Metabolic Profiling as a Tool for Understanding Mechanisms of Toxicity

Abstract

Metabolic profiling (metabolomics/metabonomics) is the measurement in biological systems of the complement of low-molecular-weight metabolites and their intermediates that reflects the dynamic response to genetic modification and physiological, pathophysiological, and/or developmental stimuli. The measurement and interpretation of the endogenous metabolite profile from a biological sample (typically urine, serum, or biological tissue extract) have provided many opportunities to investigate the changes induced by external stimuli (e.g., drug treatment) or enhance our knowledge of inherent biological variation within subpopulations. This article will focus on the basic principles of metabolic profiling and how the tools (nuclear magnetic resonance [NMR], liquid chromatography–mass spectrometry [LC-MS]) can be applied in toxicology and pathology. Metabolic profiling can complement conventional methodologies and other “omics” technologies in investigating preclinical drug development issues. Case studies will illustrate the value of metabolic profiling in improving our understanding of phospholipidosis and peroxisome proliferation. A key message will be that metabolic profiling offers huge potential to highlight biomarkers and mechanisms in support of toxicology and pathology investigations in preclinical drug development.

Introduction

Metabolic profiling is the measurement in unicellular to multicellular biological systems of the complement of low-molecular-weight metabolites and their intermediates that reflects the dynamic response to genetic modification (e.g., transgenic or viral) as well as physiological (e.g., gender), pathophysiological (e.g., disease morbidity), and/or developmental stimuli (e.g., aging). The measurement and interpretation of the endogenous metabolite profile from a biological sample (typically urine, serum, or biological tissue extract) have provided many opportunities to investigate the changes induced by external stimuli (e.g., drug treatment) or enhance our knowledge of inherent biological variation within subpopulations (the metabolic phenotype).

This short article will review briefly the main metabolic profiling technologies and their applications in the safety assessment of drug development.

At the cellular level, stimuli such as drugs, dietary changes, endogenous hormones, or chemicals can provoke a sequence of biochemical reactions involving receptor activation, gene transcription, and translation to protein products. The myriad of cellular biochemical interactions generates a vast and complex array of metabolites. Many of these metabolites will pass into the systemic circulation and also urine, where they will be accessible to assay by metabolic profiling technologies. Of course, a biological sample may contain metabolites from any source, but the metabolic profiling described here is concerned with endogenous metabolites and not those generated directly from administered drugs.

Various definitions of metabolic profiling have been proposed, including the “measurement of the dynamic multiparametric metabolic response of living systems to pathophysiological stimuli or genetic modification” (Nicholson et al., 1999, p. 1181), whereas Lindon et al. (2003, p. 1140) pronounced that “measures the fingerprint of biochemical perturbations caused by disease, drugs, and toxins.” Metabolic profiling relates to the relatively low-molecular-weight molecules generated within a cell, such as amino acids, lipids, and carbohydrates.

The term metabolome has been coined to represent the entire metabolite collection within a body and is analogous to the genome and transcriptome terms already in common use. In the current literature, the terms metabonomics and metabolomics are used to describe this field, sometimes with fine distinctions of definition, and at other times seemingly interchangeably. Whereas the origins of these separate terms are to be found either in plant biology mass spectrometry (metabolomics) or biofluid nuclear magnetic resonance (metabonomics), we would propose that metabolic profiling is an understandable and equally useful overall term to describe this field.

Among the battery of new technologies, sometimes referred to colloquially as the “omics,” metabolic profiling is concerned with detecting analytes of relatively low molecular weight (up to ~1000 Da) and not with the much larger proteins, glycoproteins, and nucleotides (mRNA and DNA) that are the concern of proteomics and transcriptomics (Figure 1). This pool of small, detectable, metabolic analytes includes amino acids, oligopeptides, sugars, bile acids, simple fatty acids, and intermediates of many biochemical pathways, notably the tricarboxylic acid cycle and glycolysis. Profiling the “trail” of metabolites allows the pathologist or toxicologist to explore a biological or pathological process in a different but complementary manner to that provided by other methodologies. It is pertinent to note that a metabolite profile provides a very recent snapshot of what has actually occurred in a biological pathway and succeeds the more upstream sequence of events defined by gene, transcript, and protein changes (Goodacre, 2005).

Metabolic Profiling “Tools”

The principal technologies, or platforms, that have been developed to detect metabolites are based on nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS). Both of these are capable of being used in an “open” manner that will profile the full complement of analytes within a sample to generate an all-inclusive spectrum that can be mined for further information on components. However, MS is more amenable to being used in a “closed” manner, assaying for specific, known biochemical entities and classes such as lipids or amino acids. The selection of an “open” and/or “closed” metabolic profiling approach will depend on a priori knowledge of the biochemical pathways associated with any investigation. Another way of defining these two approaches is to consider whether the investigation is hypothesis-generating (“open”) or hypothesis-driven (“closed”). Typically, NMR and MS are used in a complementary way to maximize the opportunity for identifying and quantifying metabolites.

