Plant growth and development are significantly affected by various stress conditions. A direct consequence of a stress perturbation is an alteration in the metabolic behavior of the cell, leading to a cascade of molecular and biochemical events that facilitate a new steady state to be reached. Depending on the development stage of the plant or type of stress applied, the phenotypical output can be associated with reduced or delayed germination efficiency [1], decreased vegetative growth [2], delayed or arrested cell cycle activity [3,4], earlier or delayed transition to flowering [5,6], decreased formation and viability of reproductive organs (for review, [7]), or accelerated senescence [2,8]. Small molecules (metabolites) play an important role during these transitions and adaptations. Perhaps the most well documented changes are the accumulation of certain metabolites, such as proline (induced upon drought, cold, osmotic, and salinity stresses, [9,10]), soluble major and minor sugars (induced upon drought, light, cold, and nutrient stresses; see also Section 3.3), glycine betaine (induced upon salt, drought, and cold stresses [11],) and amine containing compounds such as putrescine (induced upon cold and oxidative stresses, [12,13,14]), spermine (induced upon cold, oxidative, and salinity stress, [13,14]), and γ-aminobutyric acid (GABA) (induced upon salinity [15], anoxia [16], and carbon deprivation [17]). The osmotic compatibility within the cell is maintained by the accumulation of some of these metabolites (e.g. proline, glycine betaine, GABA, and sugars), decreasing the entropy levels within the cell, and allowing for folded native tertiary proteins structures to be maintained. The accumulation of other metabolites (e.g. ascorbate, glutathione, vitamin B₁, and B₆, [18,19,20,21]) also significantly reduces the harmful effect of reactive oxygen species (ROS) generated by abiotic stresses while ROS itself might act as an important messenger during stress responses (for review [22]). Metabolites can also serve as important allosteric regulators during abiotic stresses. Allosteric regulation involves the direct binding of a metabolite to a protein, which modifies the interaction, localization, stability, or substrate affinity of the gene product. Furthermore, the contribution of metabolites (particularly phytohormones and sugars (Section 3.3)) as important messengers during stress signal transduction under adverse growth conditions has also been well documented (for example [23]). As the metabolome (metabolite complement of a cell) represents a snapshot of the prevailing biochemical state of a particular organ or tissue under investigation, and the current estimation of the chemical complexity across the plant lineage approximates to ~200,000 chemical diverse fingerprints [24], the scope for identifying and elucidating more signaling networks upon abiotic stresses remain to be discovered. Moreover, metabolic activities respond to stress more quickly than transcriptional responses [25], suggesting that the elucidation of metabolite changes will increase our knowledge of how complex metabolic networks interact and how they are modified upon specific stresses [25]. Using metabolomics as an important diagnostic tool within the right discovery context, thus, provides a powerful means to gain novel biological insight into the metabolic regulatory network. For this purpose, this review does not attempt to detail individual metabolite changes in response to abiotic stress conditions; for a recent comprehensive review covering these aspects we refer to [25,26]. In addition, for analytical techniques associated with plant metabolite profiling, and data analyses and processing pipelines, excellent reviews have recently covered these [24,25,26,27,28]. Here, rather, we focus on the interaction between, and wealth of information that can be uncovered in studying, metabolomics combined with other molecular platforms during abiotic stress responses. The overall result is a more thorough and intricate understanding of the global adaptation patterns and hierarchical regulation structures that prevail during abiotic stress conditions.

Key to our understanding of the central dogma of molecular biology is the linear progression of molecular events associated with it; initiated with the encoding and replication of the genomic information contained in the DNA of an organism. This gives rise to the production of messenger RNA (mRNA) that will allow for differential gene expression patterns to emerge during the process of transcription. The mRNAs are subsequently processed by an array of splicing and editing mechanisms before exiting the nucleus to the cytoplasm. Within the cytoplasm, the mRNAs will interact with large ribosomal complexes which will translate the information contained on the mRNA molecules, leading to protein synthesis. Each protein synthesized can then catalyze a set of biochemical reactions leading to the dissipation of a given substrate (metabolite x) to its biochemical (end) product (metabolite y). Until recently, this widely perceived interpretation of a metabolic pathway was commonly applied (and perhaps naively the only considerable viewpoint) when considering and interpreting the biological meaning of transcriptomic, proteomic, or metabolomic datasets when viewed in isolation. The pitfalls associated with this assumption have especially become apparent when combining two or more –omic technologies in order to obtain a more holistic view of metabolism. For a number of examples metabolic changes do follow the conventional transcriptional response. For instance, characterization of the core aerobic and anaerobic responses in rice have identified that decreased transcript abundances in sucrose catabolism or tricarboxylic acid (TCA) cycle reactions corresponds to decreased hexose equivalents and organic acids, respectively [29]. Consequently, a metabolic switch to fermentative metabolism is apparent.

