A review of metabolomics approaches and their application in identifying causal pathways of childhood asthma

Abstract

As asthma is a disease that results from host x environment interactions, an approach which allows assessment of the impact of the environment on the host is needed to understand disease. Metabolomics has appealing potential as an application to study pathways to childhood asthma development. The objective of this review is to provide an overview of metabolomics methods, and their application to understanding host x environment pathways in asthma development. We reviewed recent literature on advances in metabolomics and their application to study pathways to childhood asthma development. We highlighted 1) the potential of metabolomics in understanding the pathogenesis of disease and the discovery of biomarkers, 2) choice of metabolomics techniques, biospecimen handling, and data analysis, 3) the application to studying the role of environment on asthma development, 4) review of metabolomics applied to the outcome of asthma, 5) recommendations for application of metabolomics based –omics data integration in understanding disease pathogenesis, and 6) limitations. In conclusion metabolomics allows use of biospecimens to identify useful biomarkers and pathways involved in disease development, and subsequently to inform a greater understanding of the disease pathogenesis and endotypes, and predicting the clinical course of childhood asthma phenotypes.

Introduction - What is Metabolomics?

Metabolomics is the study of the metabolite composition, the metabolome, of a cell type, tissue, organ, or organism.¹ The metabolome is the collection of endogenous small molecules that mark specific fingerprints of cellular biochemistry.² Metabolomics measures global sets of low molecular weight metabolites (including amino acids, organic acids, sugars, fatty acids, lipids, steroids, small peptides, vitamins, etc.), thus providing a “snapshot” of relevant biological processes. It provides a readout of metabolic activity status in relation to genetic variations, gene expression, or external stimuli.³ Such external stimuli include infections⁴ and allergens⁵, where specific metabolome profile marks the interaction between the environmental agent and host molecules, i.e. gene x environment (Deoxyribonucleic acid (DNA), ribonucleic acid (RNA), proteins, lipids, and other enzymes). Metabolites, in addition to being produced directly by the host organism, can be derived by host microbiota, as well as transformed from xenobiotic, dietary, and other exogenous sources.⁶

Situated at the end of the -omics tetralogy (Figure 1), metabolomics provides an opportunity to answer questions that are not possible with other -omics technologies. The spectrum from genomics to proteomics can provide information on what might be happening in a cell (probable cause of phenotype), processes that are subject to epigenetic regulation and post-translational modifications. Metabolomics on the other hand provides a snapshot of the entire physiology of the host and its response to the environment, which can be associated with the outcome phenotype (for example healthy vs disease) and endotypes.^3,7 Metabolomics combines high-throughput analytical techniques with bioinformatics tools to provide information on a large number of metabolites simultaneously.^2,8

Figure 1.

Tetralogy of -omics. The metabolome is at the end of the -omics spectrum and is influenced both by the changes to the upstream -omics (epigenome, post transcription and translation modifications) and environmental stimuli. The metabolites also play a role in regulation of the upstream -omics functions through feedback loops.

Why apply metabolomics approaches to studying host x environment pathways?

Metabolomics is an appealing application to monitor environment-host interaction since measured metabolites reflect alterations/dysfunctions of metabolic fluxes of various organs and cells.⁴ This method can also be used to trace back upstream molecular pathways, and as such reveal gene-environment interactions in disease development pathways.⁹ Thus, metabolomics offers the potential to identify biomarkers for host susceptibility, to assess response to environmental risk factors, to monitor for subsequent development of persistent wheeze and asthma, and elucidate biologic pathways.

How metabolomics may address the challenges of studying human infant environmental exposure and overcome limitations of animal models

Studying early life respiratory morbidity in infants and children is challenging. In vivo experiments with environmental exposures such as viral infection in infants are not possible due to obvious ethical issues.²⁵ Thus, investigations to understand the pathogenesis are commonly conducted in either animal models²⁶ or adult volunteers.²⁷ However, experiments in adult human populations cannot be extrapolated to infant populations because of the difference in immune systems, lung development, prior exposures, and the impact of and response to environment in different age hosts.

