555 COVID-19 patients were analyzed. The majority of samples reflect pre-COVID-19 timepoints (441), while 86 were drawn during and 82 post-COVID-19 diagnosis (only 5 patients had samples available at all three timepoints, see Supplementary Figure S1). Severity levels ranged from 0 to 3, according to WHO guidelines. Statistically significant metabolites in each analysis were identified as those with FDR-adjusted p-values < 0.05. Metabolite associations with COVID-19 severity were determined using Equation (1), stratified by each time of sample collection. Sets of interest are described in the table, where checkmarks indicate that metabolites must be significant and share directionality of association for that comparison to be included in the same set. Metabolite associations with time of sample collection, Equation (2), were determined using all time points. Checkmarks for these associations indicate significant metabolites in each comparison.

Figure 2

Metabolite classes measured and associated with COVID-19 severity. (A) Representation of total metabolite classes detected by the Metabolon comprehensive metabolomics platform used in regression modeling (lipids = 447, unknown = 264, amino acids = 225, cofactors and vitamins = 39, nucleotides = 39, peptides = 32, partially characterized molecules = 27, carbohydrates = 25, energy = 10). (B) Metabolite classes for metabolites significantly associated with severity (FDR-adjusted p-value < 0.05; for the severity term, see Equation (1)) that fall into our three metabolite sets of interest: (1) predisposition metabolites, identified from pre-COVID-19 samples only; (2) predisposition-and-acute metabolites, defined as those significant and with the same directionality of association with severity in pre- and during COVID-19 samples; and (3) persistently affected metabolites, defined as those significant and with the same directionality of association with severity in pre-, during, and post-COVID-19 samples.

Figure 3

Metabolites significantly associated with COVID-19 severity associations that map to biological pathways. Each panel represents a pathway cluster (group of similar pathways). Log odds of significant associations between metabolites and COVID-19 severity (Equation (1)). Sets of interest include predisposition metabolites (only significantly associated with severity (FDR-adjusted p-values < 0.05) in pre-COVID-19 samples (n = 441)) (see Supplementary File S3); predisposition-and-acute metabolites (significantly associated with severity (FDR-adjusted p-values < 0.05) in pre- (n = 441) and during (n = 86) COVID-19 samples) (see Supplementary File S4); and persistent metabolites (significantly associated with severity (FDR-adjusted p-values < 0.05) in all samples (n = 441, 86, and 82 for pre-, during, and post-COVID-19 samples, respectively) (see Supplementary File S5). Metabolites can appear in multiple pathway clusters because some metabolites map to multiple pathways. * Identities of these metabolites are confidently assigned by Metabolon but without authentic standards.

Figure 4

Metabolite abundance changes over time in samples collected prior to, during, and after COVID-19 for metabolites that map to biological pathways. Change in effect sizes (coefficient of the time term of Equation (2) where the change is relative to the latter group of the compared pairs) of metabolites that reflect (A) transient changes during and (B) lingering changes after COVID-19 infection. For (A), metabolites were significantly associated with time in pre- (n = 441) vs. during (n = 86) COVID-19 and during vs. post-COVID-19 (n = 82) (FDR-adjusted p-value < 0.05) but not pre- vs. post-COVID-19 comparisons, and mapped to pathways using RaMP-DB. For (B), metabolites were significantly associated with time in pre- vs. post-COVID-19 (FDR-adjusted p-value < 0.05) comparisons and mapped to pathways using RaMP-DB. See also Supplementary Files S6 and S7 for FDR-adjusted p-values.