Title NMR-based metabolic profiling provides diagnostic and prognostic information in critically ill children with suspected infection

Abstract

Sepsis, defined as life-threatening organ dysfunction caused by infection is difficult to distinguish clinically from infection or post-operative inflammation. We hypothesized that in a heterogeneous group of critically ill children, there would be different metabolic profiles between post-operative inflammation, bacterial and viral infection and infection with or without organ dysfunction. 1D ¹H nuclear magnetic resonance spectra were acquired in plasma samples from critically ill children. We included children with bacterial (n = 25) and viral infection (n = 30) and controls (n = 58) (elective cardiac surgery without infection). Principal component analysis was used for data exploration and partial least squares discriminant analysis models for the differences between groups. Area under receiver operating characteristic curve (AUC) values were used to evaluate the models. Univariate analysis demonstrated differences between controls and bacterial and viral infection. There was excellent discrimination between bacterial and control (AUC = 0.94), and viral and control (AUC = 0.83), with slightly more modest discrimination between bacterial and viral (AUC = 0.78). There was modest discrimination (AUC = 0.73) between sepsis with organ dysfunction and infection with no organ dysfunction. In critically ill children, NMR metabolomics differentiates well between those with a post-operative inflammation but no infection, and those with infection (bacterial and viral), and between sepsis and infection.

Figure 1

Metabolite abundance comparison. Each comparison is shown in column (C control, DV definite viral, DB definite bacterial). The numbers and colours denote log ratio of group means in statistically significant comparisons only (t-test, alpha = 0.05). Red means increased in DB:C or DV:C, purple lower in DB. DV relative to C or lower in DB relative to DV.

Figure 2

Sparse partial least squares discriminant analysis (a) scores plot of model including DB (N = 25) and C (N = 58); (b) scores plot of model including DV (30) and C (N = 58); the most contributing metabolite signals (VIP scores); (c) C and DB; (d) C and DV. Associated AUC = 0.94 in DB-C and 0.83 in DV-C. DB—definite bacterial, DV – definite viral, C control, VIP variable importance score.

Figure 3

PLS-DA (a) scores plot of model including DB (N = 25) and DV (N = 30) samples. (b) Most contributing metabolite signals (VIP scores). Associated AUC = 0.78.

Figure 4

Comparison of metabolic profiles for MOD, no MOD and control cohorts. (a) Metabolite abundance comparison. Each comparison is shown in column (C—control, Yes—multiple organ dysfunction, No—no multiple organ dysfunction). The numbers and colours denote log ratio of group means in statistically significant comparisons only (t test, alpha = 0.05). Mean log ratios of MOD-positive and MOD negative group metabolic profiles with controls.

Figure 5

Metabolite profiles were investigated by pairwise comparisons of cohorts with and without MOD and fitting a PLS-DA model for selection of discriminating metabolites. The results show (a) scores plot (1 component) of model including MOD positive patients with infection (blue) (N = 25) vs MOD negative patients with infection (red) (N = 30). (b) Metabolite importance (VIP) plot. Associated AUC = 0.73.