As an example, the use of murine models to study the pathogenesis of respiratory syncytial virus (RSV) as an early life environmental asthma risk factor has not fulfilled its promise largely due to differences in immune system²⁸ and anatomy²⁹ between humans and animals. In addition, animal models of human RSV infection are semi-permissive and the most permissive primates like chimpanzees, African green monkeys, and baboons still require a large virus inoculum to establish a significant infection.³⁰ Thus these viruses are not natural pathogens in animals, and these experimental models do not replicate what occurs in human disease.

With recent development of -omics methods, previously unrecognized molecular level host-environment interactions can be investigated with repeated collection of less invasive bio-specimens and integration of high-throughput metabolomics and other -omics data using bioinformatics tools.³¹ Metabolomics presents a chance to augment in vivo studies in primates and monitor the natural course of response to risk factor exposure. Metabolomics is also intrinsically suitable for in vitro study of primary airway cells and for ex vivo experimental models on explanted airway tissues, although ex vivo methods are limited by ethical issues of tissue collection and short-term viability of the tissue.³²

Despite the promise of discovery of novel molecular biomarkers and new insights into disease pathobiology, there are important limitations and design issues that must be considered. The metabolic profile of an organism is dynamic, thus it is important to recognize that metabolites only represent a snapshot of current bodily or cellular activity. While a strength of this approach is that the measure of the metabolome represents the reaction to an exposure or what is happening at a single point in time in a dynamic system, the dynamic flux of metabolites must be considered. Metabolomics experiments are also hampered by methodological difficulties including very low molecular concentrations, variability and lack of standardized sampling, as well as complex analytical and data processing methods.^33,34

Metabolomics experimental approaches

Metabolomics experiments are roughly categorized into two approaches: untargeted (global) and targeted. These strategies differ in many aspects, first and foremost in study objective (discovery vs hypothesis testing), but also in the level of quantitation, complexity of sample preparation, experimental accuracy and precision, and number of metabolites detected.

Untargeted metabolomics is an unbiased analysis of the metabolite composition of the biological entity in a specific physiological state under given environmental conditions.³⁵ We should note that with limitations of current analytical platforms, as well as the conditions under which the samples are collected and processed, it is impossible to cover all metabolites in an unbiased manner. However, an untargeted approach can be still regarded as unbiased because no metabolite is identified prior to sample analysis. The hypothesis tested in this scenario is not associated with a particular metabolite or group of metabolites (metabolome), though knowledge of metabolite classes of biological interest helps in selecting an appropriate analytical platform and the sample preparation method to enhance detection of metabolites of interest.³⁶ The global nature of untargeted metabolomics experiments enables novel areas of metabolism to be identified, but it is currently technically infeasible to positively identify all detected molecules.³⁶ Despite the technological progress, the main disadvantage of untargeted metabolomics is the time required to process the extensive amounts of raw data, the difficulties in identifying and characterizing unknown small molecules, the reliance on the intrinsic analytical coverage of the platform employed, and the bias towards detection of high-abundance molecules.³⁷ In general, epidemiological studies using untargeted approaches pose challenges as to platform selection and metabolite identification, and further studies are needed to explore replication, synthesis, and impact of current results.³⁸ Nevertheless, untargeted metabolomics has been used to discriminate severity and phenotypes of asthma. ^{39, 40}

Targeted metabolite profiling is an approach aimed at quantification of a prior known subset of metabolites that usually are of related chemical structure and/or biological activity.^38,33 The targeted approach takes advantage of the understanding of a vast array of metabolic enzymes, their kinetics, end products, and the known biochemical pathways in which the set of metabolites takes part.³⁷ Bias toward low abundance molecules is reduced with methods such as triple quadrupole mass spectrometer (TQMS) that are quantitatively reliable and allow for quantification of low-concentration metabolites that are difficult to detect.¹ In addition, sample preparation can be optimized to reduce the dominance of high-abundance molecules.³⁸ Targeted metabolomics have been applied to distinguish asthma patients from controls without disease. The metabolites measured in these targeted studies include adenosine, Adenosine monophosphate, purine, alkanes, aldehydes, ketones, and volatile organic compounds (VOCs) in exhaled breath condensate from children.^42,43

Metabolomics analytical technologies

The two most common analytical approaches for the generation of metabolomics data are nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS). NMR is a spectroscopic technique based on the principle of energy absorption and re-emission of the atom nuclei due to variations in an external magnetic field.⁴⁴ NMR produces spectral data that allow for quantification of the concentration and for characterization of chemical structure of metabolites.

