Data analysis strategies for targeted and untargeted LC-MS metabolomic studies: Overview and workflow

Highlights

• Data analysis is the “bottleneck” of metabolomic LC-MS studies.

• Huge amounts of data are produced by LC-MS metabolomics.

• LC-MS data analysis strategies vary according to the type of metabolomic study: targeted or untargeted.

• Numerous data analysis methodologies exist with different capabilities.

• This review shows distinct data analysis strategies for LC-MS metabolomic studies.

Abstract

Data analysis is a very challenging task in LC-MS metabolomic studies. The use of powerful analytical techniques (e.g., high-resolution mass spectrometry) provides high-dimensional data, often with noisy and collinear structures. Such amount of information-rich mass spectrometry data requires extensive processing in order to handle metabolomic data sets appropriately and to further assess sample classification/discrimination and biomarker discovery.

This review shows the steps involved in the data analysis workflow for both targeted and untargeted metabolomic studies. Especial attention is focused on the distinct methodologies that have been developed in the last decade for the untargeted case. Furthermore, some powerful and recent alternatives based on the use of chemometric tools will also be discussed. In general terms, this review helps researchers to critically explore the distinct alternatives for LC-MS metabolomic data analysis to better choose the most appropriate for their case study.

Introduction

Metabolomics [1], [2], [3] is one of the categorical platforms that constitute omics [4] (see Fig. 1). Omics is a field that aims at the study of the abundance and (or) structural characterization of a broad range of molecules in organisms under distinct scenarios. In the clinical field, high-throughput omic technologies are used for the characterization of diseases to better predict the clinical course of organisms and to evaluate the efficacy of existing or under-development therapies [5]. In food science, omics plays a significant role in the light of an improvement of human nutrition [6]. In the environmental field, omic studies aim at the evaluation of the alterations that organisms might suffer after exposure to environmental stressors [7], [8].

In all cases, the expressed molecules are involved in most crucial biological processes, and principally comprehend deoxyribonucleic acid (DNA) (genomics [9], epigenomics [10]), ribonucleic acid (RNA) (transcriptomics [11]), proteins (proteomics [12]), and other small molecules (metabolomics [1], [2], [3]). In more recent years, another categorical omic platform named fluxomics [13], [14], which aims at the study of the fluxome, or the total set of fluxes in the metabolic network of the biological specimen, has gained relevance. Apart from these categorical omic platforms, a variety of omic subdisciplines have also emerged (e.g., lipidomics [15], glycomics [16], foodomics [6], [17], interactomics [18], and metallomics [19]), showing that omics is a constantly evolving discipline. Among all these omic platforms, metabolomics is becoming increasingly popular and is used to detect the perturbations that disease, drugs or toxins might cause on concentrations and fluxes of metabolites involved in key biochemical pathways [20]. Due to its importance and relevance, the current study concentrates on metabolomic data.

Several analytical techniques have been developed for each of the omic platforms (see Fig. 1), including DNA microarray-based and RNA-sequencing techniques [21], nuclear magnetic resonance (NMR) spectroscopy [22], [23] and mass spectrometry (MS) methods [24], [25]. In the field of metabolomics, both NMR and MS techniques are the most popular. High-resolution proton NMR spectroscopy (¹H-NMR) has proved to be one of the most powerful technologies for examining biofluids and studying intact tissues, producing a comprehensive profile of metabolite signals without separation, derivatization, and preselected measurement parameters [26], [27]. On the other hand, MS methods, both by direct injection [28] or coupled to chromatographic techniques [29], have also evolved into a powerful technology for metabolomics due to their ability in the analysis of low molecular weight compounds in biological systems. These two approaches (i.e., NMR and MS) are complementary, and the integration of both technologies to provide more comprehensive information is now pursued in the metabolomics field. Nevertheless, this study concentrates on MS-based metabolomic data.

Concerning MS instrumentation, high-resolution mass spectrometers are the most powerful analysers due to their ability to improve accurate mass determination. In fact, spectrometers such as time-of-flight (TOF) [30], quadrupole time-of-flight (Q-TOF) [31], and Fourier transform ion cyclotron resonance (FT-ICR) [32] spectrometers and orbital ion traps [33], have substituted in many cases the conventional low-resolution quadrupoles and linear ion traps (IT), due to their ability to resolve isomeric and isobaric species and elucidate elemental composition [34]. Regarding chromatographic techniques, early metabolomic studies were commonly based on gas chromatography (GC), since it is a highly efficient, sensitive and reproducible technique [35]. However, GC has the drawback that only volatile compounds or compounds that are made volatile after derivatisation can be analysed, and extensive sample preparation is often required. In contrast, high-performance liquid chromatography (HPLC) and ultra high-performance liquid chromatography (UHPLC) are considered to be more comprehensive than GC since they allow the analysis of a wider range of metabolites without the requirement of derivatisation [36], [37], [38], [39]. Hence, liquid chromatography coupled to mass spectrometry (LC-MS) has lately gained popularity in the metabolomics field in detriment of gas chromatography coupled to mass spectrometry (GC-MS), this being the reason why this study is focused on the former technique.

The improvement of analytical techniques has gradually caused metabolomic data sets to become larger with more intricate inner structures [40]. Mass spectrometric based techniques generate highly complex data, due to the vast number of measurements (i.e., MS spectrum at each retention time) related to the number of observations (i.e., samples). In the case of LC-MS analysis (see Fig. 1), data generated from each chromatogram are arranged in data sets containing information of mass-to-charge (m/z), retention times and intensities. Hence, massive amounts of information-rich MS data are generated in the analysis of every sample, thus requiring specific standard approaches for its study and interpretation [41].

In general terms, data analysis strategies are classified in two groups: data analysis strategies for targeted (Fig. 2) and untargeted (Fig. 3) metabolomic studies. The reason for such differentiation is due to the different types of data generated in these two approaches, which require being handled accordingly. Targeted studies [42] focus the research on a set of known metabolites whereas untargeted studies [43] allow a more comprehensive evaluation of metabolomic profiles. Most of the methodologies used in early targeted studies just allowed the identification of a few number of metabolites [44]. Nevertheless, recent targeted methodologies enable large-scale metabolic profiling, including hundreds of compounds [45], [46], [47]. However, the number of compounds analysed in untargeted studies is even larger. This is so because one must process entire data sets including thousands of metabolite signals, and among these, few are finally identified as candidate biomarkers [48]. Therefore, data analysis strategies for untargeted studies require highly-extensive processing of LC-MS chromatograms. A large number of data analysis strategies are found in the literature but none of them can be singled out as the optimal choice in all cases, which makes data analysis an open task in the bioinformatics research. In fact, the field of MS-based metabolomics is rather young, and new methods, software and platforms are being regularly published or updated [49], [50].

A recent review of Yi et al. [51] summarizes recent and potential advances in chemometric methods in relation to data processing in untargeted metabolomic studies. Various aspects, including raw data pre-processing, metabolite identification, and variable selection and modeling are accurately discussed and presented there. The present review complements the previous one with some data analysis steps not covered or partially covered by the former (e.g., data acquisition, data storage and conversion, data import, data compression and feature detection or peak resolution), presents novel and little known chemometric tools for data analysis and includes an overview of the data analysis strategies for targeted studies. Moreover, it is intended to contribute to the state-of-art by providing comprehensive information on bioanalytical and data processing tools rather than describing the principles of the chemometric methods that can be used in LC-MS metabolomic data analysis.