NMR spectroscopy examines the proton spectrum of a sample, and although quantitative, is relatively insensitive (with detection at ~10⁻⁵ M). NMR is especially good for providing structural information on constituents, that being its conventional role in pharmaceutical structural identification (chemistry and drug metabolism) laboratories. It has the advantage of being a nondestructive process and is relatively rapid, generating spectra from a biological sample within about 10 minutes. NMR spectra of biological samples may comprise thousands of peaks, with each peak or multiplet representing a proton in a different chemical environment and a peak area directly proportional to concentration (Figure 2). Put very simply, the spectrum range is based on the resonance of C-H bonds, with relatively stable structures such as aromatic rings positioned to the left of the spectrum and more flexible structures such as aliphatic methyl groups positioned on the right. Other biochemical entities (nonaromatic ring structures) such as sugars will be located in intermediate areas of the plot. The biochemical complexity of a sample typically generates a spectrum displaying crowded and overlapping peaks, but individual fractions of a spectrum can be “exploded” using standard data-handling programs and observed at an increased resolution to aid analysis.

MS is used usually in tandem with preceding separation techniques such as liquid chromatography (LC) or gas chromatography. Efficiency of chromatographic separation is related to column type (hydrophobic/hydrophilic), solvent systems, and column pressure. A routine, high-performance or ultra-performance LC method might use a C18 column that will separate out polar components at the solvent front, with more nonpolar/hydrophobic components at the end of the run. Fractions of interest, separated on the basis of retention time on the column, are ionized and “sprayed” into the mass spectrometer. MS provides a highly sensitive readout of mass, based on a biochemical component’s ability to be ionized and therefore “fly” on a spectrometer, giving a limit of detection at ~10⁻¹² M. This greater level of sensitivity, compared with NMR, enables MS to detect small chemical moieties such as sulphate, amine, and carboxyl groups. Typically, might take 20 to 30 minutes to generate a LC-MS spectrum from a single sample (Figure 3).

The ubiquitous plots and spectra that represent the raw data of the metabolic profiling scientist can appear overwhelming and impenetrable to scientists from other disciplines. The journey from these complicated graphics to an understanding of a pathobiological event is not usually easy, and good analytical chemistry skills are prerequisite to this process. A typical NMR profile may contain 2 to 3,000 peaks, but only about 10% of these will map to known chemical entities. Thus, unknown peak areas may require further assessment by technologies such as MS, which produces a much enhanced resolution profile, containing possibly 15,000 peaks. But even here, only about 5% of these will map to known components. However, serial analysis by MS is possible, with the opportunity to gain increasingly more detail on the identity of spectral fragments. This serial exercise is referred to as MS/MS or MSⁿ, depending on the extent of fragmentation needed to assist identification of a chemical entity.

A single spectrum generated from an individual sample is complex enough, but biological interrogations usually demand the comparison of multiple spectra from either different groups or serial sampling of an individual animal. The enormous sets of data generated can be compared by sophisticated statistical algorithms such as principal component analysis (PCA) to seek patterns of similarity and difference. Because metabolic profiling methodologies can detect biochemical changes to a high degree of sensitivity, is possible to identify gender differences and even subtle diurnal or dietary changes in these profiles. Physiological and other “background” changes can complicate a profile, so knowledge of such “metadata” can be valuable in sorting out more relevant changes related to treatment (Lindon et al., 2007). This is a particular concern when metabolic profiling is applied to clinical samples, in which the variability provided by genomic and lifestyle influences as well as comedications must be given due consideration (Lenz et al., 2003). Data regions that contribute to differentiation can be explored via analytical chemistry, with the intent of identifying components. The limited size of the database of known metabolites is a current restriction and contrasts somewhat with the more complete nature of the databases used for transcriptomic and proteomic identification. Thus, there is no guarantee that specific metabolites can be identified, and may be that a “metabolite profile” pattern of multiple peaks of unknown or incomplete provenance is the most feasible outcome to attribute to a change.

If individual chemical components can be identified, a good knowledge of biochemical pathways allied to some detective work can aid the development of a hypothesis about the changes. This is particularly valuable for chemical entities that are perceived as having likely rather than definitive identities. Currently, there is development work around assessing “probability of closeness” within pathways and even the possibility of automating the process. Thus, for an optimal result, the analysis and interpretation of metabolic profiling data require contributions from analytical chemistry as well as statistical and biological expertise—skills rarely found in any one individual. Good teamwork and communication between these disciplines is essential for an effective outcome. In particular, the biological interpretation is a critical part of the process. If the proposed pathway or change does not make sense in the context of the known biology or pathology, is unlikely to be of value in an investigation or search for a marker.