Mass spectrometry acquires spectral data in the form of a mass-to-charge ratio (m/z) and a relative intensity of the ionized compound.⁴⁵ MS based metabolomics is generally preceded by a separation step, which reduces the complexity of the biological sample and allows the MS analysis of different sets of molecules at different times.⁴⁶ The most common separation techniques in MS technology are liquid chromatography (LC) and gas chromatography (GC) columns, termed as LC-MS and GC-MS techniques.⁴⁷

The choice of metabolomics techniques should be based on the objective of the study, type of sample, and resource availability. NMR is highly selective, highly reproducible, requires less sample preparation, and it produces spectra that correlate directly and linearly with compound concentration.⁴⁸ However, NMR has relatively low sensitivity, and accordingly only the most abundant species can generally be detected.⁴⁹ On the other hand, mass spectrometry, when combined with effective sample preparation and chromatographic separation, has high sensitivity and specificity, as well as good dynamic range, which makes it advantageous especially for targeted metabolomics.⁴⁹ Among major weaknesses of MS in metabolomics is quantification; the type of sample preparation used and its molecular environment affects signal intensity.⁴⁹

While NMR and MS based analytical platforms are currently time-consuming for clinical and translational application, there are new advances in measuring VOCs in real-time, specifically for exhaled breath components. Semiconductor metal oxide (SMO)-based chemiresistive sensors are being used widely due to their amenability to miniaturization, for example development of electronic nose (E-nose), which enhances mobility and cost effectiveness.⁵⁰ SMO measures the electrical resistance signal from the buildup of VOC at the surface of the polymer inducing swelling of the polymer film.⁵¹ However, low sensitivity and less selectivity of SMO-based sensors are limitations for precise phenotype and endotype identification of diseases.⁵²

The choice of biospecimen for the study of the metabolomics of respiratory disease

Biospecimens that can be used in metabolomics experiments for studying asthma development, response to environmental exposures, and asthma morbidity, include urine, stool, blood (plasma, serum, and whole blood), sputum, saliva, exhaled breath (condensate (EBC) and direct (EB)), nasal lavage fluid, and bronchoaveolar lavage fluid. The choice of biospecimen should be based on relevance and suitability to the research question and ease of sample collection procedures. Figure 2 provides a visual of rank of the biospecimen with regard to ease of collection (access, abundance, and invasiveness of collection method), specificity to the pathophysiology, and comprehensiveness of metabolite composition. The detail of biospecimen properties along with advantages and disadvantages for metabolomics experiment to study the development of asthma and response to environmental exposure are described in Table 1.

Figure 2.

Schematic depiction of the position of the biospecimen with regard to the invasiveness of collection (x-axis), relevance to airway environmental exposure (y-axis), and comprehensiveness of metabolites composition (represented by color gradient) and abundance of the biospecimen for collection (represented by size of the bubbles). The quantification of the biospecimen properties is subjective judgment of the authors on the scale of 1 to 10. EB=Exhaled breath, NLF= Nasal lavage fluid, and BALF=Bronchoalveolar lavage fluid.

Table 1.

Characteristics and physiological relevance of common biospecimen used to study metabolomics of environmental response, asthma pathogenesis, and asthma phenotypes and endotypes.

