technology feature

Time to apply 119 Building pathways 120 Box 1: Metabolomics in situ 121

Metabolomics: from small molecules to big ideas Monya Baker

The focus of metabolomic studies is shifting from cataloging chemical structures to finding biological stories.

In 2004, when Alan Saghatelian wanted to know what reactions a particular enzyme catalyzed in vivo, he had to plow through a stack of paper two feet high. Standard in vitro techniques to uncover enzymes’ functions would have been much more straightfor- ward, but such studies work only when the right substrates are screened and even then might represent reactions that only occur in a test tube. To get a read on enzymatic activity in actual cells, Saghatelian, then a postdoc in Benjamin Cravatt’s laboratory at The Scripps Research Institute, developed an ‘untargeted metabolomics’ strategy to find previously unidentified substrates. Saghatelian lysed

cells with and without the enzyme, ran the Diego San California, of University laboratory, Palsson Lewis, Nathan lysates through a mass spectrometer and Genomic, proteomic and metabolomic data can all be integrated using genome-scale metabolic compared the small molecule profiles found network reconstructions. in each sample. Nature America, Inc. All rights reserved. All rights Inc. America, Nature 1 This approach, which they termed dis- organism. The assumption was that a list Unlike genes, transcripts and proteins, covery metabolite profiling, yielded sub- of human metabolites could be as useful metabolites are not encoded in the genome.

© 201 strates unlikely to have been found in vitro as a list of human genes. Such a list could They are also chemically diverse, consist- and led to the discovery of a new class of help researchers probe multiple pathways ing of carbohydrates, amino acids, lipids, mammalian lipids as substrates for an simultaneously, and collections of metabo- nucleotides and more. enzyme considered well characterized1. lites consistently associated with biological Extraction, separation and analytic Identifying the substrates, though, was traits could become biomarkers for identi- techniques that work for one class of slow going. “I had to manually look at each fying diseases and optimal treatments. metabolites are often useless for others. mass range [of chromatograms from the Researchers have only recently been With nucleic acids and proteins one detec- spectrometer], looking for ions that had able to integrate all the necessary steps to tion technique will usually suffice, but the changed between the two samples,” says produce profiles consistently. “It’s been a metabolomic community must rely on Saghatelian, now at Harvard University. decade-long progression to measure more a suite of detection techniques, includ- Things are different now, he says. Not only metabolites more reliably,” says Joshua ing capillary electrophoresis–mass spec- are mass spectrometers better at resolving Rabinowitz at Princeton University. trometry, liquid chromatography–mass and measuring compounds, instrument Profiling the complete metabolome spectrometry, gas chromatography–mass vendors all now sell software for compar- remains an impossible task for now, but spectrometry and nuclear magnetic reso- ing samples automatically. with dependable methods in hand, he nance, all paired with various enrichment The word ‘metabolomics’ was coined in says, the field is shifting from cataloging and separation technologies. 1998. It refers to biochemistry writ large, metabolites to asking broader biological “Over the last 10 years, we have learned the study of the entire set of small mole- questions about how metabolites reflect that there is no one platform; there is no cules in a biological sample. In 2001, when and affect cell function. one method,” says Rima Kaddurah-Daouk a handful of biochemists organized them- at Duke University, one of the scientists selves into the Metabolomics Society, their The known unknowns who founded the Metabolomics Society. In chief goal was to produce a catalogue of the Small molecules are less tractable to cata- fact, she says, it is hard even to pin down metabolites that existed in any particular loging than are the objects of other ‘omics’. how many compounds a sample contains.

nature methods | VOL.8 NO.2 | FEBRUARY 2011 | 117 technology feature

a few thousand they make themselves as researchers. Members of Oliver Fiehn’s lab well as nutrients and pollutants from their at the University of California, Davis, for environment and substances produced by example, have ten different kinds of mass microbes in the gut. spectrometers and five protocols geared to “We use three different detection various compound classes. methods, and we still miss stuff,” says And detection is just part of the chal- Michael Snyder of Stanford University, lenge. Many compounds can be detected describing his work to detect metabo- lites that interact with proteins. Though researchers in Snyder’s laboratory have now developed techniques to look at hydrophilic compounds, their initial studies looked at hydrophobic ones not because these were more likely to be interesting but because these were technically easier to analyze2. Even so, he says, developing these assays required years of optimization and quality