However, what happens if such a clear progression (correlation between transcript (or protein) and metabolite levels) is not eminent in the dataset? How do metabolomics (and associated bioinformatic tools) aid in resolving such (apparent) discordances, and how will this influence how we evaluate metabolite data in future? Perhaps, the first important aspect is that the unequivocal identification and robustness in the dataset(s) have to establish. Currently, metabolite profiling efforts primarily focus on hyphenated chromatography based technology coupled to mass spectrometry (gas chromatography mass spectrometry (GC MS) or liquid chromatography mass spectrometry (LC MS)) high throughput techniques to profile highly to moderately abundant metabolites. In order to achieve this, unequivocal metabolite identification relies on the establishment and verification of custom metabolite libraries that rely on retention time correction indexes (retention index, RI) and matching mass-to-charge (m/z) identifiers. While some search algorithms are using more quantitative means of matching spectra to specific metabolites, the overall generalization still prevail to match 4–6 of the most unique m/z identifiers to a specific metabolite within a specific RI window. However, metabolites that share similar m/z identifiers, and elute in close proximity will either be misidentified using this methodology or would not be distinguished from metabolites sharing similar chemical properties. Improvements in recent MS technologies have led to the development of mass accuracy instrumentation, which significantly enhance the elucidation of metabolite identities that share similar spectra information. Moreover, technologies such as nuclear magnetic resonance (NMR) provide the most unambiguous information regarding the structural properties of a small molecule. While the use of this instrumentation has several limited capacities in terms of the sensitivity of detection, new applications such as coupling NMR and hyphenated technologies will significantly advance the profiling and identification of distinct metabolites, and allow for more robust metabolite data reporting. Improved peak resolution by using two-dimensional (or multi-dimensional) GCxGC time of flight mass spectrometry (TOF MS) has also been achieved to significantly improve separating power, and allow for a higher resolution capacity [30]. Several artifacts may also arise from derivatization steps during sample preparation of nonvolatile metabolites for GC MS analyses [31]. As an example, the conversion of arginine to the trimethylsilyl (TMS) derivative of ornithine has been previously documented [32], while the systematic detection and documentation of artifacts arising particularly from GC MS based experiments have been collated into in house databases (e.g. [33]). One approach to address such artifact formation is through the classical addition of authentic standards or spiking experiments. The annotation and quantification of metabolites can also be improved by different stable isotope labeling strategies that, in combination with standardized database searches and elemental mass calculations, allows for reduced chemical background noise and increased confidence in the elemental formula annotation [34]. A highly sensitive, ultrafast (15s) isotope-derivatization reagent, d₀-/d₆-2, 4-dimethoxy-6-piperazin-1-yl pyrimidine (DMPP), has also been developed to specifically derivatize carboxylic analytes, which will reduce chemical noise [35], leading to a greater confidence in the unequivocal identification of these acidic compounds. Computational annotation methods to assist in refining metabolite annotations have also been developed (e.g. [36]), and will continue to improve the accuracy that metabolites are identified with.

Apart from metabolite identification, any other apparent discrepancy in metabolic changes also has to be rigorously evaluated in the biosynthetic context. Customized in-house and publicly available software packages have been developed for standalone metabolomic data and pathway enrichment analyses that greatly aid in statistically assessing whether metabolite changes corresponds to whole pathway enrichment analyses (e.g. [37,38,39]) or graph-clustering algorithms in order to understand the organization of metabolic functional modules [40]. While such analyses have not been strictly applied to plant specific abiotic stress metabolite profiles to date, the integration of these applications greatly accelerates and creates confidence in the information generated from metabolite profiles to determine the metabolic pathway activity associated with the particular perturbation. Other bioinformatic tools to discriminate and interrogate metabolite data include both supervised and unsupervised visualization techniques, including batch learning self-organizing map analysis (BL SOM) [41], orthogonal projection to latent structures discriminant analysis (OPLS DA), principal component analysis (PCA) and Bayesian independent component analysis (ICA) in order to reduce the dimensionality or non-overlapping information in the data [42]. In addition, in a complementary approach, metabolomics and other—omic technologies could be performed in parallel and, for a combined metabolomics and transcriptomic approach, the output can be visualized; overrepresentation and other interrogative statistical analyses performed [43,44,45,46]. A number of studies have successfully integrated data from metabolomics to transcriptomics, and even enzyme activities [47,48,49,50], attesting to the value in using parallel approaches to understand plant metabolism (see also below). This review highlights several of these combined approaches in order to understand specific abiotic stresses (or abiotic stress responses in general), and the advancement of knowledge made in understanding the molecular context of metabolite changes through both targeted and untargeted profiling techniques. These metabolic changes are specifically discussed for their alternative roles in metabolism, ranging from structural modifiers of nucleic acids to important mediators or indicators of signal transduction mechanisms, concepts largely ignored when interpreting the context of metabolic changes during stress perturbations.

