Metabolomics for Biomedical Research

Metabolomics for Biomedical Research brings together recent progress on study design, analytics, biostatistics and bioinformatics for the success of metabolomics research. Metabolomics represents a very interdisciplinary research prominent in the functional analyses of living systems; hence, this book focuses on translation and medical aspects. The book discusses topics such as biomarkers and their requirements to be used in medical research, with the parameters and approaches on how to validate their quality; and animal models and other approaches, as stem cells and organoid culture. Additionally, it explains how metabolomics may be applied in prediction of individual response to drug or disease progression.

This book is a valuable source for researchers on systems biology and other members of biomedical field interested in metabolism-oriented studies for medical research.

• Focuses on metabolomics in translational and medical research
• Provides basics for, and concepts of, contemporary translational personalized medicine research with metabolomics
• Brings the major recent progresses on design, analytics, biostatistics and bioinformatics relating to the success of metabolomics research

• Language: English
• Publisher: Academic Press
• Release date: Mar 20, 2020
• ISBN: 9780128127858

Book preview

Germany

Preface

Jerzy Adamski, Helmholtz Zentrum München, Research Unit Molecular, Endocrinology and Metabolism, Neuherberg, Germany

Metabolomics represents a very interdisciplinary research area, which is prominent in the functional analyses of living systems and environment. This book focuses on metabolomics applied to translational and medical research and provides basics and concepts of contemporary research. There was major progress in many disciplines (including study design, analytics, biostatistics, and bioinformatics) critical for the success of metabolomics. However, the knowledge in different disciplines is dispersed and there is a lack of unified resources. This book aims to bring these disciplines together and provides information on:

•Strategies in study design, sample randomization, contingency plan, and quality assurance.

•Description of preanalytical procedures assuring sample quality by proper collection, storage, and compliance to standardization.

•Analysis of the impact of confounders on metabolic phenotypes.

•Survey of mass spectrometry-based analytical methods applied to metabolomics.

•Overview of applications in animal models and cell culture.

•The use of metabolomics in drug development.

•Procedures and requirements in biomarker discovery.

•Assessment of the impact of the genome on the metabolome.

•Comparison of different bioinformatic strategies for metabolomics.

•Analyses of the impact of metabolomics in personalized medicine.

The content of this book is designed for researchers interested in metabolism-oriented studies for medical research. The book will address geneticists, biochemists, cell biologists, physiologists, epidemiologists, clinical, translational, and drug discovery researchers. The content is addressed to researchers planning, performing and interpreting experiments, and furthermore to management in foundations, academia, and pharma research.

Chapter 1

Introduction to metabolomics

Jerzy Adamski    Helmholtz Zentrum München, Research Unit Molecular Endocrinology and Metabolism, Neuherberg, Germany

Abstract

Metabolomics addresses a comprehensive view of metabolites depicting biological and nonenzymatic processes. The metabolomics requests integrative application of epidemiology, analytics, and bioinformatics and provides information on specific metabolic signatures in health and disease. Processes behind a metabolomics experiment are described including study design, sample preparation, choice of analytical methods, and data processing. Major trends in future research are reviewed.

Keywords

Metabolomics; Study design; Confounders; Preanalytics; Quality assurance; Randomization; Replication; Precision medicine

Acknowledgments

I would like to thank my collaborators Dr. Janina Tokarz and Dr. Alexander Cecil from Helmholtz Zentrum München for their comments and critical review of this chapter.

1 What is metabolomics?

Living organisms are predetermined by their biological structures in how they could react to intrinsic or environmental challenges. While the genome contains a blueprint of an organism, the proteome and the metabolome are the functional agents. Thus, genomics can read out the predictive information, but proteomics and metabolomics depict ongoing processes [1, 2].