Denise Applewhite, Princeton Office of Communications of Office Princeton Applewhite, Denise control. Graduate student Xiyan Li often Josh Rabinowitz wants to remake the classic ran assays six times to make sure everything biochemistry diagrams to include quantitative was working and highly reproducible. information. Scientists in a few highly dedicated laboratories can look at all sorts of com- Most bacteria are believed to produce a few pounds, from volatile compounds to Oliver Fiehn is creating software to show how hundred metabolites; a plant may produce complex lipids, but that capacity requires different mass spectrometry setups will portray a few thousand. Animals contain perhaps extensive equipment and well-trained the same molecule. Nature America, Inc. All rights reserved. All rights Inc. America, Nature 1 © 201

118 | VOL.8 NO.2 | FEBRUARY 2011 | nature methods technology feature

researchers identify genetic variation that predicts how a patient will respond to certain antidepressants, says Fiehn, one of several authors of a study showing how metabolomics can inform genomic studies3. Metabolomic profiling found that levels of the amino acid glycine dif- fered between individuals who stopped responding to antidepressants and those who continued to benefit from them. This led researchers to analyze genetic varia- tion in a range of metabolic genes in these individuals, focusing on genes associated with glycine synthesis and degradation. The team ultimately identified genetic variations that could help predict how an Alan Saghatelian says metabolomic profiling can uncover enzymes and pathways with therapeutic relevance. individual will respond to certain antide- pressants. Without metabolomic profiling, but not identified. Researchers can make experiments,” says Saghatelian. “If you Fiehn says, his colleagues would have been informed guesses and usually deduce com- find something that hasn’t been identified drowning in data. “It lets you know what pounds’ molecular formulas, but typically yet, you don’t know what to do with that. to look for,” he says. only about 25% of observed compounds When you find a new metabolite, you can’t In a different approach, Jeremy Nicholson can be tentatively identified. “Compound just knock it out like a gene or a protein; of Imperial College London is applying ana- identification is improving the fastest as it you have to figure out the pathways that lytical tools developed for genetic research had the furthest to come,” says Steven M. make it and break it down, and what it to metabolomics. Metabolite-wide asso- Fischer, senior applications specialist at binds to.” ciation studies, he says, could be equally Agilent Technologies. Recently, software But profiling studies can have big pay- or more useful than genome-wide associa- from Gary Siuzdak and Colin Smith at The offs. The metabolites present in a cell offer tion studies to classify disease and predict Scripps Research Institute has automated a better snapshot of what is happening at a responses to treatment. data analysis, making metabolomic analy- given moment than genes or even proteins, Metabolites themselves may have a sis much more accessible. Still, confirm- says Saghatelian. “What’s nice about these more important biological role than ing an unknown metabolite’s identity often metabolic profiling experiments is that you many scientists have assumed. Whereas requires not just puzzling over spectra can measure some changes you can’t see in the goal of hundreds of studies has been and running samples against known stan- genome sequencing, but it’s still clear that a to understand regulatory networks by Nature America, Inc. All rights reserved. All rights Inc. America, Nature 1 dards but also synthesizing compounds. pathway is important.” hunting for protein-protein and protein- Such unknowns can discourage research- DNA interactions, few studies have been

© 201 ers from looking broadly at metabolites. Time to apply designed to look at protein-small molecule “If you’re not a synthetic chemist, you’re Combining metabolomic profiling interactions. In the first high-throughput not likely to want to do these kinds of with other omics techniques has helped study to do so, researchers led by Stanford