Biospecimen	Characteristics and physiological relevance to airway infection and diseases	Advantage	Disadvantage
Urine	Stable composition Relatively less complex than serum and plasma52 Reflects both physiological and pathological changes in proximal tissues and from blood perfusing distant organs53 Composition differs by age, gender, and seasons of collection53 Low concentration of total protein and high concentration of salts54 Composed of both endogenous and exogenous metabolites	◆Noninvasive ◆Easy pediatric collection ◆Abundant for repeated collection ◆Integrative view of physiological and environmental metabolites	▪Lack of proximity and specificity to airway physiology ▪Requires transport and short-term storage on ice ▪Low concentration of total protein and high concentration of salts is challenging for MS technologies
Blood	Composed of all molecules that are being secreted, excreted or discarded by different tissues in response to different physiological needs or stresses Metabolomics studies often use plasma and serum. Whole blood is commonly used for inborn errors of metabolism Maintains a normal homeostasis in the body by constant regulatory mechanisms Contains a substantial portion of large molecular weight proteins and lipoproteins 55 Mostly composed of endogenous metabolites56	◆Provides an integrative view of the instantaneous metabolic status ◆Amenable to most analytic techniques and platforms	▪Invasive collection ▪Lack of specificity to airway physiology ▪Less abundance for repeated pediatric collection ▪Analytes are tube-additive dependent ▪Identification of small molecule metabolites, especially using NMR is 55 difficult
Induced Sputum	Complex biofluid Has high viscosity and uneven consistency57 Composed of a mixture of mucins and other exudates Composed of protein/peptide components that may reflect disease presence or severity58	◆Relatively less invasive ◆Can be easily repeated ◆Physiologically relevant to the airways physiology	▪Sample collection requires trained personnel ▪Inducing saline solution may irritate airways ▪Sampling is variable ▪Contaminated with plasma, cells, saliva, and microbes58 ▪High viscosity requires further processing
Saliva	Clear and complex biofluid Equivalent to serum in reflecting the physiological state of the body Contains a variety of secreted enzymes, hormones, antibodies, antimicrobial constituents, growth factors, and low molecular weight molecules59 Dominated by short-chain organic acids, with acetic acid being the most abundant60	◆Noninvasive ◆Easy collection ◆Abundant for repeated collection ◆Easy transport	▪Endogenous metabolites concentrations are low compared with levels in the blood ▪Contaminated by metabolites from food and oral microflora, especially bacterial putrefaction
Exhaled breath (condensates and non-condensates)	EB represents the airway lining fluid, which makes it very useful to study the biochemical and inflammatory molecules in the airway62 Relatively simple composition and contains both volatile and nonvolatile molecules63 VOC can be measured with GC-MS from condensates or non-condensates EB and directly with SOM from non-condensates EB	◆Non-invasive and easily obtainable ◆Can be repeated ◆Relevant to airway physiology ◆Suitable to analyze both volatile and non-volatile metabolites64	▪Sample collection is difficult in pediatric population ▪Affected by exercise, mode and rate of breathing, nasal contamination, environmental temperature, and humidity ▪Contaminated by exogenous environmental exposures and ammonia and sulfur-containing compounds from the oral cavity
Nasal lavage fluid (NLF)	Obtained either via direct suction of nasal secretions, or using a nasal wash The upper respiratory tract is the barrier against environmental exposure Composed of large quantity of serum proteins and low concentration of nasal proteins and small molecules66 Contains secretions that are involved during early stages of viral infection and inflammation	◆Relatively accessible ◆Relatively noninvasive sample collection ◆Can be repeated ◆Relevant to airway physiology	▪Sample collection requires trained personnel ▪Use of hypertonic saline solution may irritate and produce mucus (high total protein concentration) ▪Low concentration nasal wash proteins ▪Contains contaminants from environmental exposures ▪Variation in concentration
Broncho aveolar lavage fluid (BALF)	Thin liquid layer containing peptides and proteins that contribute to host defense and other functions67 Represents the extracellular alveolar excretion from airway epithelial cells68 The small molecule metabolites in BALF represent both those measured in serum and those that are lung specific69	◆Most physiologically relevant sample for studying lower airway respiratory diseases	▪Inaccessible ▪Sample collection requires highly trained personnel ▪Very invasive ▪Requires the introduction of exogenous fluid into alveolar space ▪High protein and salt concentration and low concentrations of metabolites
Stool	Stool is comprised of endogenous human metabolites, gut microbiota metabolites, and residues or metabolites of digested materials	◆Noninvasive ◆Easy collection ◆Abundant for repeated collection ◆Metabolites represent chemical interactions between host and gut microbiota	▪Less relevant to respiratory diseases ▪Not commonly obtained in non clinical setting ▪Difficulty distinguishing nutrition, endogenous, and microbiota metabolites

Study design

Metabolomics epidemiological studies face unique challenges of bias starting with study design and selection of target population. Like other clinical-translational investigations, metabolomics studies should first have a well-defined question and include a study population of representative subjects from a well-defined population in whom unrelated factors that impact the measure of the metabolome can be held constant (e.g. medications, diet, timing, etc.).⁷¹ As an example, if a study is designed to include both control and disease subjects, the subjects need to be similar at baseline with regard to such characteristics as hospitalization, medication, nutrition, genetic variants, and physical activity.⁷¹ Heterogeneity in severity of the disease among selected subjects should also be similar to the target population. If the study is designed as a time series (repeated measure within subjects), the effect of age on metabolites should additionally be considered. However, the initial biomarkers that are identified under the minimum difference between case and control populations need to be verified in heterogeneous populations to ascertain real world clinical application of the biomarkers.