Metabolomics signatures (including the presence, concentrations, and changes of them) of metabolites originate from a biological system. This living system is represented by a sample taken from an organism, a cell culture or an environment under specific conditions. Further, the sample can be taken from an environment where organic life is no longer present but has left metabolites behind, for example, deep sea sediments or raw oil. The term metabolome is used to address the complete set of metabolites in a selected sample.

Developments in metabolomics resulted in improvement of analytical methods in their sensitivity and reproducibility, larger metabolite coverage, and higher sample throughput. This, in turn, promoted a growing number of applications in biomedical research, as well as applications in crop and food quality. Metabolomics is a very integrative research activity (Fig. 1).

Fig. 1 Metabolomics is an integrative research discipline.

The research in metabolomics combines elements of study design, analytics, and bioinformatics. Metabolomics profits from strategies and tools developed in each discipline. The study design benefits from approaches in epidemiological research, while bioinformatics profits from that in genomics. The specific contributions may vary depending on the scientific question addressed and have to be adapted to the requirements in a metabolomics experiment. These requirements may request different sample randomization [3] or data imputation [4, 5]. Because of the presence of multiple variables in sample preparation and analytics, the research in metabolomics requests specific (i.e., distinct to that of other omics) quality control (QC) and assurance procedures [6]. Analytical technologies that are used in metabolomics include nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry coupled either to gas chromatography (GC-MS) or to liquid chromatography (LC-MS) [1, 7]. The number of compounds identified in metabolomics approaches exceeds over 100,000 individual metabolites but there are many compounds not resolved beyond their mass spectra [8]. Contrary to genomics, the metabolomics does not depict the whole metabolome in a given sample as yet.

2 Flow chart of metabolomic research

There are several phases of development of a metabolomic project (Fig. 2).

Fig. 2 Workflow of a metabolomics experiment. Different steps are interdependent and request distinct expertise.

Distinct phases are schematically placed over a spiral drawn in Fibonacci numbers proportions. This representation does not imply any mathematical structure in study design. The spiral rather reflects the impact of changes at a single level of freedom which then projects into the increase in complexity in the next step of experiment. Fibonacci approach originally explained the development of population [9]. This approach found further practical use in addressing study design in clinical studies [10, 11] and description or prediction of molecular lipid species [12].

2.1 Scientific question

The metabolomics project may start in any research field requesting assessment of changes or functional explanation of a phenotype by the use of metabolite analyses. Such requests are usually addressing an issue unsolved so far. The scientific question has to be very specific and seek for explanations than sole differences. A simple comparison of control with challenged sample will always show some differences but will not take full advantage of metabolomics. The latter has its strength in mechanistic analyses of the effects. In this step, the decision on the study model (i.e., human or animal) and type of metabolite detection (i.e., NMR or LC-MS) has to be made. The decision on analytics determines as well future replication approaches.

2.2 Study design: General aspects

The process of study design is a very critical step in a metabolomic experiment. It has to ensure that phenotype resolution, metabolite coverage, quality-driving resources, and logistic processes are considered as depicted in Box 1.

Box 1

Elements of study design for metabolomics.

2.3 Study design: Pilot study and power calculation

The clarity of data interpretation is facilitated by an adequate study design. A pilot experiment might be very instrumental in verifying the scientific question and the analytical approach chosen. This activity might be omitted if enough published data are present to perform a replication study. Otherwise, a pilot experiment shall address several specific aspects like pre-analytics, compatibility of the matrix with the chosen analytics and variability of metabolic signatures. On the overall, a pilot experiment shall be kept small in sample numbers. The sample selection in this step is very critical and should avoid comparison of extreme phenotypes but include intermediary phenotypes. The strategies are depicted in Box 2.

Box 2

Examples for sample selection for pilot experiments.