Table 1 | Bioinformatic tools for metabolomics Name and web address Lead developer, affiliation Description BiGG database Bernhard Palsson, ‘In silico resonstructions’ that integrate published data from genome-scale http://bigg.ucsd.edu/ University of California, San Diego metabolic networks BinBase Oliver Fiehn, Tool to process and analyze data from different mass spectrometry platforms http://sourceforge.net/projects/binbase/ University of California, Davis BioCyc Peter Karp, Databases describing biochemical machinery of many organisms, plus http://biocyc.org/intro.shtml/ SRI International software to construct, query and publish pathway databases online Chemical Translation Service Oliver Fiehn, Software to allow queries across metabolomic publications and databases http://cts.fiehnlab.ucdavis.edu/ University of California, Davis that use different compound identifiers Human Metabolome Database David Wishart, Database containing nearly 8,000 metabolites; each entry contains 110 http://www.hmdb.ca/ University of Alberta fields covering clinical, biochemical and other data MassBank Takaaki Nishioka, Repository of high-resolution mass spectral data for metabolites http://www.massbank.jp/ Keio University METLIN Gary Siuzdak, A database and repository of mass spectral data on over 25,000 metabolites http://metlin.scripps.edu/ The Scripps Research Institute XCMS/metaXCMS Gary Siuzdak, Open-source software to identify and analyze metabolites from raw mass http://metlin.scripps.edu/xcms/ The Scripps Research Institute spectrometry data; metaXCMS allows comparisons across several sample classes http://metlin.scripps.edu/metaxcms/

nature methods | VOL.8 NO.2 | FEBRUARY 2011 | 119 technology feature

25 indicated potential therapeutic interven- systems analysis, is really emerging now,” 20 15 tions. In one oft-cited example, research- says Bernard Palsson, at University of 10 ers at Agios Pharmaceuticals investigated California, San Diego. A metabolic profile is Fold change 5 mutations in genes for the enzyme isoci- really a disassembled jigsaw puzzle of cells’ 70 60 800 trate dehydrogenase that are frequently activity. Omics and informatic tools can 50 600 700 40 500 Retention time 30 300 400 m/z observed in glioblastomas. Metabolic pro- help researchers put that puzzle together. (min) Members of Saghatelian lab, lab, Saghatelian of Members University Harvard filing showed that the mutations bestowed Global metabolic analysis is both eas- A three-dimensional plot can be used to compare a new catalytic function on the enzyme5. ier and more complex than the research ratios of various metabolites across samples. Notably, this mutated active site was not that revealed the Krebs’ cycle some seven present in benign cells. decades ago, says Rabinowitz. “We can see University’s Snyder isolated dozens of pro- Metabolomics’ beginnings in bio- in the mass spectrometer all the things teins from yeast and then carefully collect- chemistry offer a distinct advantage for that had to be inferred or painstakingly ed and analyzed hydrophobic metabolites validating protein targets, says Princeton measured before,” he says. Now, however, associated with each protein2. Ergosterol, University’s Rabinowitz, a co-author on researchers are also looking at metabolism the yeast equivalent of cholesterol, bound the study5. Decades of work to understand in a broader way, not only what molecules many proteins in stoichiometric ratios how products of one enzyme become sub- are present and how they interconvert but that suggested a regulatory role. Though strates for the next provide crucial con- understanding how they are controlled. no such interactions had previously been text for interpreting metabolomic data. Palsson’s approach involves collating described, 21 of 100 kinases were associ- And researchers who want to investigate genes known to catalyze biochemical ated with hydrophobic small molecules, the role of particular metabolic enzymes reactions and building them into genome- and additional work on one kinase showed in disease can frequently draw on small- scale metabolic in silico reconstructions, the association was required for enzy- molecule tools, providing a fast-track for which are being used as models to pre- matic activity. Preliminary work indicates drug-discovery efforts. dict the effects of drugs. At this point, that the regulatory roles for metabolites Instead of focusing on metabolite though, models are generally built using translate to human cells, says Snyder: “our amounts, many researchers prefer to look gene expression and proteomics data, research results suggest a highly connected for changes in metabolic activity. Metabolite says Nathan Lewis, a graduate student in network of interactions in which the small amounts can change for multiple reasons, Palsson’s lab. Metabolomic data are used metabolites add an extra dimension of explains Rabinowitz. A downregulated for validating and extending the model regulatory information.” pathway might produce less of a particu- rather than constructing it. In retrospect, such results should have lar metabolite; an upregulated pathway Peter Karp of Stanford Research Institute been expected; the small molecule cyclic might consume more, so opposite mecha- (SRI) International’s Artificial Intelligence AMP (cAMP) has long been known to reg- nisms could produce the same effect. To Center also starts with genomic data ulate many pathways. Other researchers distinguish between these possibilities, to construct databases describing bio- are now looking for other small molecules Rabinowitz developed a technique called chemical machinery of many organisms. Nature America, Inc. All rights reserved. All rights Inc. America, Nature 1 that might have similar roles. Whereas kinetic fluxomics, which uses isotope- Experimental data provided by collabora- Snyder’s strategy was to look for metabo- labeled precursors such as [13C]glucose or tors are computationally analyzed to identi- 13 15 © 201 lites that bound a defined set of proteins, [ C- N] glutamine to gauge pathways’ fy metabolites that change in sync with gene Anne-Claude Gavin and colleagues at the activities over time. European Molecular Biology Laboratory, One such study showed that viral infec- Heidelberg, selected 56 lipids and inter- tion boosted rates of fatty-acid synthesis. PC 16:0/18:1 mediate metabolites and searched for the Adding drugs that reduced fatty-acid syn- Choline