Determining the sample size

Like all experiments, metabolomics studies require a certain number of samples to be able to effectively estimate the significance of the difference between study groups. Currently there is no standard power calculation approach for metabolomics, in part because the effect size of the metabolites on phenotype and number of metabolites to be identified are unknown a priori.⁷² While an adaptive design can help in overcoming the challenges of a priori unknown effect size in determining sample size, challenges stemming from the large dimensional nature of the data remain, especially in the case of untargeted metabolomics where the number of metabolites detected exceeds the number of samples. Furthermore, if metabolites belong to the same metabolic pathways, their abundances are highly correlated.^73,72 Recently, Blaise and colleagues⁷³ introduced a new approach, based on multivariate simulation, which deals with the highly correlated structure and high-dimensionality of metabolic phenotyping data.

Sample collection, handling, and analysis

For metabolomics studies, the sample collection, sample processing, and appropriate sample storage procedures should ensure that all samples are treated in the same manner.^74,75,76 Uniformity in type and brand of collection tubes, aliquot tubes, and pipette tips is crucial. It is important to freeze the samples immediately and minimize factors such as, freeze-thaw cycles, contamination, and differing preservation methods to reduce small changes to the metabolic profiles.^77,78,79 Freezing immediately is also critical as some metabolites transform very rapidly if enzymatic activity is not stopped completely.^80,81

Studies using biospecimen repositories should ensure that samples have identical histories in terms of collection, storage, and processing, which require careful planning especially for samples that are collected as part of large multipurpose studies. Like most biological molecules, metabolites degrade overtime even when frozen. Thus the earliest possible analysis and avoiding samples with freeze-thaw history prevents potential loss of biomarker analytes.^81,82

Biological samples undergoing metabolomics analyses require a standardized metabolite extraction protocol.^48,83 The protocol needs to be optimized and validated to specific biospecimen type and metabolite species of interest. The intrinsic characteristics of the biospecimen such as presence of water, salt, and protein can pose challenges depending on the analytical platform, thus additional steps are required to suppress any undesired noise.⁸⁴ Pooled quality control and blank samples must be included with each filtration/extraction batch of samples to allow for the monitoring of variability in sample processing and acquisition, especially for studies which include longitudinal and recurrent sampling.^85,86

Metabolomics data preprocessing and analysis

Analysis of metabolomics data poses challenges common to other –omics and some challenges unique to metabolomics, including high dimensionality, use of multiple analytical methods (NMR, LC/GC-MS, etc.), high degree of collinearity between features, non-random missing data, nonlinearity, and non-normality. High-throughput NMR and MS spectral data need to undergo preprocessing in order to ensure the quality and replicability.⁸⁷ Once the metabolite features are filtered, detected, identified, quantified, normalized, and scaled, data analysis can be performed either on the spectra itself (normalized peak intensity) or on the resultant concentration table. Parametric univariate tests such as t-test and analysis of variance (ANOVA), and nonparametric univariate tests such as Mann Whitney U and Kruskal Wallis one way analysis⁸⁸ can be applied to identify biomarkers associated with the outcome of interest. It is important to recognize that univariate tests suffer from the multiple test problem and need correction.⁸⁹ Unsupervised multivariate statistics such as principal component analysis (PCA), self-organizing map (SOM), and hierarchical cluster analysis (HCA) are useful to explore pattern and clusters and to assess data quality problems such outliers and batch effects.^90,91 Supervised methods such as partial least square discriminant analysis (PL-SDA) and orthogonal PLS-DA^92,93 can be applied both to concentration tables and spectra for discriminating between disease phenotypes and endotypes, prediction, and biomarker identification. Machine learning tools such as hidden Markov, Bayesian, support vector machine (SVM), random forest, and neural network are useful for prediction and biomarker identification in large samples.^94,95

The use of metabolomics data to identify biomarkers using supervised multivariate and machine learning methods has limitations, including overfitting, which denotes the over specification of a model on the basis of irrelevant variables that separate the groups, but are not related to the classification criterion itself. To remedy the overfitting problem, adequate sample size are needed, the model should be calibrated with the application of internal cross-validation or bootstrapping, and the results should be validated in an independent external sample. Ideally, an independent team conducts the external validation step on independent subjects and on an independent analytical platform, but it is sufficient if blinded validation finds correlation between biomarker and the clinically important outcome.