The number of biological replicates has to be determined according to the expected variance of metabolites of interest (the latter provided by a pilot experiment or public records). For the same scientific question, for example, metabolomic signatures of aging, the number of individuals enrolled in the project would be quite different in cross-sectional studies (high numbers required) and longitudinal studies (lower numbers required). The latter would represent the most robust design and the interindividual variability would not be masking the progression of aging. If the resources for a pilot experiment are limited, then bioinformatic procedures to select fewer samples such as k-medoid clustering might be chosen [13]. The pilot experiment provides essential information for the calculation of the requested sample number by power calculation [14]. A pilot study is pivotal in ascertaining or determining the power level of any statistical testing after the measurements are done. The power calculation for a larger project can be further refined using clustering methods [15, 16].

2.4 Study design: Sample identity

Samples should be unequivocally identifiable to allow tracing and phenotype assignment. This can be achieved by many measures including machine and human readable barcoding of tubes, and further by integration of other data like the genotype, RNA profiles, biomarkers of disease or even confounders. Outliers visible in metabolomics signatures might be due to a labeling mismatch, technical analytical issues or extreme phenotypes. Comparison of data for the same sample from different sources helps to understand the origin of metabolite variance. For example, a wrong sex assignment could be discovered by miRNA analyses in human samples [17]. Human nonresponders to medication might have fast drug metabolism (identified by drug metabolites) or might avoid actually taking the drug (identified by lack of drug or its metabolites). In cell culture experiments, cell identity should be checked by STR or karyotype analyses [18].

2.5 Study design: Preanalytics

Conditions and procedures of sample collection and storage effects may contribute to a huge metabolite variance [19, 20] and skew metabolic phenotypes. Therefore, compliance with standard operating procedures (SOPs) is a prerequisite. In prospective or de novo studies, the implementation of SOPs is rather straightforward, but in multicenter or retrospective studies the differences in SOPs pose a challenge to data consistency [21]. Samples stored under conditions not arresting the metabolism will be useless for metabolomics experiments. Storage at less than − 80°C is definitely preferable [22].

2.6 Study design: Sample matrix

Metabolites are present in different matrices which have to be considered for study design and analytical procedures [23]. For the optimal phenotype resolution, the matrix (e.g., plasma or cell culture media) with the most prominent effect expected has to be chosen as a primary target of analyses. When analyzed for the first time, each matrix has to be adopted for extraction and processing procedures to ensure maximal output and analytical performance of metabolites [24]. The concentrations of metabolites in different types of matrices cannot be compared directly due to extraction or ion suppression effects [24, 25].

2.7 Study design: Confounders

Metabolic phenotype analyses in a human population are very challenging. Contrary to animal models or cell culture approaches, there are many common factors co-modulating metabolism. Factors not causing the phenotype but changing its appearance or penetrance are called confounders [26]. Strong confounders in healthy humans are sex, nutrition, age, or ethnicity. Further environmental confounders include smoking or alcohol consumption and physical activity [27]. During study design, the potential confounders have to be identified and a decision should be made either to prematch the individuals in the cohort to be collected (applicable to small cohorts) or deliberately ensure that all confounders are frequently present (large cohorts). There are different approaches for deconvoluting confounder impact on metabolome [27].

2.8 Study design: Time schedule

Specific phases of metabolomics experiment request completion of previous steps. For example study design, power calculation, and verification of preanalytical parameters should be accomplished before the measurements start.

Further, distinct steps in metabolomics experiment request time to accomplish. Some activities might be decoupled, like those of sample collection, measurement, and bioinformatic analysis, allowing for other activities in between. These might include phenotype retrieval from data repositories, sample collection identity validation, maintenance of equipment or server time booking. The overall time requested to accomplish all steps of a metabolomics experiment is frequently underestimated. The latter may cause problems in the coordination of analyses and late deliverables to different parts of project flow.

