- Email: [email protected]
Metabolomics in the pharmaceutical industry
Drug Discovery Today: Technologies
Vol. 13, 2015
Editors-in-Chief Kelvin Lam – Simplex Pharma Advisors, Inc., Boston, MA, USA Henk Timmerman – Vrije Universiteit, The Netherlands DRUG DISCOVERY
TODAY
TECHNOLOGIES
Metabolomics in Medicinal Chemistry
Metabolomics in the pharmaceutical industry Michael D. Reily*, Adrienne A. Tymiak Pharmaceutical Candidate Optimization, Bristol-Myers Squibb Pharmaceutical Co., Princeton, NJ, USA
Metabolomics has roots in the pharmaceutical industry that go back nearly three decades. Initially focused on
Section editor: Pascal de Tullio, University of Lie`ge, Lie`ge, Belgium.
applications in toxicology and disease pathology, more recent academic and commercial efforts have helped advance metabolomics as a tool to reveal the molecular basis of biological processes and pharmacological responses to drugs. This article will discuss areas where
modalities, applications of highest relevance to pharmaceutical research and development, and remaining challenges for metabolomics to deliver on its full potential.
metabolomic technologies and applications are poised to have the greatest impact in the discovery and development of pharmaceuticals.
Introduction In the 1980’s, measurement of endogenous metabolite levels in biological tissues and fluids using NMR enabled pharmaceutical applications, with a particular emphasis on the study of pathological conditions. As a result, early adoption by the industry drove advancement of this field [1–3]. Subsequent streamlining of the pharmaceutical industry led many companies to scale back on dedicated exploratory technology groups, including those supporting metabolomics. At about the same time, government funding for metabolomics research began ramping up globally and while the number of metabolomics publications coming from the industry continue to rise, they are greatly overshadowed by academic-only publications (see Fig. 1). This academic surge is significantly advancing the field, creating new interest in metabolomics as a useful tool for the elucidation of biochemical mechanisms and setting the stage for broader applications in the industry [4]. Herein, we will discuss the predominant analytical *Corresponding author.: M.D. Reily ([email protected]) 1740-6749/$ ß 2015 Elsevier Ltd. All rights reserved.
A multidisciplinary science The mammalian metabolome is comprised of thousands of sub-kiloDalton molecules with physicochemical properties ranging from small highly polar carboxylic acids, amines and amino acids to large neutral lipids. Metabolomic studies measure this breadth of molecules and allow correlation of endogenous biochemical composition with different states of an organism to provide (1) a deeper biochemical understanding of phenotypes, (2) a facile platform for mechanistic hypothesis generation and testing, and (3) potential biomarkers for monitoring safety or efficacy. In the context of most pharmaceutical applications, it is desirable to reliably measure as many endogenous metabolites as possible to maximize the coverage of expected biochemical pathways while gaining insight into any unexpected perturbations [4]. More than just an analytical exercise, a successful metabolomic study requires close coordination amongst a diverse set of experts who are engaged at different times, but have familiarity with all aspects of the study. First, the in vivo study design and execution are critical to ensure collection of high quality samples at the appropriate time points, often using sophisticated equipment (e.g. specialized cages designed to provide refrigerated urine collection) [5]. Second, sample preparation requires attention to detail, to ensure that
http://dx.doi.org/10.1016/j.ddtec.2015.03.001
25
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
Vol. 13, 2015
Pfizer
85
Bristol-Myers Squibb
38
Novartis
30
Merck
26
Eli Lilly Roche
14
Bayer
14
GlaxoSmithkline
All 2000
Pharma
13 11
Novo Nordisk
10
Amgen
9
Servier
9 7
AstraZeneca Pharmacia
1500
1000
5
Takeda
Publications/year (All)
Merck KGaA
4
3
Sanofi
3
Johnson & Johnson
2
DuPont
2
Ono
2
3
5 6
5
4
Abbott
3
CROs
2
2
2
Publications/year (Pharma & CROs)
17
50
40
30
20
5
500
10
27
14 7
6
Publication Year 0 2000
2005
2010
2015
0
Drug Discovery Today: Technologies
Figure 1. Metabolomics publication metrics from 1999 through 2014. The bar chart (upper left) shows the total number of publications for the largest pharmaceutical companies. The line graph (lower right) shows 3 year running averages for annual publications that list authors from all affiliations (blue line), pharmaceutical companies (red line) and contract research organizations that provide metabolomic sample analysis services (green line). Numerals over each red data point indicate the percentage of total metabolomics publications with pharmaceutical industry authorship. These data are based on the following search queries conducted in Scopus: 1) blue line ‘TITLE-ABS-KEY((metabolomics OR metabonomics))’; green line ‘(TITLE-ABSKEY((metabolomics OR metabonomics)) AND AFFILORG ((metabolon OR metanomics OR metabometrix OR biocrates)))’; red line and bar chart ‘(TITLE-ABS-KEY((metabolomics OR metabonomics)) AND AFFILORG ((glaxo OR pfizer OR roche OR nordisk OR dupont OR pharmacia OR astra OR servier OR squibb OR merck OR schering OR johnson OR lilly OR bayer OR boehringer OR sanofi OR novartis OR abbott OR amgen OR takeda OR baxter)))’.
differences between samples are biological in nature, and not due to systematic errors. To minimize the numbers of animals required in a study, peripheral biofluids such as urine, plasma or serum are used most often, but occasionally it is desirable to evaluate tissue extracts. For each target fluid or tissue, simple and robust methods must be developed and standardized procedures followed carefully in order to obtain reliable metabolomic data. Third, generating high quality data, often containing millions of data points per sample, requires a considerable time investment from analytical experts and their instruments. Fourth, these complex data sets must be analyzed in some customized manner since available software tools simply do not provide all the needed functionality. Metabolomic data analysis requires computational and programming experts with a good fundamental understanding of the analytical data and its molecular significance. Fortunately, much of this infrastructure exists within pharma and 26
www.drugdiscoverytoday.com
groups dedicated exclusively to metabolomics are not necessary (nor desirable), assuming that the resources are working together in an effective matrix. Finally, the end user, a biologist, toxicologist or physician, is equally important to a successful outcome, as they possess the most intimate knowledge of the system of interest and are best positioned to interpret the metabolomic results in the context of other study endpoints. While multivariate statistics has its place, we have found that a table of quantitative changes in a comprehensive and specific list of metabolites, essentially an expanded clinical chemistry panel, is often much more useful to the customer than a PCA plot.
Analytical modalities Nuclear magnetic resonance (NMR) and mass spectrometry (MS) have emerged as the best platforms for providing metabolomic information at the molecular level. These techniques
Vol. 13, 2015
stand out by offering comprehensive, annotatable, quantitative and reproducible measurements that are amenable to automation. Since these two modalities measure components of the metabolome via completely distinct biophysical phenomena, they are highly complementary to each other. Furthermore, since a portion of the measured metabolome overlaps between the methods, the use of both offers a comforting degree of redundancy. The scope of this commentary necessitates only a passing mention of these sophisticated modalities, while several recent reviews [4,6–14] and other contributions in this issue provide more detail. As already mentioned, urine and plasma or serum are the most common matrixes for metabolomic evaluation, since these can be obtained in a serial manner without sacrificing the animal under study and are readily translated into the clinic. Although both types of fluids can be analyzed by MS and NMR, it is most expedient to analyze urine by NMR and plasma or serum by MS. Urine osmolality and inorganic salt concentration can vary widely and complicate MS quantitation because of variable ion suppression, while having little effect on NMR response. Conversely, the broad lipoprotein signals in serum or plasma and additional anti-coagulant signals in some plasma samples can complicate small molecule assessments by NMR, even in protein-precipitated samples [15]. To account for the typical variability in dilution expected for urine samples, numerous normalization approaches have been developed for both LC–MS [16] and NMR [4]. Mass spectrometry is currently the predominate analytical platform for metabolomic studies [10,12,17,18] and comprises a family of highly sensitive detection approaches based on the mass of the analyte (or derivatives thereof) coupled with one of a variety of sample introduction methods (e.g. direct sample infusion, liquid or gas chromatography, capillary electrophoresis). This family ranges from highly selective (e.g. a calibrated triple quadrupole system) which can deliver absolute quantitation on tens of metabolite simultaneously, to high resolution accurate mass (HRAMS) systems that are useful for measuring hundreds of components with relative quantitation. The latter are increasing being used in metabolomics studies and can deliver a mass accuracy of a few parts per million (ppm). In conjunction with chromatography and soft ionization techniques such as electrospray ionization (ESI) or atmospheric pressure chemical ionization (APCI), intact molecular ions measured with this level of accuracy are able to dictate specific molecular formulas, thereby greatly aiding analyte annotation. The mass spectrometer continuously counts all ions across a preset range (e.g. 80–1000 Daltons), that are eluting at any given moment. This HRAMS data then consists of an array of millions of data points that must be further processed to measure individual components [19]. Each molecule has a characteristic elution time and ion signature that can be compared with a library of known compounds to aid in automated identification.