Knowledge based methods for biological data interpretation

Following robust statistical analysis, the selected metabolites should be interpreted within the context of relevant metabolic pathways. Metabolic pathways are groups of metabolites and associated genes and proteins that are related to the same biological process and directly or indirectly related by one or multiple enzymatic reactions.^96,97 Several curated biological databases have been constructed and referenced by metabolomics investigators over the last decade.^98,99,100 Open source and commercial bioinformatics tools are available to access, draw, edit, and visualize the pathway networks related to the metabolites identified.^101,102

In addition, metabolites representation in metabolic pathways can be objectively assessed with metabolite set enrichment analysis (MSEA) and quantitative enrichment analysis (QEA).¹⁰³ MSEA and QEA evaluate the probability that metabolites are represented in the pathway more than expected by chance or due to phenotype. MSEA uses a preselected set of metabolites while QEA is based on the concentration of metabolites from experiments. The latter accounts for correlation between metabolites, whether a few metabolites significantly change, or whether there is a pattern of change in which a large number of metabolites change slightly but consistently.¹⁰⁴

Application of metabolomics to study the impact of environment on the host

Using the example of infant respiratory viral infection as an early life exposure and asthma risk factor it is easy to demonstrate how metabolomics sheds light on the impact of an environmental exposure on the host, on elucidating disease pathways, and on identifying biomarkers of disease.

Metabolic profiling has been utilized in infectious disease models to enhance prognostic or diagnostic methods, and to gain insight into disease pathogenesis. ^105,106,107 Evidence shows that viruses takeover and reprogram metabolic pathways, demonstrating modified metabolism or new metabolic pathways that enhance virulence. ¹⁰⁷.^108,109 Although to date, very few studies have applied metabolomics methods to investigate infant RSV and human rhinovirus (HRV) infections, it is one area in which metabolomics is beginning to be applied to characterize disease pathogenesis and endotypes. Atzei and colleagues¹¹⁰ studied a group of preterm neonates hospitalized due to RSV bronchiolitis comparing two 1H-NMR urine spectra from two bronchiolitis patients, which revealed alterations in several compounds like creatinine, betaine and glycine.¹¹⁰ Our group has developed the only other data to date profiling and contrasting the urinary metabolome of healthy and RSV infected infants, demonstrating that five metabolites were significantly decreased during acute RSV infection, represented in essential and nonessential amino acids, leukotriene, and urea cycle, and vitamin B metabolism pathways.¹¹¹ Schee and colleagues profiled VOC in exhaled breath of preschool children who wheezed (with and without HRV infection) and did not wheeze demonstrating that the VOC profile between preschool children with and without acute respiratory wheeze differ, and a sustained and distinct VOC profile was observed in children with HRV-induced wheeze after resolution of symptoms.¹¹² Fowler and colleagues have also demonstrated that VOC can be used to distinguish those with high and low pathogen load in lower respiratory tract among intensive care unit intubated and ventilated patients.¹¹³

Review of metabolomics applied to the outcome of childhood asthma

The first pediatric asthma metabolomics investigation was carried out by Carraro and colleagues¹¹⁴ less than a decade ago. They applied untargeted NMR metabolomics on EBC samples of 25 asthma patients and 11 controls between the ages of 7 and 15 years. Metabolomics profiles clearly discriminated children with and without asthma, with a 95% success rate in their classification, and identified acetylated and oxidized compounds that distinguished the children. In subsequent studies, researchers have applied untargeted NMR¹¹⁵, targeted LC-MS (targeting adenosine monophosphate and purine¹¹⁶ and leukotriene¹¹⁷), untargeted LC-MS¹¹⁸ and targeted GC-MS (targeting VOC^{119,120,121,43} and alkanes, alkenes, aldehydes, and ketones⁴²) metabolomics technologies in EB^{120,121,42,43} and EBC^116,117,118 samples to distinguish between asthmatic and healthy children,^{116,117,120,121,43} phenotypes of asthma,¹¹⁸ wheezers and non wheezers,⁴³ and transient wheezers and asthmatics⁴². In addition, E-nose VOC pattern recognition in EB has been successfully used to discriminate between asthmatic and control, as well as between mild and severe asthma in children.¹²²