2.9 Study design: Randomization

Samples selected for analyses have to go through sample processing (like extraction or derivatization) and measurement(s). Not all of these steps will be accomplished on the same day and the measurements will be done in batches. Inevitable procedure or instrument drift occurring despite SOPs will cause confounding for the measurement day. To avoid this effect contributing to the creation of artificial differences between batches, randomization of samples has to be undertaken. Metabolomics requires different randomization to that used in genomics where all samples are to be randomly distributed. Instead, stratified randomization is recommended [3]. In this approach, dependent samples (e.g., longitudinal samples of the same patient, samples prior and post-challenge) are not randomized between batches but kept together. Randomization is then performed for major confounders which may include study center, sex, age, and BMI among others.

2.10 Study design: Budgeting and resources

With a growing number of samples, a detailed analysis of costs and resources required for the successful completion of metabolomics project becomes inevitable. Costs can be accommodated by (a) the choice of specific analytical methods addressing selected metabolites at low costs or (b) by reduction of sample number. For the latter, power calculation and selection of representative samples have to be performed. To avoid batch effects, all samples have to be moved in the same experimental flow. This applies for example to the transfer of samples from collection/storage center to the analytical center with safe logistics and sample tracing possibilities in both centers.

2.11 Study design: Contingency plan

As all complex systems, the metabolomics experiment flow can be disrupted unexpectedly, too. In case of resources not being sustainable (e.g., unique human samples), a contingency plan based on risk analyses and recommendations on how to avoid issues always has to be implemented (Box 3). In all preparations for the metabolomics experiment the SOPs shall be in place.

Box 3

Suggested contingency plan elements.

2.12 Study design: Legal aspects

In both human and animal studies, ethics permission is to be collected prior to collection of samples [28].

A unique ID of the sample has to be used in all steps of the experiments to avoid mismatch. In human studies this research-grade ID has to be anonymized for patient origin [29, 30]. This has to ensure that only qualified personnel (not the metabolomics laboratory) have access to all the phenotype data. The latter can be used anonymized during bioinformatic analyses. This procedure has to be implemented prior to sample release from the collection center.

Animal samples or human samples might have different biological safety levels because of their origin. Human samples might be potentially infectious whereas animal samples might be genetically modified or veterinary regulations might be in place. During the study design, the corresponding regulations have to be considered and the agreements signed beforehand.

In all types of metabolomics studies it has to be further clarified if services claimed or tools used during the project are impacting dissemination (e.g., publication or patenting) of discoveries. Multiple suggestions on how to prepare a legal agreement between academic and business collaborators have been prepared already [31].

2.13 Study design: Governance

The metabolomics requests multiple talents to be involved and a lot of distinct activities have to be undertaken. The productivity in the team will be increased when specific responsibilities, monitoring, and management structures are agreed and implemented [32]. During governance, structuring any bottlenecks in all critical processes have to be determined to prepare a contingency plan and exclude ineffective approaches [33].

2.14 Study design: Replication

All discoveries need to be replicated to become a significant and reliable scientific contribution. To allow replication, several procedures have to be implemented during discovery phase including quality assurance [34–36], implementation of SOPs and standardization [6, 37, 38] and an accessible and annotated database of results [39–41].

2.15 Analytics

The measurement of metabolites highly relies on quality assurance procedures which include hardware maintenance prior to experiments, strict adherence to SOPs for transport, storage and measurement, and further verification protocols for sample identity. Instrumental in many aspects are matrix-matched reference samples used across different experiments to ensure monitoring of analytical variability. Within this step, analytical QC supports further decisions on remeasurement and handling of outliers. In Box 4, the key features for different wet lab metabolomics approaches are provided to facilitate their choice.

Box 4

Key elements of different types of metabolomic analytical methods.

The decision on which type of analytical approaches is to be used depends on the scientific question, the time schedule, and the resources available. Whereas in discovery phase all metabolomics approaches have proven to be successful, the present policy for clinical applications requests absolute quantification of metabolites.

2.16 Data validation

Immediate data analyses post-measurement addresses verification of results according to technical parameters like variability of metabolites in the reference samples, determination of metabolite missingness in a given matrix, analysis of batch effects, and determination of technical and biological outliers. All of these parameters could be as well used for determination of sample quality. In this step, some samples might be allocated for remeasurement to support robust data set creation.