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
Currently, only a small percentage of the ions obtained with this approach are assigned, but such data are frequently reinterrogated as new analytes are discovered or new hypotheses are presented for testing. Early work utilizing NMR spectroscopy to study biofluids [20,21] recognized the extremely rich and easily obtainable information contained in a simple high-field proton NMR spectrum. More recent advancements in NMR technology further increased its utility for analyzing whole biological fluids and tissues [11,22]. Many atoms have a nuclear spin and therefore can be monitored using NMR spectroscopy, and hundreds of useful multidimensional experiments to verify molecular structures have been developed [23,24]. However, because proton NMR is highly sensitive and hydrogen is present in nearly every metabolite of interest, simple 1 dimensional (1D) 1H NMR spectroscopy will be the primary source of NMR-based metabolomics data now and into the foreseeable future. Although NMR cannot compete with MS in terms of sensitivity, its advantages include being largely unaffected by matrix, highly reproducible [25] and inherently quantitative. As with mass spectrometry, NMR provides a dataset that can be retrospectively evaluated for molecular entities of interest long after the data have been acquired.
Preclinical applications Ultimately, the goal of preclinical research is to identify new molecular entities (NMEs) predicted to be safe and efficacious in humans. Metabolomics can contribute to this through model characterization, target discovery, target engagement studies, safety testing and biomarker discovery. A fundamental premise of translational medicine is that preclinical research should ‘translate’ into the clinic. While in vitro approaches are leveraged heavily in drug discovery, animal models remain a basic requirement for evaluating NMEs before they are introduced into humans. A drug that works well in an animal model but fails in the clinic represents an enormous opportunity cost. But how well do animal models predict human responses? There are numerous examples in the literature highlighting the caveats with animal models [26] and metabolomics can serve as a bridging tool to compare human outcomes to preclinical models and vice versa. As a simple example, most protocols for animal studies include an overnight fast prior to necropsy. While this might seem innocuous and consistent with clinical situations where humans are also frequently required to fast before providing blood or urine samples, an overnight fast has a much larger impact on the rodent metabolome than that of the human [4]. Similarly, the metabolomic signatures relative to control in liver extracts from humans with nonalcoholic steatohepatitis (NASH) are not in complete concordance with those from a common rat NASH model [27]. We take it for granted that our inbred laboratory animals are all alike. However, www.drugdiscoverytoday.com
27
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
metabolomics successfully identified that rats obtained from adjacent rooms in a large animal supplier clearly had distinct gut microflora with the potential to affect drug tolerability and/or exposure [28,29]. Diabetes and insulin resistance is an area where metabolomics has identified branched chain amino acids (BCAs) and 2-aminoadipic acid (2-AAA) as strong indicators of current or impending disease [30,31]. Results from our laboratory indicate that some animal models mirror these changes, whereas others do not; therefore, a more thorough comparison of existing metabolic syndrome models is still needed. Clearly, these types of findings indicate that a detailed biochemical understanding of preclinical models is crucial for meaningful translational correlations. Another important objective of pharmaceutical discovery is the identification and validation of new targets for therapeutic intervention. While metabolomics is not commonly used for target discovery, it can play a role in identifying downstream or off-target effects of a therapeutic that might have unanticipated biochemical consequences. Since the analytical methods described above collect data in an unbiased fashion, metabolomics can be used to screen for peripheral pathway perturbations and provide mechanistic insights. For example, using metabolomics, it was shown that infecting cultured cells with human cytomegalovirus (CMV) resulted in increased flux through the cell’s acetyl-CoA carboxylase pathway. Inhibition of this enzyme, which is completely unrelated to the virus itself, blocked CMV replication [32]. Although entirely a clinical example, recent metabolomic analysis of human liver samples from patients with non-alcoholic steatohepatitis (NASH) showed changes in lysophospholipids which were consistent with those noted in hepatocelluar carcinoma, revealing a potential mechanistic link in the progression from NASH to carcinoma [33]. The value of this approach will need to be proven by successful clinical development of an intervention of one of these metabolomics-derived targets. Modern pharmaceuticals in development are usually selected to act at specific molecular targets. While definitive proof of the drug’s efficacy may take months or even years of clinical studies, a reliable biomarker of target engagement can build confidence in the short term and streamline clinical development plans. Of course, rational target engagement biomarkers such as substrates or products in the target pathways will continue to be the most valuable, but, there are numerous reasons why these might be difficult to identify or measure. There is an expanding literature on the use of metabolomics to identify potential biomarkers. In this endeavor, it is important to recognize that non-specific biomarkers (‘usual suspects’) can change in response to many physiologic perturbations [34]. Also, while a change in one common metabolite may not be diagnostic, changes in a panel of several metabolites might provide a signature for a specific pathway perturbation. While there are few (if any) 28
www.drugdiscoverytoday.com
Vol. 13, 2015
published examples where metabolomics has been applied for this purpose, it would seem to be well positioned to identify peripheral or indirect biomarkers of target engagement. As previously mentioned, metabolomics was first applied in the pharmaceutical industry for safety assessment [35,36]. The COMET consortium was an early effort by multiple pharmaceutical companies and Imperial College in London to use this new technology to generate a spectral database of biofluids from animals treated with known toxins from which predictive models could be generated [37]. More recent work demonstrates how metabolomics can be used to actually discover preclinical and translational safety biomarkers. In an exploratory study of Cerivastatin-induced myopathy, metabolomics demonstrated the presence of N-acetylated 1- and 3-methylhistidine in the urine of treated rats [38]. While 3-methylhsitidine had been promulgated as a biomarker of muscle turnover for many years [39], the use of methylhistidines as biomarkers of myotoxicity had not yet been explored. Liver P450 enzyme induction is often an impediment to progressing a drug into pharmaceutical development. Metabolomics identified increases in ascorbate and gluonic acid in rats treated with a preclinical compound, which was later shown to be an enzyme inducer. Follow-up work further demonstrated that (1) these two molecules related to ascorbic acid biosynthesis were also elevated in rats treated with known inducers such as Phenobarbital and diallylsulfide and (2) these elevations were correlated with transcriptional changes in enzymes regulating ascorbic acid biosynthesis [40]. These molecules are now useful early indicators of P450 induction. In preclinical studies evaluating a cannabanoid-1 receptor antagonist, dogs developed a drugrelated, microscopically observable lipid accumulation in muscle tissue. Urinary metabolomics revealed a coincident ethylmalonic acid accumulation in the urine, which was subsequently carried into phase 1 clinical trials as a safety biomarker [41]. Xenobiotic metabolites are also part of the metabolome, and these can also be characterized through metabolomic studies. The surprising complexity of the longknown nephrotoxin, 2-bromoethanamine, has been elucidated through MS based stable isotope tracer analysis where some 20 metabolites of this small molecule were identified, some of which were previously unknown [42]. These are just a few of the examples that demonstrate that metabolomics has added and will continue to add value to preclinical pharmaceutical research.