proteins that bound them. They identi- thesis slowed replication of cytomegalovi- Plasmenyl-PC 36:0 PC 18:1/18:1 Betaine

fied over 500 protein-lipid interactions, rus as well as influenza A, the cause of most Carnitine PC 16:0 /16:0 intensity 6 C2-Carnitine over two-thirds of which were unknown serious viral influenzas . As there are many LC-QTOF MS Data acquisition or unexpected4. In follow-up studies steps in lipid synthesis, says Rabinowitz, 6 HILIC retention time (min) 16.0 749 they used live-cell imaging to validate there are potentially many ways to inter- 87.6% the functional interactions of a subset of fere with it and so block viral replication. 625 PC 18:1/18:1 metabolites, the mysterious sphingolipids, Researchers in other labs are also finding m/z 808.7 LipidBLAST revealing a potential new type of protein- that unanticipated pathways are involved Identification →MS/MS 526 m/z Score binding domain. with disease, says Rabinowitz. “No one Snyder calls such work interaction pro- thought that lipid metabolism had any- teomics, which is different from classic thing to do with viruses, and there are other metabolomic profiling experiments. “It’s pathways that people thought had nothing not your typical experiment to look for a to do with certain diseases that are actually biomarker,” he says. “We are trying to work apparently quite central to them.” MetaMapp

out pathways.” In fact, metabolic profiling Pathway enrichment is increasingly moving from identifying Building pathways Davis California, of University laboratory, Fiehn of Members potential biomarkers to putting them in “Metabolism, not in the old sense of dis- Identifying metabolites in biological samples can context. In some cases, this work has even covering molecules, but in terms of doing help researchers home in on pathways.