2.17 Data release

This step ensures that the data set can be further processed with deep drilling biostatistics methods. The results are not only provided in an omics-compatible format but are as well accompanied by a listing of SOPs applied, any deviations from SOPs, assays applied, randomization [3], normalization, and data imputation [4] (if any) protocols used. This data set is frozen, that is, no further changes are made to the data (Box 5).

Box 5

Minimal checklist for a metabolomic experiment

3 Future trends

3.1 Standardization

At present, the comparability of metabolomics results is hampered by the diversity of procedures used to collect, store, measure, and analyze data [42]. There are different approaches worldwide to address this diversity by making SOPs for specific processes and general flow publicly available [6]. Standardization is a prerequisite for successful replication and promotion of discoveries into biomedical or clinical use.

3.2 Coverage and speed

As depicted in Box 4, there is a counterbalance in the speed of measurement and coverage of metabolites at present. The tendency in the hardware and software development, however, is to provide fast analytical methods covering large amount of metabolites with high resolution [41, 43]. There is currently a process of convergence between targeted and nontargeted metabolomics because in both approaches the number of metabolites identified is above 1000 compounds.

3.3 Accessibility

Metabolomics data deposition has not yet reached those standards implemented in genomics. This observation applies to both the way the data are described and the annotation of metabolites to pathways. Activities represented by MetaboLights for deposition of raw data [44], Rhea for assignment of metabolites to specific biochemical reactions [45] or MetaboAnalyst for pathway analyses [46] are illustrative for developments to come.

3.4 Mechanisms of disease

Analyses of the human metabolome reveal that despite its dynamics, the metabolome is quite stable over months [47] and several years [48]. Further, observed deviations from stable signatures indicate specific influence originating from BMI changes, aging, or disease [49]. Common human diseases like diabetes type 2, cancer, cardiovascular, lung, or neurological disorders are not explained by genetics. The development of diseases might originate from a somatic mutation [50, 51], or from an epigenetic [52, 53] or environmental background [54]. Metabolomics facilitated disease etiology studies explaining changes in signal transduction pathways [55], epithelial-mesenchymal transition [56], impact of toxic exposure [57, 58], gender-specific issues in disease [59, 60], transgender aspects [61], mental health [62, 63], and aging signatures [64, 65]. Microbiome research perfectly addresses nutritional and symbiotic effects of microbiota on human health and disease [66, 67] and correlation to disease progression [68].

3.5 Biomarkers

Metabolomics depicts endpoints of biological processes in health and disease [69]. Therefore, it delivers a huge amount of information for individualized phenotyping of human disease [70]. Changes in concentrations of selected metabolites act as biomarkers. Several metabolites might be used to calculate their ratios. The latter are overcoming individual variabilities of the model and represent normalized parameters with higher phenotype explanation [71, 72]. Research on future precision medicine profits a lot from metabolomics-derived information on drug efficacy prediction [73, 74], early disease diagnosis [75–77], identification of distinct signatures specific to different disorders of the same organ [2, 78], or disease progression [2, 79, 80]. The search for predictive biomarkers by metabolomics [73, 81] supports research in preventive or personalized medicine [74, 82].

References

[1] Tokarz J., Haid M., Cecil A., Prehn C., Artati A., Moller G., et al. Endocrinology meets Metabolomics: achievements, pitfalls, and challenges. Trends Endocrinol Metab. 2017;28(10):705–721.

[2] Hocher B., Adamski J. Metabolomics for clinical use and research in chronic kidney disease. Nat Rev Nephrol. 2017;13(5):269–284.

[3] Jonsson P., Wuolikainen A., Thysell E., Chorell E., Stattin P., Wikstrom P., et al. Constrained randomization and multivariate effect projections improve information extraction and biomarker pattern discovery in metabolomics studies involving dependent samples. Metabolomics.