Table 1.
LC-MS datasets details.
Fig 1.
Diagram representing the study workflow.
A. Peaks were detected from input LC-MS data (mzML format), including the internal standards and serum samples, using MZmine2. Peakset lists containing ion information (m/z, RT, intensity) were obtained. B. Internal standards analysis. A reference dataset was selected and the RT drift in the other datasets was calculated and modelled using GP regression. C. Sample analysis. Based on the GPR models obtained for each dataset, the RT was corrected in each peakset list and alignment was done using MZmine2. Afterwards, statistical analysis focused on the intensity differences between the control and infected samples was performed using limma R package. This was followed by annotation and pathway analysis using mummichog.
Fig 2.
Modelling the RT drift in the two datasets DM and DVL.
A. Modelling the drift in DM using an RBF kernel. Graph (a): Mean and 95% posterior confidence with optimised hyperparameters: RBF variance = 0.07, RBF lengthscale = 6.84. Graph (b) illustrates the RT drift in DM before and after correction of the retention times using the GPR model. B. Modelling the drift in DVL using a composite RBF + MLP kernel. Graph (a): Mean and 95% posterior confidence with optimised hyperparameters: MLP variance = 3.99, MLP weight variance = 2.02e7, MLP bias variance = 5.56e-309, RBF variance = 7.44, RBF lengthscale = 9.41. Graph (b) illustrates the RT drift in DVL before and after correction of the retention times using the GPR model.
Fig 3.
Boxplots of putatively annotated compounds which were upregulated in infected patients.
Overview of putatively annotated metabolites which are statistically significant (p-val<0.05) and present a general upward trend in all three datasets, i.e higher intensities in infected patients. The metabolites are grouped based on their class or subclass according to HMDB. In this case, metabolites from amino acids and derivatives, lipids and derivatives, sugars and pyrimidines were identified. Values from both positive and negative ESI modes are presented from left to right in ascending order of their p-value. Metabolites in italic font are only annotated using mummichog.
Fig 4.
Boxplots of putatively annotated compounds which were downregulated in infected patients.
Overview of putatively annotated metabolites which are statistically significant (p-val<0.05) and present a general downward trend in all three datasets, i.e. lower intensities in infected patients (with p-value<0.05). The metabolites are grouped based on their class or subclass according to HMDB. In this case, metabolites from amino acids and derivatives, carboxylic acids and indoles were identified. Values from both positive and negative ESI modes are presented from left to right in descending order of their p-value in each group. Metabolites in italic font are only annotated using mummichog.
Fig 5.
Boxplots of putatively annotated compounds which were downregulated in infected patients (continuation).
Overview of putatively annotated metabolites which are statistically significant (p-val<0.05) and present a general downward trend in all three datasets, i.e. lower intensities in infected patients (with p-value<0.05). The metabolites are grouped based on their class or subclass according to HMDB. In this case, metabolites from lipids and derivatives and nucleotides and derivatives were identified. Values from both positive and negative ESI modes are presented from left to right in descending order of their p-value in each group. Metabolites in italic font are only annotated using mummichog.
Table 2.
Significantly altered metabolic pathways (p-val<0.05) following mummichog analysis of the negative and positive ionisation mode data.
Fig 6.
Tryptophan metabolism and the changing metabolites from each dataset.
Intensity values are represented as lg2 values. The metabolites were mapped against the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway map hsa00380. Metabolites in italic font are annotated following mummichog analysis or the HMDB matching method, while the rest are annotated using the internal standard metabolites information. The boxplots represent the intensities of all the samples in each condition (red = infected, blue = control) in all three datasets.
Fig 7.
Peakset annotated as Kynurenine (M+H[1+]).
using internal standards matching method, mummichog and HMDB matching method. The spectrum belongs to DVL (resolver obtained from ms2lda.org) and it was matched against experimental LC-MS MS2 information from MassBank compound KO003269 with a cosine similarity score of 0.54 (fragment tolerance = 0.2) (Metabolomics spectrum resolver plot).
Fig 8.
Peak annotated as L-Tryptophan fragment with loss of ammonia (M-NH3+H[1+]).
using internal standards matching method, mummichog and HMDB matching method. The loss of ammonia from protonated tryptophan was observed as the primary fragmentation pathway in gas-phase reactions [28]. The spectrum belongs to DZ and it was matched against experimental LC-MS MS2 information from MassBank BML01191 compound with a cosine similarity score of 0.55 (fragment tolerance = 0.2) (Metabolomics spectrum resolver plot).