120 | VOL.8 NO.2 | FEBRUARY 2011 | nature methods technology feature

Box 1 Metabolomics in situ

Where a metabolite is can be much more important than how then analyzed much is there. “If the metabolite is in the vacuole, it is not immediately.” exposed to nuclear or cell-surface receptors,” says Wolf Frommer Because there at the Carnegie Institution for Science, who has developed is no matrix genetically encoded sensors that can be placed precisely in the deposition as in endoplasmic reticulum, nuclear membrane and elsewhere to matrix-assisted identify metabolites. Last year, researchers in his laboratory laser desorption/ reported using fluorescence resonance energy transfer (FRET) ionization and and microfluidics to monitor intracellular glucose and ATP thus no matrix in yeast cells exposed to varying amounts of nutrients7. The background, sensors can also be used to measure metabolite uptake and spectra are easier release in mammalian and plant cells; moreover, FRET is much to interpret. The simpler to implement than many biologists seem to believe, difficulty, says says Frommer. “You don’t even need a microscope,” he explains. Siuzdak, is that Whereas the technique can be used to monitor only one or two the chips are not metabolites at once, Frommer anticipates that studies can be commercially miniaturized and parallelized so that hundreds of metabolites available yet.

can be monitored in time and space. For metabolomics Science for Institution Carnegie laboratory, Frommer of Members For tissue, mass spectrometry approaches can be used. in general, the past Glucose levels in the cytosol of leaf epidermal Gary Siuzdak at Scripps Research Institute recently introduced 5 to 10 years have cells of Arabidopsis thaliana, a model plant, as nanostructure-initiator mass spectrometry, an extraction been characterized visualized and quantified with a FRET sensor. method that can be used to probe metabolites in complex by the advent of tissues. Researchers in Siuzdak’s lab recently used it to detect more accurate machines allowing more accurate identification cholesterol in the cerebrum, cerebellum and brainstem of of the contents of sample lysates, says Arthur Castle, Program mice used to study a neurocognitive disorder that results from Director of Metabolomics and Informatics at the US National faulty cholesterol synthesis8. The implementation, he says, is Institutes of Health. Emerging technologies show not just trivial: “tissues can be placed directly on the nanostructure- compounds’ identities but also their location. “Imaging is not that initiator mass spectrometry chip without any preparation and big yet,” he says, “but I expect it to get big.”

expression. This information is then used identify the unknown metabolites would peaks, but not all of them are important to flesh out maps of metabolic pathways. speed his and others’ research, he says, and metabolites,” says Rabinowitz. Establishing Nature America, Inc. All rights reserved. All rights Inc. America, Nature 1 Separately, Karp uses metabolomic data to several collaborative efforts are underway a set of molecular standards to distinguish search for correlations between metabolites to reduce the number of unknowns and to metabolites from procedural artifacts, he

© 201 and drug responses. put them in context (Table 1). Other tech- says, would be somewhat equivalent to hav- Unfortunately, many metabolites from niques are being developed to precisely locate ing an organism’s genetic sequence; it will collaborators’ data sets cannot be resolved metabolites in cells and tissues (Box 1). be useful, but he is not stopping to wait for into chemical structures, Karp says. “We Ultimately, researchers need to know it. “One thing we’ve learned from sequenc- can’t map them to our pathways right not only what metabolites are where but ing the genome is that you should try to now, so we disregard them.” The ability to also which are relevant. “There are a lot of push forward to interesting science stories as soon as possible.”

1. Saghatelian, A. et al. Biochemistry 43, 14332– 14339 (2004). 2. Li, X., Gianoulis, T.A., Yip, K.Y., Gerstein, M. & Snyder, M. Cell 143, 639–650 (2010). 3. Ji, Z., et al. Clin. Pharm. Therap. 89, 97–104 (2011). 4. Gallego, O. et al. Mol. Syst. Biol. 6, 430 (2010). 5. Dang, L. et al. Nature 462, 739–744 (2009). 6. Munger, J. et al. Nat. Biotechnol. 26, 1179– 1186 (2008). 7. Bermejo, C., Haerizadeh, F., Takanaga, H., Chermak, D. & Frommer, W.B. Biochem. J. 432, 399–406 (2010). 8. Patti, G.J. et al. Neuroscience 170, 858–864. (2010).

Monya Baker is technology editor for

Agilent Nature and Nature Methods More accurate mass spectrometers and data-analysis software encourages metabolic profiling. ([email protected]).

nature methods | VOL.8 NO.2 | FEBRUARY 2011 | 121