Clinical applications While the greatest impact of metabolomics has been in preclinical pharmaceutical research, numerous intriguing clinical applications are emerging. These include translational research (mentioned above), patient stratification and patient status evaluation. Of course, the metabolome
Vol. 13, 2015
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
N-methylnicotinamide
Glucose
Hippuric acid
0.3 0.04
1
0.2
0.02
0.5 0.1
0 Count Median
0
0 FASTED 15 0.00201
OGTT 15 0.00195
Count Median
FASTED 15 0.08565
OGTT 15 0.09397
Count Median
FASTED 15 0.00825
OGTT 15 0.00844
Drug Discovery Today: Technologies
Figure 2. Metabolic outliers from a clinical study on 15 normal healthy subjects as determined by NMR assessment of urine in the fasted and post-oral glucose challenge states. Boxes indicate upper and lower quartiles, vertical bars indicate upper and lower adjacent values and horizontal bars indicate the median values. The ordinate represents NMR peak spectral areas divided by the peak area of creatinine.
represents the integrated and selective expression and transport of metabolites from all cellular processes, symbiotic and parasitic organisms (e.g. gut microflora and infective microbes), and environmental exposure [43]. This complicates interpretation of metabolomic results from a heterogeneous human population, but also provides an advantage over other ‘omics, which normally only report on the biochemical compartments cascaded down from the host genome. While large epidemiological studies remain the domain of academe, and the ultimate translational impact for metabolomics will be the discovery of novel biomarkers that can be validated for use in clinical trials, direct applications are still of interest in the industry, particularly in early development. Perhaps the most promising area of metabolomic work is in the area of patient stratification, which relies on an individual’s preexisting metabolome. The concept of pharmacometabonomics was first described based on the observation that pre-dose metabolomic measures in rat urine could be used to predict the post-dose outcomes after acetaminophen treatment [44]. This approach was later demonstrated in humans where samples taken shortly after dosing with acetaminophen predicted later hepatic sensitivity [45]. In the context of patient stratification, the challenge is to extract information about the metabolites relevant to how an individual will respond to a given treatment in the presence of a background which is both complex and highly variable [46]. As
mentioned earlier, BCAs and 2-AAA have been reported to precede presentation of diabetes in humans by as much as 12 years [30]. It will be interesting to see if new preventative interventions will be developed based on a simple biomarker measurement taken a decade in advance of disease onset. Additional information that emerges from evaluating patient or healthy volunteer samples prior to initiation of a clinical trial can report on the status of the patient. Phase 1 clinical trials typically recruit healthy volunteers and most Phase 1 and 2 trials have significant restrictions on patient behavior. Previously unpublished metabolomic results from a small clinical trial involving 15 normal healthy volunteers indicated that 4 of the 15 had levels of certain urinary metabolites that differentiated them from others in the study (Fig. 2). In this study, patients were forbidden from taking any co-medications or supplements and urine and plasma samples were taken just before and 2 h after a fasting glucose challenge. In one individual, levels of N-methylnicotinamide (a vitamin b3 metabolite) were 180 fold above the median for the group, suggesting that the subject might not have adhered to the supplements exclusion requirement. In two of the volunteers, hippuric acid levels were 10–15 fold higher than normal, which is likely associated with a different gut microflora or diet in these individuals. Perhaps the most significant observation was in subject 13, where urinary glucose was elevated, especially post challenge. Although postprandial urinary glucose was not a criterion for www.drugdiscoverytoday.com
29
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
exclusion, it is almost certain that this individual is prediabetic or has a genetic predisposition to urinary elimination of glucose. In this small study, 24% of the enrolled patients presented an unusual metabolomic profile and while many of these differences might be benign, exclusion of such metabolic outliers could help avoid complications in interpreting overall study results. How much noise could be eliminated in larger studies were individual data points could be excluded, or at least contextualized, based on observations such as this? While the principle has been demonstrated, solid clinical evidence that metabolomics-based patient stratification in the evaluation of new candidate drugs is still lacking. Hopefully, we will see such evidence in the next few years.
Biologics production applications A non-biomedical application of metabolomics that has been embraced by the industry is in monitoring the many supplied and excreted small molecules as genetically engineered cells are cultured in bioreactors to produce protein therapeutics. Unlike conventional biochemical assays, a metabolomics approach allows the simultaneous quantification of many metabolites related to cell growth and viability. In practice, observed depletion of an essential nutrient can trigger media augmentation, or excessive build-up of deleterious metabolites may serve as an early indicator of a failing culture. Both NMR [47,48] and mass spectrometry [49] based metabolomic techniques have been developed for this purpose. The effect of dextran sulfate, a commonly used anti-aggregation agent, and volume of the bioreactor were shown to impact specific media nutrients and cellular components [50] and recently, the biochemical reprogramming of two different variants of CHO cells by temperature shift has been reported [51]. This application of metabolomics avoids many of the pitfalls of studying whole animals, since there are fewer concurrent processes and nutrient influx and efflux can be more closely controlled and measured. When compared to the few selected metabolites and parameters that are traditionally monitored in a protein production setting, metabolomics can dramatically increase the information available on the biochemical changes occurring within the culture.
Conclusions and outlook From this discussion, it is clear that metabolomics has demonstrated utility in the pharmaceutical industry. Examples of how metabolomic findings can help elucidate disease mechanism, reveal novel targets for therapeutic intervention, and serve to identify biomarkers are convincing in some cases and need more testing in others. It is not difficult to understand why metabolomics efforts have waxed and waned in industrial settings while they continue to flourish in academia. Like many emerging technologies, the promise of metabolomics may have been overestimated in its early days. Initial ideas that conventional toxicity endpoints could be replaced with 30
www.drugdiscoverytoday.com
Vol. 13, 2015
multivariate mathematical models based on metabolomic spectral data have now been largely supplanted with the understanding that direct and detailed biochemical information is actually the most useful product of metabolomic studies. Furthermore, the availability of commercial outsourcing options may obviate the need for large in-house analytical efforts. The biggest impediments to fully realizing the potential of metabolomics in our industry involve the limited breath of coverage and our ability to annotate analytical signals. Many of the unbiased metabolomic profiling techniques, while impressive in breadth, still only sample a small fraction of the metabolome and usually provide relative concentrations of metabolites. Such untargeted metabolomics techniques currently produce more data on un-annotated signals than meaningful information on assigned metabolites, so methods for rapid identification of the remaining unassigned signals is desirable. While these and other issues remain to be solved, a number of developments in the past few years’ foster optimism that metabolomics will continue to be useful throughout all phases of pharmaceutical research and development.
Conflict of interest The author(s) have no conflict of interest to declare.
References 1. Nicholson JK, Wilson ID. High resolution nuclear magnetic resonance spectroscopy of biological samples as an aid to drug development. Progr Drug Res 1987;31:427–79. 2. Lindon JC, Nicholson JK, Everett JR. NMR spectroscopy of biofluids. Ann Rep NMR Spect 1999;38:1–88. 3. Robertson DG. Metabonomics in toxicology: a review. Toxicol Sci 2005;85:809–22. 4. Robertson DG, Reily MD. The current status of metabolomics in drug discovery and development. Drug Dev Res 2012;73:535–46. 5. Robertson DG, Reily MD, Lindon JC, Holmes E, Nicholson JK. Metabonomic technology as a tool for rapid throughput in vivo toxicity screening. In: Vanden Heuvel JP, Perdew GJ, Mattes WB, Greenlee WF, editors. Comprehensive toxicology. Elsevier Science BV; 2002. p. 583–610. 6. Reily MD, Lindon JC. NMR spectroscopy: principles and instrumentation. In: Robertson JCL, Holmes E, Nicholson JK, editors. Metabonomics in toxicity assessment. Taylor & Francis; 2005. p. 75–104. 7. Lindon JC, Holmes E, Bollard ME, Stanley EG, Nicholson JK. Metabonomics technologies and their applications in physiological monitoring, drug safety assessment and disease diagnosis. Biomarkers 2004;9:1–31. 8. Lindon JC, Nicholson JK, Everett JR. NMR spectroscopy of biofluids. Annu Rep NMR Spectr 1999;38:1–88. 9. Wishart DS. Advances in metabolite identification. Bioanalysis 2011;3:1769–82. 10. Theodoridis G, Gika HG, Wilson ID. Mass spectrometry-based holistic analytical approaches for metabolite profiling in systems biology studies. Mass Spectr Rev 2011;30:884–906. 11. Dunn WB, Broadhurst DI, Atherton HJ, Goodacre R, Griffin JL. Systems level studies of mammalian metabolomes: the roles of mass spectrometry and nuclear magnetic resonance spectroscopy. Chem Soc Rev 2011;40:387–426. 12. Want EJ, Nordstro¨m A, Morita H, Siuzdak G. From exogenous to endogenous: the inevitable imprint of mass spectrometry in metabolomics. J Proteome Res 2007;6:459–68.