The application of metabolic profiling in pathology and toxicology is usually related to the investigation of a lesion in a preclinical species. Typically, a drug-development project may require assistance in identifying a mechanism and/or a biomarker to facilitate the monitoring of the onset, progression, and possible recovery of a change. Metabolic profiling need not be applied in all such circumstances, of course, but can provide unique data as part of a complementary approach within an investigation.

The design of a metabolic profiling investigation will depend on a multitude of factors including the species and organ systems affected, available knowledge and materials, and the timing of the development program. The diversity of samples that are available for profiling include plasma or serum, urine, and many of the tissues sampled for histology. Other samples such as cerebrospinal fluid, seminal fluid, bile, and tissue extracts are not available typically but are amenable to profiling. Metabolic profiling can be applied to any preclinical species, although the interpretation will be limited by the extent of metabolic profiling knowledge in that species. Clinical samples may also be available but are usually limited to biofluids for obvious reasons.

Metabolic profiling has been used to address a variety of pathology and toxicity problems such as cardiovascular injury with phosphodiesterase inhibitors (Zhang et al., 2006), hepatotoxicity with receptor antagonists (Mortishire-Smith et al., 2004), peroxisome proliferation with peroxisome-proliferator activated receptor (PPAR) agonists, and phospholipidosis related to cationic amphiphilic drugs (CADs). Two examples exploring the use of metabolic profiling in phospholipidosis and PPAR projects will be considered.

Phospholipidosis

Membrane lipids recycle in endosomes as part of the cellular homeostatic process, so some accumulation of the lipids in untreated animals and humans is normal. However, abnormal or excessive accumulation of lipids, manifested as multilamellar bodies, can develop into phospholipidosis and can cause pathology in multiple organs. The current gold standard to identify and monitor phospholipidosis is the electron microscopic examination of tissues. This depends on an invasive sampling process that requires a particular methodology and is relatively expensive and slow. Furthermore, is very difficult to acquire accurate quantitative data from such samples. Therefore, a specific and sensitive biochemical marker present in an accessible biofluid would be very valuable to monitor the change.

An exploratory project to identify a marker of phospholipidosis established an initial NMR investigation of urine from rats treated with various CADs. The NMR profiles were used to identify phenylacetyl glycine (PAG) as a possible biomarker, as was present after a 6-month dosing period but was absent in the urine of controls and in treated rats that had been off dose for 3 months (Nicholls et al., 2000). Whereas the initial data were encouraging, the metabolic pathway linkage between phospholipid accumulation and urinary PAG was unclear, so a strong biological hypothesis could not be constructed. Further investigation of the pathways involved identified the possibility that dietary and drug effects on the microflora in the rat gut could be responsible for generating PAG (Delaney et al., 2004). The utility of PAG as a marker of phospholipidosis for the rat and particularly other species was therefore compromised.

A new phase of the search used tissues and biofluids from rats treated with a known CAD, amiodarone, for 7 days. Phospholipids within the various samples were separated initially by high performance liquid chromatography (HPLC). Treatment-related changes were identified in the HPLC spectra, but one lipid peak in particular was elevated in all the tissues examined (peripheral blood mononuclear cells, serum, lung, and liver) and absent in control samples (Figure 4). The peak was further analyzed via MS/MS and was identified as lyso-bis-phosphatidic acid (LBPA). This phospholipid is known to be located in late endosomes and lysosomes and represents less than 1% of the total cellular phospholipids. Examination of the literature revealed that LBPA is raised in the tissues of patients with some lysosomal storage diseases (such as Tay-Sachs and Niemann-Pick disease). Furthermore, there were some existing reports that LBPA is raised in animals and humans following CAD treatment. The molecule also has a relatively long half-life, giving some ideal properties as a potential biomarker.

Once a putative marker has been identified, further work may or may not involve metabolic profiling technologies in the lengthy but necessary validation process. It may be that the identified candidate is amenable to measurement via other methods or provides a pathway lead to a marker that may be a more suitable biomarker. In this instance, the identification of LBPA has triggered a range of further work still using conventional technologies and skills, such as histopathology and electron microscopy, but also bringing in transcriptomic analysis and flow cytometry to complement metabolic profiling. The objective will be to assess if LBPA can be demonstrated as a useful biomarker preclinically and/or clinically and to determine which technology or technologies may be most suited to rapid and accurate assay of LBPA.

