Temporal metabolic response yields a dynamic biosignature of inflammation

Abstract

Chronic low-grade inflammation is a subclinical condition directly and indirectly linked to the development of a wide range of diseases responsible for the vast majority of morbidity. To examine mechanisms coupled to chronic disease, a group of overweight and obese human subjects without known inflammatory diseases participated in a high-fat meal challenge as an acute inflammation stimulus. Analysis of serum metabolites grouped by baseline cytokine levels revealed that single samples had little power in differentiating groups. However, an analysis that incorporated temporal response separated inflammatory response phenotypes and allowed us to create a metabolic signature of inflammation which revealed metabolic components that are crucial to a cytokine-mediated inflammation response. The use of temporal response, rather than a single time point, improved metabolomic prediction of high postprandial inflammation responses and led to the development of a dynamic biosignature as a potential tool for stratifying risk to a wide range of diseases.

Keywords: Metabolomics; Pathophysiology; Systems biology.

Graphical abstract

Figure 1

Grouping of participants by cytokine response (A) K-means analysis of the cytokine concentrations for participants in the cohort. The “elbow” of the optimal clusters graph shows that three groups is the best way to bin the subjects based on cytokine response. (B) Plots for pro-inflammatory cytokines used in the study at each time point from fasting to four hours after meal. Error bars are included and indicate 95% confidence intervals. See also Table S1.

Figure 2

Static metabolic profiling of high and low inflammation groups (A) A principal component analysis (PCA) of response groups at one hour after meal shows little to no separation. (B) Heatmap for time point two showing the top 25 discriminating features. The dendrogram at the top of the figure shows little clustering of the inflammation groups. See also Figure S2.

Figure 3

Feature interaction based on time and inflammation response (A) Important features for the interaction between response groups over the time course found by combining SPE scores and leverage. (B) Cp plot of well-modeled features in a linear regression model predicting cytokine response. Columns represent mass features and rows show BIC values where the best model is on the top row.

Figure 4

Composite model Plots of the five metabolites found in the composite model including error bars indicating 95% confidence intervals. Relative concentrations are shown for the entirety of the time course. Creatine, hydroxymethyluracil, and histidinylhistidine represent the three most common metabolites, found in three cytokine models and the composite model while carbamic acid and the unknown metabolite were found in one model each and the composite model. Note the slope change between the high and low response groups at one hour after meal.

Figure 5

Six of the metabolites from the cytokine models Each of these metabolites were found in specific cytokine models but were not included in the composite model. 4-Guanidinobutyrate, butyrylglycine, and threitol were part of two cytokine models. The other metabolites were found in one model each. Relative concentrations are shown for the entirety of the time course with error bars indicating 95% confidence intervals.