Vol. 13, 2015
13. Lindon JC, Holmes E, Nicholson JK. Metabonomics techniques and applications to pharmaceutical research & development. Pharma Res 2006;23:1075–88. 14. Fiehn O. Combining genomics, metabolome analysis, and biochemical modelling to understand metabolic networks. Comp Funct Genomics 2001;2:155–68. 15. Reily MD, Shipkova P, Hnatyshyn S. LC–MS in endogenous metabolite profiling and small molecule biomarker discovery. In: Lee MS, Zhu M, editors. Mass spectrometry in drug metabolism and disposition: basic principles and applications. John Wiley; 2011. 16. Warrack BM, Hnatyshyn S, Ott KH, Reily MD, Sanders M, Zhang H, et al. Normalization strategies for metabonomic analysis of urine samples. J Chromatogr B: Anal Technol Biomed Life Sci 2009;877(5–6):547–52. 17. van der Greef J, van der Heijden R, Verheij ER. The role of mass spectrometry in systems biology: data processing and identification strategies in metabolomics. Adv Mass Spectr 2004;16:145–65. 18. Fiehn O. Combining genomics, metabolome analysis, and biochemical modeling to understand metabolic networks. Comp Func Genom 2001;2:155–68. 19. Hnatyshyn S, Shipkova P. Automated and unbiased analysis of LC–MS metabolomic data. Bioanalysis 2012;4(5):541–54. 20. Nicholson JK, Buckingham MJ, Sadler PJ. High resolution proton NMR studies of vertebrate blood and plasma. Biochem J 1983;211(3):605–15. 21. Bales JR, Higham DP, Howe I, Nicholson JK, Sadler PJ. Use of highresolution proton nuclear magnetic resonance spectroscopy for rapid multi-component analysis of urine. Clin Chem 1984;30(3):426–32. 22. Nicholson JK, Connelly J, Lindon JC, Holmes E. Metabonomics: a platform for studying drug toxicity and gene function. Review 86 refs. Drug Discov 2002;1(2):153–61. Nature Reviews. 23. Chylla RA, Hu K, Ellinger JJ, Markley JL. Deconvolution of twodimensional NMR spectra by fast maximum likelihood reconstruction: application to quantitative metabolomics. Anal Chem 2011;83(12): 4871–80. 24. Reily MD, Lindon JC. NMR spectroscopy: principles and instrumentation. In: Robertson JCL, Holmes E, Nicholson JK, editors. Metabonomics in safety assessment. Taylor & Francis; 2005. 25. Keun HC, Ebbels TMD, Antti H, Bollard ME, Beckonert O, Schlotterbeck G, et al. Analytical reproducibility in 1H NMR-based metabonomic urinalysis. Chem Res Toxicol 2002;15(11):1380–6. 26. Shanks N, Greek R, Greek J. Are animal models predictive for humans? Philos Ethics Hum Med 2009;4(1). 27. Robertson DG, Frevert U. Metabolomics in drug discovery and development. Clin Pharmacol Therap 2013;94(5):559–61. 28. Robosky LC, Wells DF, Egnash LA, Manning ML, Reily MD, Robertson DG. Communication regarding metabonomic identification of two distinct phenotypes in Sprague-Dawley (Crl:CD(SD)) rats. Toxicol Sci 2006;91(1):p309. 29. Robosky LC, Wells DF, Egnash LA, Manning ML, Reily MD, Robertson DG. Metabonomic identification of two distinct phenotypes in SpragueDawley (Crl:CD(SD)) rats. Toxicol Sci 2005;87(1):277–84. 30. Wang TJ, Ngo D, Psychogios N, Dejam A, Larson MG, Vasan RS, et al. 2Aminoadipic acid is a biomarker for diabetes risk. J Clin Invest 2013;123(10):4309–17. 31. Newgard CB, An J, Bain JR, Muehlbauer MJ, Stevens RD, Lien LF, et al. A branched-chain amino acid-related metabolic signature that differentiates obese and lean humans and contributes to insulin resistance. Cell Metab 2009;9(4):311–26. 32. Rabinowitz JD, Purdy JG, Vastag L, Shenk T, Koyuncu E. Metabolomics in drug target discovery. Cold Spring Harbor Symp Quant Biol 2011;76: 235–46. 33. Clarke JD, Novak P, Lake AD, Shipkova P, Aranibar N, Robertson D, et al. Characterization of hepatocellular carcinoma-related genes and
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
34. 35.
36.
37.
38.
39.
40.
41.
42.
43. 44.
45.
46.
47.
48.
49.
50.
51.
metabolites in human nonalcoholic fatty liver disease. Toxicol Sci 2012;126(1):p429. Robertson DG. Metabonomics in toxicology: a review. Toxicol Sci 2005;85(2):809–22. Robertson DG, Reily MD, Sigler RE, Wells DF, Paterson DA, Braden TK. Metabonomics: evaluation of nuclear magnetic resonance (NMR) and pattern recognition technology for rapid in vivo screening of liver and kidney toxicants. Toxicol Sci 2000;57(2):326–37. Lindon JC, Holmes E, Nicholson JK. Metabonomics: systems biology in pharmaceutical research and development. Curr Opin Mol Ther 2004;6(3):265–72. Lindon JC, Keun HC, Ebbels TMD, Pearce JMT, Holmes E, Nicholson JK. The Consortium for Metabonomic Toxicology (COMET): aims, activities and achievements. Pharmacogenomics 2005;6(7):691–9. Aranibar N, Vassallo JD, Rathmacher J, Stryker S, Zhang Y, Dai J, et al. Identification of 1- and 3-methylhistidine as biomarkers of skeletal muscle toxicity by nuclear magnetic resonance-based metabolic profiling. Anal Biochem 2010;410(1):84–91. Young VR, Haverberg LN, Bilmazes C, Munro HN. Potential use of 3methylhistidine excretion as an index of progressive reduction in muscle protein catabolism during starvation. Metabolism 1973;23(2):1429–36. Aranibar N, Bhaskaran V, Ott KH, Vassallo J, Nelson D, Lecureux L, et al. Modulation of ascorbic acid metabolism by cytochrome P450 induction revealed by metabonomics and transcriptional profiling. Magn Res Chem 2009;47(Suppl. 1):S12–9. Tomlinson L, Tirmenstein MA, Janovitz EB, Aranibar N, Ott KH, Kozlosky JC, et al. Cannabinoid receptor antagonist-induced striated muscle toxicity and ethylmalonic-adipic aciduria in beagle dogs. Toxicol Sci 2012;129(2):268–79. Shipkova P, Vassallo JD, Aranibar N, Hnatyshyn S, Zhang H, Clayton TA, et al. Urinary metabolites of 2-bromoethanamine identified by stable isotope labelling: Evidence for carbamoylation and glutathione conjugation. Xenobiotica 2011;41(2):144–54. Nicholson JK. Global systems biology, personalized medicine and molecular epidemiology. Mol Syst Biol 2006;2. Clayton TA, Lindon JC, Cloarec O, Antti H, Charuel C, Hanton G, et al. Pharmaco-metabonomic phenotyping and personalized drug treatment. Nature 2006;440(7087):1073–7. Winnike JH, Li Z, Wright FA, MacDonald JM, O’Connell TM, Watkins PB. Use of pharmaco-metabonomics for early prediction of acetaminopheninduced hepatotoxicity in humans. Clin Pharmacol Therap 2010;88(1): 45–51. Jackson J, Townsend R, Reily M, Soares H, Averbuch S. Scientific domains of personalized medicine. In: Hawana J, Runkle D, editors. Personalized medicine: prescriptions and prospects. The Food and Drug Law Institute; 2011. p. 1–16. Aranibar N, Reily MD. NMR methods for metabolomics of mammalian cell culture bioreactors. In: Portner R, editor. Methods in molecular biology. 2014. p. 223–36. Bradley SA, Ouyang A, Purdie J, Smitka TA, Wang T, Kaerner A. Fermentanomics: monitoring mammalian cell cultures with NMR spectroscopy. J Am Chem Soc 2010;132(28):9531–3. Chong WPK, Goh LT, Reddy SG, Yusufi FNK, Lee DY, Wong NSC, et al. Metabolomics profiling of extracellular metabolites in recombinant Chinese Hamster Ovary fed-batch culture. Rapid Commun Mass Spectr 2009;23(23):3763–71. Aranibar N, Borys M, Mackin NA, Ly V, Abu-Absi N, Abu-Absi S, et al. NMRbased metabolomics of mammalian cell and tissue cultures. J Biomol NMR 2011;49(3–4):195–206. Wagstaff JL, et al. Masterton RJ, Povey JF, Smales CM, Howard MJ. 1H NMR spectroscopy profiling of metabolic reprogramming of Chinese hamster ovary cells upon a temperature shift during culture. PLoS ONE 2013;8(10).
www.drugdiscoverytoday.com
31
Vol. 13, 2015
Editors-in-Chief Kelvin Lam – Simplex Pharma Advisors, Inc., Boston, MA, USA Henk Timmerman – Vrije Universiteit, The Netherlands DRUG DISCOVERY
TODAY
TECHNOLOGIES
Metabolomics in Medicinal Chemistry
Metabolomics in the pharmaceutical industry Michael D. Reily*, Adrienne A. Tymiak Pharmaceutical Candidate Optimization, Bristol-Myers Squibb Pharmaceutical Co., Princeton, NJ, USA
Metabolomics has roots in the pharmaceutical industry that go back nearly three decades. Initially focused on
Section editor: Pascal de Tullio, University of Lie`ge, Lie`ge, Belgium.
applications in toxicology and disease pathology, more recent academic and commercial efforts have helped advance metabolomics as a tool to reveal the molecular basis of biological processes and pharmacological responses to drugs. This article will discuss areas where
modalities, applications of highest relevance to pharmaceutical research and development, and remaining challenges for metabolomics to deliver on its full potential.
metabolomic technologies and applications are poised to have the greatest impact in the discovery and development of pharmaceuticals.
