Method development in metabolomics and its application as diagnostic tool

Metabolomics is known as the comprehensive study of metabolites in biological systems. Furthermore, the metabolome is defined as the complete set of low-molecular-weight biomolecules that provides biologically relevant endpoints of metabolic processes. Metabolomics is a powerful tool in disease research and biomarker discovery. These biomarkers are measurable biological read-outs that can be used to diagnose, monitor or predict risk of diseases. Different analytical techniques are available to study biomarkers. Nowadays, the most established technique is RPLC coupled with high resolution MS. An upcoming technique in this domain is CE coupled with high resolution MS, because of the increased interest in low volume samples, which are less invasive and open possibilities for self-sampling. Still, interpretation of the data remains challenging and therefore requires suitable visualization, chemometric and bioinformatic methods.
In a first part of this thesis, the suitability of CE-MS for metabolomics was investigated.For this purpose, a simulated comparative metabolic profiling study is pe oled human plasma samples spiked with a different concentration of isotope-labeled metabolites, generating two groups, classes I and II. Multivariate data analysis of the recorded metabolic profiles provided a clear distinction between classes I and II. The separation between the classes was mainly attributed to the difference in isotope-labeled compounds, emphasizing the possibilities of CE-MS in comparative metabolomics studies. Furthermore, this method was applied on volume-limited mice plasma samples to discover metabolic consequences of evoked seizures. Increased knowledge about global metabolic hallmarks related to a seizure leads to an improved understanding of seizure pathophysiology and can reveal novel pathways to control and prevent seizures. Seizures were induced in the 6Hz corneal seizure acute mice model. Similarly decreased levels of metabolites (glycine, citrulline, isoleucine, serine, proline, tryptophan, arginine, alanine, valine, asparagine and histidine), as previously observed by other research groups using other seizure models, were observed. However, for monoamines and their precursors contradictory results compared to the literature were obtained. Therefore, it would be useful to further investigate and validate their role in different animal models (electrical and chemoconvulsive) as well as in humans, using different analytical techniques. In general, the suitability of CE-MS, including sampling and sample preparation, for the analysis of volume-restricted plasma samples with acceptable precision was demonstrated. Most metabolic markers that were identified by the CE-MS study are amino acids. This could be attributed to the employed CE-MS separation conditions, which are specifically suited for the profiling of basic metabolites, including compound classes, such as amino acids, amines and nucleosides. It would be of interest to investigate the samples also with a complementary technique, such as RPLC-MS to profile less polar compounds.
In the second part of this thesis, an RPLC-MS method was developed for semi-polar metabolites by means of a screening-design approach. Considering the detected number of peaks, the following factors were found to influence the response, i.e. the scan time, low collision energy and desolvation temperature. However, several difficulties were encountered during this method development, such as variation in the chromatographic profiles for blanks, diluted plasma samples and QC samples. A possible cause of this variation could be the presence of (a) contaminant(s). After performing a thorough cleaning, an adapted screening design was executed. These adaptations included the use of additional blanks in the experimental sequence and of more nominal-level experiments
among the design experiment. Still, a large variation in the chromatographic profiles was observed resulting in unclear results. Therefore, it was concluded that in untargeted metabolomics it is more appropriate to mimick a metabolomic study by spiking known endogenous compounds in the matrix instead of taking the total number of peaks as response. This latter procedure was also executed for method validation considering linearity, matrix effects, accuracy and precision. Acceptable repeatability was observed for all eleven spiked compounds at the higher concentration levels and in the QC samples. As different plasma concentrations ranges are observerd for the eleven endogenous metabolites (nicotinic acid, kynurenic acid, arginine, creatinine, methionine, cholic acid,
hippuric acid, pyroglutamic acid, phenylalanine, tryptophan and caffeine) in different batches, it would be of interest to repeat the validation experiments with narrower concentration intervals around the basal levels. This may result in better linearity results. Intermediate precision evaluation showed a similar trend of increased variation at lower concentration levels. Furthermore, accuracy evaluation did not provide appropriate results for all concentration levels. In untargeted metabolomics it is already acceptable when a method shows some linearity (increasing signal with concentration) and precision. Further evaluation of the precipitation solvent and reconstitution solvent could be made because
of their major influence on recovery. Additionally, the inclusion of multiple isotope-labeled internal standards (IS) needs to be evaluated instead of the consideration of only one IS.
The third part of this thesis concerns the development of a diagnostic tool, by means of SIFT-MS, to discriminate asthmatic and cystic fibrosis patients from healthy children. SIFTMS is of interest as it allows analyzing exhaled breath samples in real-time. Breath sampling is non-invasive and its analysis may thus be a good alternative for sputum analysis and bronchoscopy, two other approaches often applied in lung diseases evaluation. The focus of this study was on different data analysis approaches, including a variety of pretreatment, classification and discrimination techniques to show the suitability of full-scan SIFT-MS analysis in a clinical context. Furthermore, the knowledge gathered may be of interest to the wider scientific community because data pretreatment and finding good modelling techniques is labor intensive work. Perfect models were obtained after DOSC as pretreatment. However, further investigation has shown that perfect classification of random classes is also obtained after DOSC, which makes this data pretreatment method suspicious and unreliable. Other promising classification models were obtained, but their prediction capacity, demonstrated by cross-validation results, was less optimistic. Good models were obtained after pareto scaling and a more extensive data pretreatment approach, including single value imputation, normalization, autoscaling and log transformation. Both allowed good PCA-QDA analysis models and the latter data pretreatment also resulted in promising PLS-DA analysis models. A future requirement is the collection of an extended data set, allowing a proper external validation. In general, the potential of the untargeted application of SIFT-MS spectra as rapid pattern-recognition tool is shown.
In summary, it is shown in this dissertation that the application of metabolomics is possible in different fields. Different analytical techniques can be applied, each with their strengths and weaknesses. The most important strength is their complementarity allowing to reveal more information from a given sample.