Any disagreement between transcriptomic and other –omic (proteomic and metabolomic) datasets are usually interpreted as an additional layer of regulation either occurring on a post-transcriptional or—translational level (although the direct comparison between technological platforms and sensitivity and analytical capacity of instrumentation should also be taken into account) [51]. The intriguing observation that certain metabolites can exert a transcriptional response preceding the induction or independent of its own biosynthetic gene expression has, however, emerged in the last years [49,52]. In addition, the exogenous application of certain metabolites that modulate or mimic abiotic stress responses [52,53] also attribute to the complexity of interpreting metabolic changes in its strict pathway context. The direct influence of prevailing metabolite levels on gene and protein expression or activity is usually referred to as metabolic regulation, ranging from simple allosteric regulation to the emergence of direct interaction of metabolites with nuclear gene expression (NGE) during retrograde signaling (Section 3.2). Regulation strategies are also most likely unidirectional as transcriptional regulation of metabolic changes will lead to the metabolic induction of transcriptional signal transduction pathways. In addition, overlapping abiotic stresses that act either synergistic or antagonistic in behavior might exhibit phases of both. As an example, the cold acclimation response and subsequent interaction with carbon availability exhibit both transcriptional and metabolic regulation strategies [54]. Thus, metabolites can exert its effect(s) on multiple hierarchical levels of control, displaying much more versatile roles than previously anticipated.

While this review has largely focused on transcript-metabolite cohort changes, and the signal transduction pathways they participate in, metabolomics can be applied to a diverse range of –omic platforms to gain functional insight into physiological and molecular responses during abiotic stresses. The application of metabolomics to phenomics, genome wide association (GWA) and recombinant mapped populations, genome and quantitative trait loci (QTL) sequencing efforts have provided the timely opportunity to explore the molecular and biochemical basis and pathway interactions for complex plant abiotic stress interactions [48,129]. As an example, in a GWA mapping strategy applied to 289 maize inbred lines, the biochemical composition of 26 metabolites were strongly associated with single nucleotide polymorphisms (SNPs), including 15 distinct SNP-metabolite correlations that underpinned the genetic variance, observed in these hybrids [130]. Such discovery modules will greatly accelerate the future directives to understand the complexity associated with, and strategies to improve, plant performance upon specific abiotic stresses. For Arabidopsis accessions, a similar approach has been applied upon C and N limitation and, once again, suggests that the resulting metabolic networks are highly variable and specific in terms of the limiting resource [131]. Similarly, upon cold stress, ecotype variance in Arabidopsis accessions could identify specific lipids or lipid classes (e.g. glucose/galactosylceramide (d18:1/c24:0), monogalactosyldiacylglycerol (MGDG) 34:2, 34:3, 36:7), and triacylglyceride (TAG 52:0 and 58:9)) through ultrahigh performance liquid chromatography Fourier transform MS (UPLC FT MS) technology positively correlated with low temperature acclimation; allowing for potential metabolic biomarker discovery that can be used in breeding strategies [132].

Furthermore, while perhaps less suited to proteomic studies dealing primarily with relative protein abundances, the integration of metabolomics with quantitative proteomics and, especially, enzyme activities [48], have also yielded far greater insight into the (post-) translational metabolic regulation of plants upon, for example, limited carbon availability [48]. Metabolites can serve as transient modulators of protein function and therefore have useful applications in the dynamic signal transduction pathway that lead to the activation of abiotic stress tolerance [133]. Cross et al. suggests that altered biomass and growth limitations of 24 genetically diverse Arabidopsis accessions upon carbon deprivation is not a result of absolute metabolite levels of starch, sugars, or amino acids; but, based on the measured enzymatic capacity, related to the relative flux through the metabolic system [48]. As such, fluxomics, currently defined as the measurement of isotopic (stable or radioactive) label enrichment of metabolites (or pools) across time in organs, tissue or cell types, plays a pivotal role in linking changes in metabolite levels to that of genomic and proteomic platforms. The technologies and approaches to measure these on a more comprehensive, high throughput basis are still being developed (for recent comprehensive review, [134]). One major limitation is our largely fragmented knowledge about plant metabolism. The identification of “new” metabolites or metabolite derivatives (both in primary and specialized metabolism), as well as novel metabolic biosynthetic pathways, is imperative to interpret metabolic flux labeling and account for flux balance equations. Current flux methodologies rely thus heavily on empirical mathematical modeling to obtain a reasonable goodness of fit of obtained experimental isotopic distribution to that of curated metabolic models. As a result, fluxomics are currently more specific and applicable for tailoring experiments to resolve specific fluxes with high precision by labeling with precursor isotopes that are in close proximity to the area or flux distribution point of interest, or within a well-defined pathway. Metabolic flux can also be only truly reflective of in vivo conditions if each metabolite (and its resulting label enrichment) in the pathway, starting with total label uptake and utilization, can be quantitated, and these parameters need to be readily measured and be accounted for in any mass calculation equation. One way to currently gain biological insight into dynamic metabolite states is to define the degree of label enrichment within each individual metabolite (as assessed via hyphenated MS technologies), without interfering flux dynamics from it.