Introduction In the 1980’s, measurement of endogenous metabolite levels in biological tissues and fluids using NMR enabled pharmaceutical applications, with a particular emphasis on the study of pathological conditions. As a result, early adoption by the industry drove advancement of this field [1–3]. Subsequent streamlining of the pharmaceutical industry led many companies to scale back on dedicated exploratory technology groups, including those supporting metabolomics. At about the same time, government funding for metabolomics research began ramping up globally and while the number of metabolomics publications coming from the industry continue to rise, they are greatly overshadowed by academic-only publications (see Fig. 1). This academic surge is significantly advancing the field, creating new interest in metabolomics as a useful tool for the elucidation of biochemical mechanisms and setting the stage for broader applications in the industry [4]. Herein, we will discuss the predominant analytical *Corresponding author.: M.D. Reily ([email protected]) 1740-6749/$ ß 2015 Elsevier Ltd. All rights reserved.
A multidisciplinary science The mammalian metabolome is comprised of thousands of sub-kiloDalton molecules with physicochemical properties ranging from small highly polar carboxylic acids, amines and amino acids to large neutral lipids. Metabolomic studies measure this breadth of molecules and allow correlation of endogenous biochemical composition with different states of an organism to provide (1) a deeper biochemical understanding of phenotypes, (2) a facile platform for mechanistic hypothesis generation and testing, and (3) potential biomarkers for monitoring safety or efficacy. In the context of most pharmaceutical applications, it is desirable to reliably measure as many endogenous metabolites as possible to maximize the coverage of expected biochemical pathways while gaining insight into any unexpected perturbations [4]. More than just an analytical exercise, a successful metabolomic study requires close coordination amongst a diverse set of experts who are engaged at different times, but have familiarity with all aspects of the study. First, the in vivo study design and execution are critical to ensure collection of high quality samples at the appropriate time points, often using sophisticated equipment (e.g. specialized cages designed to provide refrigerated urine collection) [5]. Second, sample preparation requires attention to detail, to ensure that
http://dx.doi.org/10.1016/j.ddtec.2015.03.001
25
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
Vol. 13, 2015
Pfizer
85
Bristol-Myers Squibb
38
Novartis
30
Merck
26
Eli Lilly Roche
14
Bayer
14
GlaxoSmithkline
All 2000
Pharma
13 11
Novo Nordisk
10
Amgen
9
Servier
9 7
AstraZeneca Pharmacia
1500
1000
5
Takeda
Publications/year (All)
Merck KGaA
4
3
Sanofi
3
Johnson & Johnson
2
DuPont
2
Ono
2
3
5 6
5
4
Abbott
3
CROs
2
2
2
Publications/year (Pharma & CROs)
17
50
40
30
20
5
500
10
27
14 7
6
Publication Year 0 2000
2005
2010
2015
0
Drug Discovery Today: Technologies
Figure 1. Metabolomics publication metrics from 1999 through 2014. The bar chart (upper left) shows the total number of publications for the largest pharmaceutical companies. The line graph (lower right) shows 3 year running averages for annual publications that list authors from all affiliations (blue line), pharmaceutical companies (red line) and contract research organizations that provide metabolomic sample analysis services (green line). Numerals over each red data point indicate the percentage of total metabolomics publications with pharmaceutical industry authorship. These data are based on the following search queries conducted in Scopus: 1) blue line ‘TITLE-ABS-KEY((metabolomics OR metabonomics))’; green line ‘(TITLE-ABSKEY((metabolomics OR metabonomics)) AND AFFILORG ((metabolon OR metanomics OR metabometrix OR biocrates)))’; red line and bar chart ‘(TITLE-ABS-KEY((metabolomics OR metabonomics)) AND AFFILORG ((glaxo OR pfizer OR roche OR nordisk OR dupont OR pharmacia OR astra OR servier OR squibb OR merck OR schering OR johnson OR lilly OR bayer OR boehringer OR sanofi OR novartis OR abbott OR amgen OR takeda OR baxter)))’.
differences between samples are biological in nature, and not due to systematic errors. To minimize the numbers of animals required in a study, peripheral biofluids such as urine, plasma or serum are used most often, but occasionally it is desirable to evaluate tissue extracts. For each target fluid or tissue, simple and robust methods must be developed and standardized procedures followed carefully in order to obtain reliable metabolomic data. Third, generating high quality data, often containing millions of data points per sample, requires a considerable time investment from analytical experts and their instruments. Fourth, these complex data sets must be analyzed in some customized manner since available software tools simply do not provide all the needed functionality. Metabolomic data analysis requires computational and programming experts with a good fundamental understanding of the analytical data and its molecular significance. Fortunately, much of this infrastructure exists within pharma and 26
www.drugdiscoverytoday.com
groups dedicated exclusively to metabolomics are not necessary (nor desirable), assuming that the resources are working together in an effective matrix. Finally, the end user, a biologist, toxicologist or physician, is equally important to a successful outcome, as they possess the most intimate knowledge of the system of interest and are best positioned to interpret the metabolomic results in the context of other study endpoints. While multivariate statistics has its place, we have found that a table of quantitative changes in a comprehensive and specific list of metabolites, essentially an expanded clinical chemistry panel, is often much more useful to the customer than a PCA plot.
Analytical modalities Nuclear magnetic resonance (NMR) and mass spectrometry (MS) have emerged as the best platforms for providing metabolomic information at the molecular level. These techniques
Vol. 13, 2015
stand out by offering comprehensive, annotatable, quantitative and reproducible measurements that are amenable to automation. Since these two modalities measure components of the metabolome via completely distinct biophysical phenomena, they are highly complementary to each other. Furthermore, since a portion of the measured metabolome overlaps between the methods, the use of both offers a comforting degree of redundancy. The scope of this commentary necessitates only a passing mention of these sophisticated modalities, while several recent reviews [4,6–14] and other contributions in this issue provide more detail. As already mentioned, urine and plasma or serum are the most common matrixes for metabolomic evaluation, since these can be obtained in a serial manner without sacrificing the animal under study and are readily translated into the clinic. Although both types of fluids can be analyzed by MS and NMR, it is most expedient to analyze urine by NMR and plasma or serum by MS. Urine osmolality and inorganic salt concentration can vary widely and complicate MS quantitation because of variable ion suppression, while having little effect on NMR response. Conversely, the broad lipoprotein signals in serum or plasma and additional anti-coagulant signals in some plasma samples can complicate small molecule assessments by NMR, even in protein-precipitated samples [15]. To account for the typical variability in dilution expected for urine samples, numerous normalization approaches have been developed for both LC–MS [16] and NMR [4]. Mass spectrometry is currently the predominate analytical platform for metabolomic studies [10,12,17,18] and comprises a family of highly sensitive detection approaches based on the mass of the analyte (or derivatives thereof) coupled with one of a variety of sample introduction methods (e.g. direct sample infusion, liquid or gas chromatography, capillary electrophoresis). This family ranges from highly selective (e.g. a calibrated triple quadrupole system) which can deliver absolute quantitation on tens of metabolite simultaneously, to high resolution accurate mass (HRAMS) systems that are useful for measuring hundreds of components with relative quantitation. The latter are increasing being used in metabolomics studies and can deliver a mass accuracy of a few parts per million (ppm). In conjunction with chromatography and soft ionization techniques such as electrospray ionization (ESI) or atmospheric pressure chemical ionization (APCI), intact molecular ions measured with this level of accuracy are able to dictate specific molecular formulas, thereby greatly aiding analyte annotation. The mass spectrometer continuously counts all ions across a preset range (e.g. 80–1000 Daltons), that are eluting at any given moment. This HRAMS data then consists of an array of millions of data points that must be further processed to measure individual components [19]. Each molecule has a characteristic elution time and ion signature that can be compared with a library of known compounds to aid in automated identification.
