Table 1.
Clinical Metadata of the Studied Patient Population.
Fig 1.
PLS-DA analysis for G0 versus G4 serum samples.
(A) Three-dimensional PLS-DA score plot. Red triangles: G0 samples (fibrosis stage F0). Green crosses: G4 samples (fibrosis stage F4). (B) Two-dimensional PLS-DA score plot. Red circles: G0 samples. Green circles: G4 samples. (C) PLS-DA classification using different numbers of components. The red asterisk indicates the best classifier. The inset table summarizes Q2, R2 and accuracy of the best model. Comps means number of components. (D) Permutation test statistics for 1000 permutations with observed statistic at p < 0.01.
Fig 2.
Significant features (binned signals) discriminating between G0 and G4 serum samples.
(A) Important features identified by PLS-DA and VIP scores. The colored boxes on the right indicate relative bin integrals for G0 and G4 samples. Variable Importance in Projection is a weighted sum of squares of the PLS loadings taking into account the amount of explained Y-variation in each dimension. (B) Heatmap of unsupervised hierarchical clustering (distance measure using Pearson and clustering algorithm using Ward). The heatmap was constructed from the most significantly differing bins (features), as identified by PLS-DA and VIP scores. Only the top 25 features are shown. Each colored cell on the map corresponds to a relative concentration value, with samples in rows (S2 Table indicates the original name of every sample) and features/compounds in columns. Red and blue colors denote increased and decreased bin integrals, respectively.
Fig 3.
Representative T2-filtered 1H- NMR spectrum of human serum sample measured at 300 K, 600 MHz.
(A) Full spectrum (0–10 ppm), (B) 35× zoom on the aromatic signal region (5.5–8.5 ppm). Signal assignments were derived by consulting the NMR metabolic profiling database (HMDB), literature references, or from NMR experiments on the pure compounds added to an average serum sample. Spectra were referenced internally against the TSP signal (δ = 0.00 ppm). Assignment numbers correspond to identified metabolites as follows: 1, citric acid; 2, cysteine; 3, lactic acid; 4, glutamine; 5, glutamate; 6, isoleucine; 7, valine; 8, leucine; 9, alanine; 10, 3-hydroxybutyrate; 11, lysine; 12, arginine; 13, Nac1 (N-acetyl of glycoproteins); 14, Nac2; 15, choline; 16, creatine; 17, creatinine, 18, glycerol; 19, TMAO (trimethylamine N-oxide); 20, isobutyric acid; 21, VLDL1 (very low density lipoproteins); 22, acetoacetate; 23, VLDL2; 24, LDL2 (low density lipoproteins); 25, Lipid; 26, GPC (glycerophosphocholine); 27, glucose β-H2; 28, glucose/sugars; 29, α-glucose; 30, lipids; 31, urea; 32, fumaric acid; 33, tyrosine; 34, histidine; 35, phenylalanine; 36, hippuric acid; 37, formic acid.
Fig 4.
Boxplot of relative concentrations for some significantly altered metabolites (p < 0.05) in serum of G4 (green) and G0 (red) patients.
Y axes are represented as relative units. Data were normalized to the total spectral area. Due to this normalization process we obtained negative scale in the Y-axis in some of the bins (Metaboanalyst program analysis). The bar plots show the normalized values (mean +/- one standard deviation). The boxes range from the 25% and the 75% percentiles; the 5% and 95% percentiles are indicated as error bars; single data points are indicated by circles. Medians are indicated by horizontal lines within each box.
Table 2.
Most important metabolites obtained from the PLS-VIP, Wilcoxon Mann Whitney test and ROC analysis.
Fig 5.
Individual receiver operating characteristic (ROC) curves.
The colored curves represent the bins at 0.83 ppm (LDL1), 3.20 ppm (choline) and 2.30 ppm (acetoacetate). The black curve represents our multivariable predictive model described by the linear combination α*2.30 + β*0.83 + γ*3.20 that reaches a cut-off value of -0.316, specificity of 80%, sensitivity of 70%, AUROC score of 0.922, and confidence interval of 95% (0.85 to 0.97).