In plasma sample based targeted LC-MS lipid metabolomics application, McGeachie and colleagues¹²³ found that monoHETE_0863 and sphingosine-1-phosphate could predict asthma control. In addition, integration of metabolomics and genetics data revealed that sphingolipid metabolism and immune response pathways were associated with asthma control. Similarly, Fitzpatrick and colleagues³⁹ used untargeted LC-MS in plasma samples from children to discriminate mild and severe asthma. Fitzpatrick and colleagues³⁹ identified two metabolic pathways (glycine, serine and threonine metabolism and N-acylethanolamine and N-acyltransferase pathways) associated with asthma severity. In addition, Saude and colleagues¹²⁴ and Mattarucchi and colleagues⁴⁰ applied targeted NMR and LC-MS to urine samples. Saude and colleagues showed that 23 and 28 metabolites separated stable asthmatic versus healthy controls, and stable asthmatics versus acute asthmatic patients, respectively. Mattarucchi and colleagues additionally found decreased excretion of methyl-imidazole acetic acid, urocanic acid and a metabolite similar to the structure of an isolusine–proline fragment in the asthmatics. These studies demonstrate that metabolomics may be used to identify biomarkers of disease and healthy states, as well as progression of disease states.

Limitations of metabolomics

Metabolomics provides a unique opportunity to study the interaction between environment and host because the metabolites represent the response to both environmental stimulation and upstream genetic and regulatory modification (epigenetics, transcription, post-translational modification). However, metabolomics by itself may not capture the range of environmental agent characteristics or the range of host responses during the interaction. First, an environmental stimulus unleashes a cascade of host molecular pathways, and it is difficult to pinpoint which pathway is precisely associated with the specific stimuli. Second, metabolic profiles are subject to random fluctuations, and can be influenced by diet, sleep patterns, age, smoking, and many other variables that mask the effects of disease or toxicity. Third, many of the changes in metabolites are subtle and below detection limits of commonly applied analytical techniques. Furthermore, there are a variety of factors that influence the type and amount of metabolites detected: timing of sample collection, the sample collection procedure, sample processing, stabilization, stability and storage, extraction procedures, dilution of sample, type and number of analytical methods used, and preferences of analytical assays for metabolites with certain physico-chemical properties. Identifying biomarkers of disease from this background noise is a complex analytical and statistical challenge.

Integration with other -omics data and future directions

Disease development occurs in complex biological systems, where genetic, regulatory, and environmental stimuli trigger a broad range of input-output cascades in the biological system. The pathogenesis of disease development needs to be examined within the context of this broad biological system. An emerging approach is to integrate data from a range of -omics studies (genome, epigenome, transcriptome, proteome, and metabolome), which may provide greater opportunity to understand and model the process of disease development in the context of complex biological systems.^125,126 Vertical and longitudinal integration of multi-omics data may reveal not only what will happen following exposure to an environmental risk factor, but also the long-term imprint of the exposure on the immune response and airway function leading to an understanding of the mechanisms through which the risk factor contributes to asthma development, and ultimately how asthma, or the disease under study, might be prevented.

Phenotyping and endotyping with multi-omics data can be performed using clustering, latent class models, and projection based methods. Further functional analyses using network and pathway inference can be performed with combining database and bioinformatics tools. Analysis of causal relationships using multi-omics data from experiments under different conditions and/or at different time points can be modeled in probabilistic causal networks. However, multi-omics data integration tools are in their infancy, and integration of data is complicated by many unresolved factors including handling of data with different measurement scales (eg. binary, ordinal, and interval) and how to manage missing data.