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
Currently, only a small percentage of the ions obtained with this approach are assigned, but such data are frequently reinterrogated as new analytes are discovered or new hypotheses are presented for testing. Early work utilizing NMR spectroscopy to study biofluids [20,21] recognized the extremely rich and easily obtainable information contained in a simple high-field proton NMR spectrum. More recent advancements in NMR technology further increased its utility for analyzing whole biological fluids and tissues [11,22]. Many atoms have a nuclear spin and therefore can be monitored using NMR spectroscopy, and hundreds of useful multidimensional experiments to verify molecular structures have been developed [23,24]. However, because proton NMR is highly sensitive and hydrogen is present in nearly every metabolite of interest, simple 1 dimensional (1D) 1H NMR spectroscopy will be the primary source of NMR-based metabolomics data now and into the foreseeable future. Although NMR cannot compete with MS in terms of sensitivity, its advantages include being largely unaffected by matrix, highly reproducible [25] and inherently quantitative. As with mass spectrometry, NMR provides a dataset that can be retrospectively evaluated for molecular entities of interest long after the data have been acquired.
Preclinical applications Ultimately, the goal of preclinical research is to identify new molecular entities (NMEs) predicted to be safe and efficacious in humans. Metabolomics can contribute to this through model characterization, target discovery, target engagement studies, safety testing and biomarker discovery. A fundamental premise of translational medicine is that preclinical research should ‘translate’ into the clinic. While in vitro approaches are leveraged heavily in drug discovery, animal models remain a basic requirement for evaluating NMEs before they are introduced into humans. A drug that works well in an animal model but fails in the clinic represents an enormous opportunity cost. But how well do animal models predict human responses? There are numerous examples in the literature highlighting the caveats with animal models [26] and metabolomics can serve as a bridging tool to compare human outcomes to preclinical models and vice versa. As a simple example, most protocols for animal studies include an overnight fast prior to necropsy. While this might seem innocuous and consistent with clinical situations where humans are also frequently required to fast before providing blood or urine samples, an overnight fast has a much larger impact on the rodent metabolome than that of the human [4]. Similarly, the metabolomic signatures relative to control in liver extracts from humans with nonalcoholic steatohepatitis (NASH) are not in complete concordance with those from a common rat NASH model [27]. We take it for granted that our inbred laboratory animals are all alike. However, www.drugdiscoverytoday.com
27
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
metabolomics successfully identified that rats obtained from adjacent rooms in a large animal supplier clearly had distinct gut microflora with the potential to affect drug tolerability and/or exposure [28,29]. Diabetes and insulin resistance is an area where metabolomics has identified branched chain amino acids (BCAs) and 2-aminoadipic acid (2-AAA) as strong indicators of current or impending disease [30,31]. Results from our laboratory indicate that some animal models mirror these changes, whereas others do not; therefore, a more thorough comparison of existing metabolic syndrome models is still needed. Clearly, these types of findings indicate that a detailed biochemical understanding of preclinical models is crucial for meaningful translational correlations. Another important objective of pharmaceutical discovery is the identification and validation of new targets for therapeutic intervention. While metabolomics is not commonly used for target discovery, it can play a role in identifying downstream or off-target effects of a therapeutic that might have unanticipated biochemical consequences. Since the analytical methods described above collect data in an unbiased fashion, metabolomics can be used to screen for peripheral pathway perturbations and provide mechanistic insights. For example, using metabolomics, it was shown that infecting cultured cells with human cytomegalovirus (CMV) resulted in increased flux through the cell’s acetyl-CoA carboxylase pathway. Inhibition of this enzyme, which is completely unrelated to the virus itself, blocked CMV replication [32]. Although entirely a clinical example, recent metabolomic analysis of human liver samples from patients with non-alcoholic steatohepatitis (NASH) showed changes in lysophospholipids which were consistent with those noted in hepatocelluar carcinoma, revealing a potential mechanistic link in the progression from NASH to carcinoma [33]. The value of this approach will need to be proven by successful clinical development of an intervention of one of these metabolomics-derived targets. Modern pharmaceuticals in development are usually selected to act at specific molecular targets. While definitive proof of the drug’s efficacy may take months or even years of clinical studies, a reliable biomarker of target engagement can build confidence in the short term and streamline clinical development plans. Of course, rational target engagement biomarkers such as substrates or products in the target pathways will continue to be the most valuable, but, there are numerous reasons why these might be difficult to identify or measure. There is an expanding literature on the use of metabolomics to identify potential biomarkers. In this endeavor, it is important to recognize that non-specific biomarkers (‘usual suspects’) can change in response to many physiologic perturbations [34]. Also, while a change in one common metabolite may not be diagnostic, changes in a panel of several metabolites might provide a signature for a specific pathway perturbation. While there are few (if any) 28
www.drugdiscoverytoday.com
Vol. 13, 2015
published examples where metabolomics has been applied for this purpose, it would seem to be well positioned to identify peripheral or indirect biomarkers of target engagement. As previously mentioned, metabolomics was first applied in the pharmaceutical industry for safety assessment [35,36]. The COMET consortium was an early effort by multiple pharmaceutical companies and Imperial College in London to use this new technology to generate a spectral database of biofluids from animals treated with known toxins from which predictive models could be generated [37]. More recent work demonstrates how metabolomics can be used to actually discover preclinical and translational safety biomarkers. In an exploratory study of Cerivastatin-induced myopathy, metabolomics demonstrated the presence of N-acetylated 1- and 3-methylhistidine in the urine of treated rats [38]. While 3-methylhsitidine had been promulgated as a biomarker of muscle turnover for many years [39], the use of methylhistidines as biomarkers of myotoxicity had not yet been explored. Liver P450 enzyme induction is often an impediment to progressing a drug into pharmaceutical development. Metabolomics identified increases in ascorbate and gluonic acid in rats treated with a preclinical compound, which was later shown to be an enzyme inducer. Follow-up work further demonstrated that (1) these two molecules related to ascorbic acid biosynthesis were also elevated in rats treated with known inducers such as Phenobarbital and diallylsulfide and (2) these elevations were correlated with transcriptional changes in enzymes regulating ascorbic acid biosynthesis [40]. These molecules are now useful early indicators of P450 induction. In preclinical studies evaluating a cannabanoid-1 receptor antagonist, dogs developed a drugrelated, microscopically observable lipid accumulation in muscle tissue. Urinary metabolomics revealed a coincident ethylmalonic acid accumulation in the urine, which was subsequently carried into phase 1 clinical trials as a safety biomarker [41]. Xenobiotic metabolites are also part of the metabolome, and these can also be characterized through metabolomic studies. The surprising complexity of the longknown nephrotoxin, 2-bromoethanamine, has been elucidated through MS based stable isotope tracer analysis where some 20 metabolites of this small molecule were identified, some of which were previously unknown [42]. These are just a few of the examples that demonstrate that metabolomics has added and will continue to add value to preclinical pharmaceutical research.