Peroxisome Proliferation

Peroxisomes are normal subcellular organelles involved in the β-oxidation of long-chain fatty acids. Peroxisomal proliferation (PP) is usually deemed to be a rodent-specific phenomenon but has been noted in nonhuman primates and in humans. Some PPARs (e.g., fibrates) have been associated with hepatic hypertrophy, hyperplasia, and tumorigenesis in rodents, so a way of monitoring PP would be valuable. Conventionally, an increase in peroxisome number is characterized using electron microscopy or special immuno-staining on tissue samples. These techniques require invasive tissue sampling and are lengthy, and there are difficulties generating accurate quantitative data. A metabolic profiling project was established to determine if a marker of PP could be identified in biofluids from rats and other species.

NMR profiling of urine from rats treated with fenofibrate and other PPAR agonists demonstrated spectral changes that were identified putatively as N-methyl-nicotinamide (NMN) and N-methyl-4-pyridone-3-carboxamide (4PY; Figure 5). Further studies using samples spiked with NMN and 4PY confirmed these identities, and a time course of the urinary elimination of the metabolites was also plotted. It was known that NMN and 4PY were end products of the tryptophan-NAD⁺ pathway; this knowledge allowed other observed changes of the NMR spectra that mapped to less definitive structural components but were known to represent other parts of the same pathway to be putatively identified as well. From these data, was possible to build a hypothesis about the PP mechanism involving perturbation of the tryptophan pathway. However, further evidence was sought using complementary technologies (transcriptomics) to identify changes in the mRNA profiles of key enzymes in the tryptophan-NAD⁺ pathway. Real-time reverse transcriptase polymerase chain reaction (Taqman) was used to confirm a substantial down-regulation of amino-carboxy-muconate-semialdehyde decarboxylase (ACMSD) and an equivocal or slight up-regulation in quinolate phosphorlbosyl-transferase (QAPRT) in the liver of treated rats (Ringeissen et al., 2003; Delaney et al., 2005). These data, together with the metabolic profiling findings, indicated a clear and rational hypothesis with a shift in the flux of the tryptophan pathway away from the glutarate pathway and through the nicotinamide pathway toward peroxisomal β-oxidation (Figure 6).

Thus, urinary NMN and 4PY were identified as putative biomakers for PP in the rat, and further work was performed to explore their relevance in other species. Using similar metabolic profiling methodologies, was found that NMN and 4PY were not distinct markers of PP in the cynomolgus monkey, and therefore, in isolation, as biomarkers could not be used to predict PP accurately in this species. However, many other spectral changes mapping to the tryptophan pathway were apparent via NMR, which did allow a “fingerprint profile” that enabled good prediction of PP in the primate. There is also clinical evidence (from NMR and MS) that tryptophan pathways are affected with PPARs.

Conclusion

Metabolic profiling can aid the mechanistic elucidation of toxicologic and pathologic changes and is also a means to identify potential biochemical markers of toxicity, either as specific chemical entities or as patterns within a profile. One can envisage their use also in lead candidate selection, screening compounds to select or deselect on the basis of comparing profiles against known toxicity profiles within a database (Robertson et al., 2000; Keun, 2006).

The use of metabolic profiling in toxicology and pathology is still in its infancy, but appropriate use can augment existing conventional endpoints represented by morphologic and clinical pathology. The technology provides an additional layer of detail to the interrogation of a biological or pathological change. Indeed, metabolic profiling can play a role in supplementing the battery of clinical pathology assays via the discovery and validation of novel biochemical markers. Metabolic profiling can also complement the other emerging technologies (transcriptomics and proteomics) and the integrated analysis of data from these and other sources, and the subsequent interpretation holds considerable promise for an improved understanding and sensitivity of detection of toxicologic mechanisms. A variety of software tools to facilitate this are in development.

A particular challenge for the application of metabolic profiling technologies is the limited database of normal and affected profiles. Much more work is required to populate this to the level of those supporting other technologies, but the potential value of a comprehensive endogenous metabolite database will be very high. Looking back at the development of more familiar technologies, is interesting to note that haematoxylin and eosin were used in combination to stain tissues for the first time in 1876. Thus, as pathologists, we are operating with the benefit and confidence of a database of more than 130 years vintage. Similarly, has taken 60 to 70 years to provide a panel of ~20 standard clinical pathology assays. Metabolic analysis has been around for fewer than 20 years and has received relatively little investment from industry. As a new technology, has considerable potential, not least because is amenable to noninvasive, serial investigations in both animals and humans. However, the route to becoming an effective mainstream technology used routinely alongside conventional methods will take some considerable time and resources.