An example of progress in bioinformatics for integrating complex data, is the development of a promising method that overcomes the scale issue (integrative phenotyping framework, iPF) by Kim and coworkers and they have demonstrated its use in feature discovery and integrative clustering in high dimensional space.¹²⁷ They applied their iPF method to a large dataset of lung samples of genomic and clinical data and identified endotypes of Chronic Obstructive Pulmonary Disease and Interstitial Lung Disease. However, the -omics data integrated using this method was limited to messenger (mRNA) and micro RNA (miRNA). There is still a need to extend such an approach to upstream (genome) and downstream (proteome and metabolome) -omics data.

Multi-omics data integration and analysis pipelines for studying the pathogenesis of disease and the influence of environmental risk factors are scarce. de Steenhuijsen Piters and colleagues were able to integrate host blood transcriptome and nasopharyngeal microbiome data to show that RSV infection immune response and disease severity are modulated by dominant colonizing microbiota.¹²⁸ On a larger scale, the complexity of multi-omics data is challenging for integration because the data is generated not only from the analysis of host biospecimens, but may also include sequencing of the host microbiome, the environmental microbiome, the virus genome, and even –omics data from controlled in vitro and ex vivo experiments in host cell or tissue cultures. The multisource -omics data integration approach can be thought of as a top-down systems approach, the integration of -omics data at a population level; and a bottom-up systems approach, which integrates experimental data from primary cell in vitro experiments (Figure 3). The bottom-up systems approach provides detailed and validating information for the pathways hypotheses generated from the top-down systems approach. The integration of the -omics data from the host and in response to environment is crucial in elucidating disease pathogenesis at a systems level. This is a novel approach, and contrary to traditional approaches where both are studied separately. In this regard, asthma consortiums and cohorts such as the Unbiased Biomarkers for the Prediction of Respiratory Disease Outcomes (U-BIOPRED) project, the Severe Asthma Research Program (SARP), and the NIH Environmental Influences on Child Health Outcomes (ECHO) Children’s Respiratory Research and Environment Workgroup (CREW) have great potential in integrating complex clinical and biomolecular data and identifying and predicting various endotypes and phenotypes of asthma.^129,130 Commercial and public collaborative initiative such as European Translational Information and Knowledge Service (eTRIKS) have developed a graph based knowledge integration tool that could help to integrate experimental and clinical knowledge to map disease pathways.^131,132

Figure 3.

Illustration of multi-omics systems approach using a birth cohort (top-down systems approach) and in vitro and ex vivo experiments (bottom-up systems approach) to study pathogenesis and biomarker discovery using an example of reparatory viral infection exposure and acute morbidity, and associated subsequent chronic respiratory disease.

In conclusion, metabolomics measures relevant biological processes, providing readout of metabolic activity status in relation to genetic variations, gene expression, or environmental stimuli. Metabolic profiles are unique in providing a snapshot of the interaction between the environmental agent and host molecules, providing an opportunity to answer questions that have not been addressed with other upstream -omics technologies. However, remaining, but an area of active development, is addressing the challenge of integrating and interpreting these complex data in order to address mechanistic questions about disease pathogenesis, disease phenotyping and endotyping, and disease prediction.

Abbreviations

ANOVA

Analysis of variance

CREW

Children’s Respiratory Research and Environment Workgroup

E-nose

Electronic nose

ECHO

Environmental Influences on Child Health Outcomes

eTRIKS

European Translational Information and Knowledge Service

EB

Exhaled Breath

EBC

Exhaled breath condensate

HCA

Hierarchical cluster analysis

HRV

Human rhinovirus

iPF

Integrative phenotyping framework

LC

Liquid chromatography

GC

Gas chromatography

RNA

Ribonucleic acid

mRNA

Messenger RNA

MSEA

Metabolite set enrichment analysis

miRNA

Micro RNA

NIH

National Institute of Health

NMR

Nuclear magnetic resonance

PL-SDA

Partial least square discriminant analysis

PCA

Principal component analysis

QEA

Quantitative enrichment analysis

RSV

Respiratory syncytial virus

SOM

Self-organizing map

SARP

Severe Asthma Research Program

MS

Spectroscopy and mass spectrometry

SMO

Semiconductor metal oxide

TQMS

Triple quadrupole mass spectrometer

SVM

Support vector machine

U-BIOPRED

Unbiased Biomarkers for the Prediction of Respiratory Disease Outcomes

VOC

Volatile organic compound