Metabolomic analyses of COVID-19 patients unravel stage-dependent and prognostic biomarkers

Abstract

The circulating metabolome provides a snapshot of the physiological state of the organism responding to pathogenic challenges. Here we report alterations in the plasma metabolome reflecting the clinical presentation of COVID-19 patients with mild (ambulatory) diseases, moderate disease (radiologically confirmed pneumonitis, hospitalization and oxygen therapy), and critical disease (in intensive care). This analysis revealed major disease- and stage-associated shifts in the metabolome, meaning that at least 77 metabolites including amino acids, lipids, polyamines and sugars, as well as their derivatives, were altered in critical COVID-19 patient's plasma as compared to mild COVID-19 patients. Among a uniformly moderate cohort of patients who received tocilizumab, only 10 metabolites were different among individuals with a favorable evolution as compared to those who required transfer into the intensive care unit. The elevation of one single metabolite, anthranilic acid, had a poor prognostic value, correlating with the maintenance of high interleukin-10 and -18 levels. Given that products of the kynurenine pathway including anthranilic acid have immunosuppressive properties, we speculate on the therapeutic utility to inhibit the rate-limiting enzymes of this pathway including indoleamine 2,3-dioxygenase and tryptophan 2,3-dioxygenase.

Fig. 1. Profound metabolomics alterations associated with COVID-19 clinical severity.

A Heatmap illustrating the changes in metabolite abundance in the plasma from control (n = 27), mild (n = 23), moderate (n = 21), and critical (n = 28) COVID-19 patients. Significant metabolites were identified by Wilcoxon rank-sum test and the false discovery rate (FDR) controlled with Benjamini–Hochberg procedure between patients with critical and mild COVID-19. Hierarchical clustering (Euclidean distance, ward linkage method) of the metabolite abundance is shown. PCaes, total abundance of the different phosphatidylcholines identified in the cohort plasma samples. B, Random forest classification model was built using main metabolites altered (p < 0.05) between critical and mild COVID-19 patients as a predicting tool. The variables importance (as the mean decrease of the Gini index) for building the model is reported in a dot plot, with dots substituted by an up-pointing triangle to indicate metabolites increased in critical vs mild COVID-19 patients, and by a down-pointing triangle in the opposite case (B), the confusion matrix (indicating model accuracy) is depicted below. OOB out-of-bag error.

Fig. 2. Effects of COVID-19 on circulating sugars and amino acids.

Modified carbohydrates (A) and amino acids (B) were profoundly altered in patients with the most severe COVID-19. Data in A and B were analyzed by non-parametric unpaired Wilcoxon test (Mann–Whitney) for each two-group comparison. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Fig. 3. Effecs of COVID-19 on polyamines, tryptophan derivatives and selected amino acids.

Polyamines, arginine (A) and tryptophane (B) pathways alterations in critically ill COVID-19 patients were representative of an immunosuppressive metabolomic state. Data in A and B were analyzed by non-parametric unpaired Wilcoxon test (Mann–Whitney) for each two-group comparison. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Fig. 4. Patients with unfavorable clinical evolution after tocilizumab, infused for worsening pulmonary involvement of COVID-19, had pre-treatment metabolomics differences compared to patients with favorable outcome.

A Heatmap illustrating pre-tocilizumab metabolite abundance in COVID-19 patients evaluable for clinical evolution after treatment (n = 21). Significant metabolites were identified by Wilcoxon rank-sum test between patients with favorable and unfavorable evolution after tocilizumab infusion. BMI Body Mass Index, WHO World Health Organization, O2 oxygenotherapy, ICU intensive care unit, OTI orotracheal intubation. B Principal component analysis biplot, showing the contribution of the most significant metabolites (p < 0.05) to the discrimination (PC1 45.9%) between patients with favorable and unfavorable evolution after tocilizumab infusion. C Random forest classification model was built using main metabolites altered (p < 0.05) in baseline samples from COVID-19 patients with favorable and unfavorable evolution after tocilizumab treatment as a predicting tool. The variable importance (as the mean decrease of the Gini index) for building the model is reported in a dot plot, with dots substituted by an up-pointing black triangle to indicate metabolites increased in patients who showed unfavorable vs favorable evolution, and by a gray down-pointing triangle in the opposite case. The confusion matrix (indicating model accuracy) is depicted. OOB out-of-bag error.

Fig. 5. COVID-19 Patients with unfavorable outcome after tocilizumab infusion did not improved lymphopenia, inflammatory, immunosuppressive, and metabolomic abnormalities instead of patients who evolved towards clinical improvement.

Patients with paired baseline and post treatment (day 7 ± 3) serum samples are represented (n = 18). The measured parameters include total lymphocyte counts (A) as well as the concentrations of IL18 (B), IL10 (C) and tryptophan (D). Wilcoxon signed-rank test was used to compare paired baseline and post treatment measures and Wilcoxon rank-sum test to compare baseline or day 7 (±3) measures between patients with response or no response. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Fig. 6. Correlations between cytokines and metabolites before tocilizumab infusion in patients with worsening COVID-19 highlighted that dysregulated metabolomic and immunologic pathways were closely related to clinical worsening of patients developing critical COVID-19.

A Correlation between cytokines and most significant metabolites at baseline was analyzed by Pearson correlation. *p < 0.05, **p < 0.01, ***p < 0.001. B Pearson correlations between IL10 and IL18 with anthranilic acid serum levels.