As an example, in Arabidopsis roots exposed to oxidative stress; ¹³C glucose isotopic labeling incorporation into TCA cycle organic acids, and sucrose was significantly decreased, although the absolute concentration of these respective metabolites remained unaltered [135]. Upon recovery, experiments from oxidative stress, the metabolic recovery shifts (both steady state and dynamic metabolite levels and enrichment, respectively) precedes the enzymatic activities of the majority of the biosynthetic gene products [136], suggesting that combined metabolomics/fluxomics and proteomic approaches act as important platforms that can pinpoint the metabolic step(s) involved in plant stress responses.

Another major area of advancement of metabolomic data through the combination of fluxomics, is the broader application of isotopes to plant systems grown in closed facilities supplied with different isotopic labeling precursors (¹³C, ¹⁵N, ³⁴S) [34,137,138,139]. This approach leads to a greater confidence in current metabolite annotation and also promise to expand the plant metabolite depository needed to improve current flux balance models. Greatly aiding these assignments is improved instrumentation, including Fourier transform ion cyclotron resonance mass spectrometry (FTICR MS) and Orbitrap mass spectrometers (see also Section 2), that allows for mass accuracy and isotope elemental composition determinations. Combining such larger scale metabolite/flux studies have not been applied to plants upon abiotic stress, however, strategies to incorporate this promise to uncover unique and diverse metabolite identities and signatures that participate in various signaling and metabolic roles.

Linking chemical genomics, fluxomics and metabolite profiling technologies have the further potential to uncover novel signal transduction pathways, and also for genetic engineering strategies to transiently perturb transcript or protein levels in response to stress stimuli (reviewed in [24,140]). In addition to these experimental integration strategies, subcellular metabolomics [134,141], tissue specific metabolite profiling [142], and the spatial compartmentation of cell specific transcriptional changes upon, for example nitrogen and iron deficiency [143,144], have also provided more technical advances in refining models and hypothesis surrounding microscale metabolic changes upon nutritional stresses to date. Combined with further technologies, including fluxomics and ionomics, multidisciplinary approaches to understand abiotic stress adaptation and tolerance are rapidly expanding, and the next challenge is the systematic integration of metabolic data into its correct molecular context.

Metabolomics has taught us a great deal about the diversity in and dynamics of systems driven biological approaches during abiotic stresses. Ranging from osmoprotectants to allosteric regulators of protein properties, as well as direct interactors and modulators of DNA and RNA structures to alter nuclear gene expression, small molecules and the study thereof (metabolomics) have illustrated that, in combination with other –omic datasets, metabolites, proteins, and transcripts do not always “play by the rules of the game”. While it challenges us to shift our assumptions and interpretation of metabolic changes in terms of how we think metabolism operates, it also greatly attests to the complexity within biological systems. Our current understanding of data structure relationships (particularly transcript-metabolite or metabolite-metabolite comparisons), and resulting hierarchical regulation structures, have been significantly improved by addressing metabolomics driven questions by careful designing experiments that will aid in understanding the complexity and regulation mechanisms during abiotic stresses. Plant metabolomics has taught us that, while metabolomics is a powerful tool to understand specific abiotic stresses (and even how these stresses overlap), we need the integration of other platforms or genetic variability in order to understand and integrate from what we measure to what the relevant molecular context is of these metabolic changes. However, metabolomics as a standalone entity will still provide valuable information regarding the identification of uncharacterized metabolites and/or derivatives, greater quantification of metabolites on both tissue specific and subcellular levels, as well as the dynamic movement of soluble metabolites between cells and tissues, especially upon a range of stresses. In conclusion, metabolomics is a crucial link to decipher the molecular functional mechanisms(s) underlying specific abiotic stress responses.