Clinical applications While the greatest impact of metabolomics has been in preclinical pharmaceutical research, numerous intriguing clinical applications are emerging. These include translational research (mentioned above), patient stratification and patient status evaluation. Of course, the metabolome
Vol. 13, 2015
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
N-methylnicotinamide
Glucose
Hippuric acid
0.3 0.04
1
0.2
0.02
0.5 0.1
0 Count Median
0
0 FASTED 15 0.00201
OGTT 15 0.00195
Count Median
FASTED 15 0.08565
OGTT 15 0.09397
Count Median
FASTED 15 0.00825
OGTT 15 0.00844
Drug Discovery Today: Technologies
Figure 2. Metabolic outliers from a clinical study on 15 normal healthy subjects as determined by NMR assessment of urine in the fasted and post-oral glucose challenge states. Boxes indicate upper and lower quartiles, vertical bars indicate upper and lower adjacent values and horizontal bars indicate the median values. The ordinate represents NMR peak spectral areas divided by the peak area of creatinine.
represents the integrated and selective expression and transport of metabolites from all cellular processes, symbiotic and parasitic organisms (e.g. gut microflora and infective microbes), and environmental exposure [43]. This complicates interpretation of metabolomic results from a heterogeneous human population, but also provides an advantage over other ‘omics, which normally only report on the biochemical compartments cascaded down from the host genome. While large epidemiological studies remain the domain of academe, and the ultimate translational impact for metabolomics will be the discovery of novel biomarkers that can be validated for use in clinical trials, direct applications are still of interest in the industry, particularly in early development. Perhaps the most promising area of metabolomic work is in the area of patient stratification, which relies on an individual’s preexisting metabolome. The concept of pharmacometabonomics was first described based on the observation that pre-dose metabolomic measures in rat urine could be used to predict the post-dose outcomes after acetaminophen treatment [44]. This approach was later demonstrated in humans where samples taken shortly after dosing with acetaminophen predicted later hepatic sensitivity [45]. In the context of patient stratification, the challenge is to extract information about the metabolites relevant to how an individual will respond to a given treatment in the presence of a background which is both complex and highly variable [46]. As
mentioned earlier, BCAs and 2-AAA have been reported to precede presentation of diabetes in humans by as much as 12 years [30]. It will be interesting to see if new preventative interventions will be developed based on a simple biomarker measurement taken a decade in advance of disease onset. Additional information that emerges from evaluating patient or healthy volunteer samples prior to initiation of a clinical trial can report on the status of the patient. Phase 1 clinical trials typically recruit healthy volunteers and most Phase 1 and 2 trials have significant restrictions on patient behavior. Previously unpublished metabolomic results from a small clinical trial involving 15 normal healthy volunteers indicated that 4 of the 15 had levels of certain urinary metabolites that differentiated them from others in the study (Fig. 2). In this study, patients were forbidden from taking any co-medications or supplements and urine and plasma samples were taken just before and 2 h after a fasting glucose challenge. In one individual, levels of N-methylnicotinamide (a vitamin b3 metabolite) were 180 fold above the median for the group, suggesting that the subject might not have adhered to the supplements exclusion requirement. In two of the volunteers, hippuric acid levels were 10–15 fold higher than normal, which is likely associated with a different gut microflora or diet in these individuals. Perhaps the most significant observation was in subject 13, where urinary glucose was elevated, especially post challenge. Although postprandial urinary glucose was not a criterion for www.drugdiscoverytoday.com
29
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
exclusion, it is almost certain that this individual is prediabetic or has a genetic predisposition to urinary elimination of glucose. In this small study, 24% of the enrolled patients presented an unusual metabolomic profile and while many of these differences might be benign, exclusion of such metabolic outliers could help avoid complications in interpreting overall study results. How much noise could be eliminated in larger studies were individual data points could be excluded, or at least contextualized, based on observations such as this? While the principle has been demonstrated, solid clinical evidence that metabolomics-based patient stratification in the evaluation of new candidate drugs is still lacking. Hopefully, we will see such evidence in the next few years.
Biologics production applications A non-biomedical application of metabolomics that has been embraced by the industry is in monitoring the many supplied and excreted small molecules as genetically engineered cells are cultured in bioreactors to produce protein therapeutics. Unlike conventional biochemical assays, a metabolomics approach allows the simultaneous quantification of many metabolites related to cell growth and viability. In practice, observed depletion of an essential nutrient can trigger media augmentation, or excessive build-up of deleterious metabolites may serve as an early indicator of a failing culture. Both NMR [47,48] and mass spectrometry [49] based metabolomic techniques have been developed for this purpose. The effect of dextran sulfate, a commonly used anti-aggregation agent, and volume of the bioreactor were shown to impact specific media nutrients and cellular components [50] and recently, the biochemical reprogramming of two different variants of CHO cells by temperature shift has been reported [51]. This application of metabolomics avoids many of the pitfalls of studying whole animals, since there are fewer concurrent processes and nutrient influx and efflux can be more closely controlled and measured. When compared to the few selected metabolites and parameters that are traditionally monitored in a protein production setting, metabolomics can dramatically increase the information available on the biochemical changes occurring within the culture.
Conclusions and outlook From this discussion, it is clear that metabolomics has demonstrated utility in the pharmaceutical industry. Examples of how metabolomic findings can help elucidate disease mechanism, reveal novel targets for therapeutic intervention, and serve to identify biomarkers are convincing in some cases and need more testing in others. It is not difficult to understand why metabolomics efforts have waxed and waned in industrial settings while they continue to flourish in academia. Like many emerging technologies, the promise of metabolomics may have been overestimated in its early days. Initial ideas that conventional toxicity endpoints could be replaced with 30
www.drugdiscoverytoday.com
Vol. 13, 2015
multivariate mathematical models based on metabolomic spectral data have now been largely supplanted with the understanding that direct and detailed biochemical information is actually the most useful product of metabolomic studies. Furthermore, the availability of commercial outsourcing options may obviate the need for large in-house analytical efforts. The biggest impediments to fully realizing the potential of metabolomics in our industry involve the limited breath of coverage and our ability to annotate analytical signals. Many of the unbiased metabolomic profiling techniques, while impressive in breadth, still only sample a small fraction of the metabolome and usually provide relative concentrations of metabolites. Such untargeted metabolomics techniques currently produce more data on un-annotated signals than meaningful information on assigned metabolites, so methods for rapid identification of the remaining unassigned signals is desirable. While these and other issues remain to be solved, a number of developments in the past few years’ foster optimism that metabolomics will continue to be useful throughout all phases of pharmaceutical research and development.
Conflict of interest The author(s) have no conflict of interest to declare.
References 1. Nicholson JK, Wilson ID. High resolution nuclear magnetic resonance spectroscopy of biological samples as an aid to drug development. Progr Drug Res 1987;31:427–79. 2. Lindon JC, Nicholson JK, Everett JR. NMR spectroscopy of biofluids. Ann Rep NMR Spect 1999;38:1–88. 3. Robertson DG. Metabonomics in toxicology: a review. Toxicol Sci 2005;85:809–22. 4. Robertson DG, Reily MD. The current status of metabolomics in drug discovery and development. Drug Dev Res 2012;73:535–46. 5. Robertson DG, Reily MD, Lindon JC, Holmes E, Nicholson JK. Metabonomic technology as a tool for rapid throughput in vivo toxicity screening. In: Vanden Heuvel JP, Perdew GJ, Mattes WB, Greenlee WF, editors. Comprehensive toxicology. Elsevier Science BV; 2002. p. 583–610. 6. Reily MD, Lindon JC. NMR spectroscopy: principles and instrumentation. In: Robertson JCL, Holmes E, Nicholson JK, editors. Metabonomics in toxicity assessment. Taylor & Francis; 2005. p. 75–104. 7. Lindon JC, Holmes E, Bollard ME, Stanley EG, Nicholson JK. Metabonomics technologies and their applications in physiological monitoring, drug safety assessment and disease diagnosis. Biomarkers 2004;9:1–31. 8. Lindon JC, Nicholson JK, Everett JR. NMR spectroscopy of biofluids. Annu Rep NMR Spectr 1999;38:1–88. 9. Wishart DS. Advances in metabolite identification. Bioanalysis 2011;3:1769–82. 10. Theodoridis G, Gika HG, Wilson ID. Mass spectrometry-based holistic analytical approaches for metabolite profiling in systems biology studies. Mass Spectr Rev 2011;30:884–906. 11. Dunn WB, Broadhurst DI, Atherton HJ, Goodacre R, Griffin JL. Systems level studies of mammalian metabolomes: the roles of mass spectrometry and nuclear magnetic resonance spectroscopy. Chem Soc Rev 2011;40:387–426. 12. Want EJ, Nordstro¨m A, Morita H, Siuzdak G. From exogenous to endogenous: the inevitable imprint of mass spectrometry in metabolomics. J Proteome Res 2007;6:459–68.
Vol. 13, 2015
13. Lindon JC, Holmes E, Nicholson JK. Metabonomics techniques and applications to pharmaceutical research & development. Pharma Res 2006;23:1075–88. 14. Fiehn O. Combining genomics, metabolome analysis, and biochemical modelling to understand metabolic networks. Comp Funct Genomics 2001;2:155–68. 15. Reily MD, Shipkova P, Hnatyshyn S. LC–MS in endogenous metabolite profiling and small molecule biomarker discovery. In: Lee MS, Zhu M, editors. Mass spectrometry in drug metabolism and disposition: basic principles and applications. John Wiley; 2011. 16. Warrack BM, Hnatyshyn S, Ott KH, Reily MD, Sanders M, Zhang H, et al. Normalization strategies for metabonomic analysis of urine samples. J Chromatogr B: Anal Technol Biomed Life Sci 2009;877(5–6):547–52. 17. van der Greef J, van der Heijden R, Verheij ER. The role of mass spectrometry in systems biology: data processing and identification strategies in metabolomics. Adv Mass Spectr 2004;16:145–65. 18. Fiehn O. Combining genomics, metabolome analysis, and biochemical modeling to understand metabolic networks. Comp Func Genom 2001;2:155–68. 19. Hnatyshyn S, Shipkova P. Automated and unbiased analysis of LC–MS metabolomic data. Bioanalysis 2012;4(5):541–54. 20. Nicholson JK, Buckingham MJ, Sadler PJ. High resolution proton NMR studies of vertebrate blood and plasma. Biochem J 1983;211(3):605–15. 21. Bales JR, Higham DP, Howe I, Nicholson JK, Sadler PJ. Use of highresolution proton nuclear magnetic resonance spectroscopy for rapid multi-component analysis of urine. Clin Chem 1984;30(3):426–32. 22. Nicholson JK, Connelly J, Lindon JC, Holmes E. Metabonomics: a platform for studying drug toxicity and gene function. Review 86 refs. Drug Discov 2002;1(2):153–61. Nature Reviews. 23. Chylla RA, Hu K, Ellinger JJ, Markley JL. Deconvolution of twodimensional NMR spectra by fast maximum likelihood reconstruction: application to quantitative metabolomics. Anal Chem 2011;83(12): 4871–80. 24. Reily MD, Lindon JC. NMR spectroscopy: principles and instrumentation. In: Robertson JCL, Holmes E, Nicholson JK, editors. Metabonomics in safety assessment. Taylor & Francis; 2005. 25. Keun HC, Ebbels TMD, Antti H, Bollard ME, Beckonert O, Schlotterbeck G, et al. Analytical reproducibility in 1H NMR-based metabonomic urinalysis. Chem Res Toxicol 2002;15(11):1380–6. 26. Shanks N, Greek R, Greek J. Are animal models predictive for humans? Philos Ethics Hum Med 2009;4(1). 27. Robertson DG, Frevert U. Metabolomics in drug discovery and development. Clin Pharmacol Therap 2013;94(5):559–61. 28. Robosky LC, Wells DF, Egnash LA, Manning ML, Reily MD, Robertson DG. Communication regarding metabonomic identification of two distinct phenotypes in Sprague-Dawley (Crl:CD(SD)) rats. Toxicol Sci 2006;91(1):p309. 29. Robosky LC, Wells DF, Egnash LA, Manning ML, Reily MD, Robertson DG. Metabonomic identification of two distinct phenotypes in SpragueDawley (Crl:CD(SD)) rats. Toxicol Sci 2005;87(1):277–84. 30. Wang TJ, Ngo D, Psychogios N, Dejam A, Larson MG, Vasan RS, et al. 2Aminoadipic acid is a biomarker for diabetes risk. J Clin Invest 2013;123(10):4309–17. 31. Newgard CB, An J, Bain JR, Muehlbauer MJ, Stevens RD, Lien LF, et al. A branched-chain amino acid-related metabolic signature that differentiates obese and lean humans and contributes to insulin resistance. Cell Metab 2009;9(4):311–26. 32. Rabinowitz JD, Purdy JG, Vastag L, Shenk T, Koyuncu E. Metabolomics in drug target discovery. Cold Spring Harbor Symp Quant Biol 2011;76: 235–46. 33. Clarke JD, Novak P, Lake AD, Shipkova P, Aranibar N, Robertson D, et al. Characterization of hepatocellular carcinoma-related genes and
Drug Discovery Today: Technologies | Metabolomics in Medicinal Chemistry
34. 35.
36.
37.
38.
39.
40.
41.
42.
43. 44.
45.
46.
47.
48.
49.
50.
51.
metabolites in human nonalcoholic fatty liver disease. Toxicol Sci 2012;126(1):p429. Robertson DG. Metabonomics in toxicology: a review. Toxicol Sci 2005;85(2):809–22. Robertson DG, Reily MD, Sigler RE, Wells DF, Paterson DA, Braden TK. Metabonomics: evaluation of nuclear magnetic resonance (NMR) and pattern recognition technology for rapid in vivo screening of liver and kidney toxicants. Toxicol Sci 2000;57(2):326–37. Lindon JC, Holmes E, Nicholson JK. Metabonomics: systems biology in pharmaceutical research and development. Curr Opin Mol Ther 2004;6(3):265–72. Lindon JC, Keun HC, Ebbels TMD, Pearce JMT, Holmes E, Nicholson JK. The Consortium for Metabonomic Toxicology (COMET): aims, activities and achievements. Pharmacogenomics 2005;6(7):691–9. Aranibar N, Vassallo JD, Rathmacher J, Stryker S, Zhang Y, Dai J, et al. Identification of 1- and 3-methylhistidine as biomarkers of skeletal muscle toxicity by nuclear magnetic resonance-based metabolic profiling. Anal Biochem 2010;410(1):84–91. Young VR, Haverberg LN, Bilmazes C, Munro HN. Potential use of 3methylhistidine excretion as an index of progressive reduction in muscle protein catabolism during starvation. Metabolism 1973;23(2):1429–36. Aranibar N, Bhaskaran V, Ott KH, Vassallo J, Nelson D, Lecureux L, et al. Modulation of ascorbic acid metabolism by cytochrome P450 induction revealed by metabonomics and transcriptional profiling. Magn Res Chem 2009;47(Suppl. 1):S12–9. Tomlinson L, Tirmenstein MA, Janovitz EB, Aranibar N, Ott KH, Kozlosky JC, et al. Cannabinoid receptor antagonist-induced striated muscle toxicity and ethylmalonic-adipic aciduria in beagle dogs. Toxicol Sci 2012;129(2):268–79. Shipkova P, Vassallo JD, Aranibar N, Hnatyshyn S, Zhang H, Clayton TA, et al. Urinary metabolites of 2-bromoethanamine identified by stable isotope labelling: Evidence for carbamoylation and glutathione conjugation. Xenobiotica 2011;41(2):144–54. Nicholson JK. Global systems biology, personalized medicine and molecular epidemiology. Mol Syst Biol 2006;2. Clayton TA, Lindon JC, Cloarec O, Antti H, Charuel C, Hanton G, et al. Pharmaco-metabonomic phenotyping and personalized drug treatment. Nature 2006;440(7087):1073–7. Winnike JH, Li Z, Wright FA, MacDonald JM, O’Connell TM, Watkins PB. Use of pharmaco-metabonomics for early prediction of acetaminopheninduced hepatotoxicity in humans. Clin Pharmacol Therap 2010;88(1): 45–51. Jackson J, Townsend R, Reily M, Soares H, Averbuch S. Scientific domains of personalized medicine. In: Hawana J, Runkle D, editors. Personalized medicine: prescriptions and prospects. The Food and Drug Law Institute; 2011. p. 1–16. Aranibar N, Reily MD. NMR methods for metabolomics of mammalian cell culture bioreactors. In: Portner R, editor. Methods in molecular biology. 2014. p. 223–36. Bradley SA, Ouyang A, Purdie J, Smitka TA, Wang T, Kaerner A. Fermentanomics: monitoring mammalian cell cultures with NMR spectroscopy. J Am Chem Soc 2010;132(28):9531–3. Chong WPK, Goh LT, Reddy SG, Yusufi FNK, Lee DY, Wong NSC, et al. Metabolomics profiling of extracellular metabolites in recombinant Chinese Hamster Ovary fed-batch culture. Rapid Commun Mass Spectr 2009;23(23):3763–71. Aranibar N, Borys M, Mackin NA, Ly V, Abu-Absi N, Abu-Absi S, et al. NMRbased metabolomics of mammalian cell and tissue cultures. J Biomol NMR 2011;49(3–4):195–206. Wagstaff JL, et al. Masterton RJ, Povey JF, Smales CM, Howard MJ. 1H NMR spectroscopy profiling of metabolic reprogramming of Chinese hamster ovary cells upon a temperature shift during culture. PLoS ONE 2013;8(10).
www.drugdiscoverytoday.com
31