Anal. Chem.

2009, 81, 24182420

Phosphoproteomics: Miles To Go Before Its Routine

Christine Piggee
PDB PROTEIN IDS: 1ERK, NATURE 1994, 367, 704 711 AND 2ERK, CELL 1997, 90, 859 869

Researchers see major technological advances but still have signicant challenges to overcome.
Not so long ago, the process for determining protein phosphorylation was a labor-intensive and serial process. Typically, a researcher would use 32P-labeling coupled with 2D gel electrophoresis phosphopeptide mapping, autoradiography, band excision, and Edman sequencing. In addition to expending much time and effort, researchers had to work with radioactivity, and the method could not be multiplexed. Now in 2009, two complementary techniques for identication of phosphorylation sites, MS and microarrays, have provided signicant increases in throughput. In discovery mode, researchers can identify thousands of phosphorylation sites with LC/MSn, and in proling mode, researchers can detect thousands of phosphorylation events and predict the substrates of various protein kinases with planar, suspension, and reverse microarrays.1 As the technology has improved and researchers have been able to glean impressive amounts of information from phosphoproteomics experiments, more labs have shown interest. I think the eld of phosphoproteomics is poised to take a big leap forward, says Timothy Veenstra of the U.S. National Cancer Institute. And the widespread enthusiasm among researchers is tangible. Its probably the most exciting time right now in phosphoproteomics, says Daniel Figeys of the University of Ottawa. [Phosphoproteomics] used to be kind of stuck a few years ago, where the numbers of phosphorylation sites that could be mapped were dreadfully small, he says. But mapping so many sites can be a problem if researchers cannot weed through the information to nd what is important in a reasonable amount of time, as Matthias Mann of the Max Planck Institute of Biochemistry (Germany) explains. Now were in the paradoxical situation that we almost have too much of a good thing. So its now possible to quantify even 10,000 phosphorylation sites in a single experiment. But in my view ... without then knowing which ones of the 10,000 are actually relevant for [a particular] biological system, then its too much, he says. With this ood of new information about protein phosphorylation, the intricacy of the cell signaling network is becoming more
2418 Analytical Chemistry, Vol. 81, No. 7, April 1, 2009

apparent. In the past with proteomics, people thought, Okay, one gene, one functionsone gene, one protein, says Albert Heck of Utrecht University (The Netherlands). It makes you wonder and makes you humble about what we know already. Researchers are learning that the cell signaling pathways are incredibly complicated. We think we understand sometimes how proteins are interacting with each other and how signaling networks are really working, says Emanuel Petricoin of George Mason University. We go into these clinical trials where we think we can turn off these pathways, and you have some compensatory feedback loop that was unknown take over and actually turn on another pathway that you didnt even know was connected. Elucidating the structure and function of these complex signaling networks will be a challenging enterprise. Fortunately, the eld of phosphoproteomics has been gaining momentum over the past
10.1021/ac802740t CCC: $40.75 2009 American Chemical Society Published on Web 03/10/2009

few years, fueled by great technological advances in identifying and detecting phosphorylation sites. Despite this monumental progress, phosphoproteomics researchers still must overcome several obstacles before these methods can see routine use. PHOSPHORYLATIONsTHE MAGIC WAND The reversible, covalent attachment of a phosphate group to one or more amino acids is the most common posttranslational modication for regulating cellular processes, and researchers estimate that as many as 30-50% of proteins are phosphorylated at any given time. Phosphorylation is like a magic wand that changes regions on a protein from hydrophobic to hydrophilic, and this can result in a conformational change of the whole protein. The phosphorylated amino acid can also form part of a structure recognized by the binding sites of other proteins, thus driving the regulated assembly and disassembly of protein complexes.2 Kinases and phosphatases are the enzymes that phosphorylate and dephosphorylate proteins, respectively, on a timescale of seconds to minutes. Many kinases are phosphoproteins themselves and are active only when phosphorylated. Cellular signaling is a cascade of reactions: phosphorylation activates kinase x, then x phosphorylates kinase y, and so on.3 Disruptions of these cascades have been linked to several diseases, including cancer. When someone says cancer is a protein pathway disease, what they mean is, its a phosphoproteomic disease, because at the end of the day, its the phosphoproteins that drive these protein pathways, says Petricoin. Compared with regular proteomics, phosphoproteomics poses some additional challenges. Phosphorylation is a dynamic, reversible process, and the ratio of phosphorylated to unphosphorylated proteins can be rather low in vivo. Mann says that phosphoproteomics experiments are more technically challenging because they require more initial separation and additional sensitivity. Moreover, the objectives for phosphoproteomics research are somewhat different than those of regular proteomics. THE MISSION The initial phosphoproteomics tasks are clearsidentify the phosphoproteins, determine the locations of the phosphorylation sites on the proteins, quantify the phosphorylation under different conditions, and determine the stoichiometry of the phosphorylation. But that is not enough, according to Michael Snyder of Yale University. One thing thats been clear from studies of individual proteins is that often, an individual phosphorylation event by itself doesnt tell you much, he says. In addition, researchers are seeking the answers to several other questions. When a protein is phosphorylated, which kinase or kinases are responsible? How does each phosphorylation t into the signaling network? And what is the ultimate biological signicance of the phosphorylation? Although MS-based phosphoproteomics originally concentrated on cataloging phosphorylation sites, the emphasis is shifting. The new [MS] technologies have obviously been fantastic for cataloging these sites, says Snyder. Where we aresas far as matching which kinases are phosphorylating what residues and how that controls gene expression, protein activities, localization, and everything else on a global scaleswe really are in the Stone Age here. Veenstra agrees: I see the focus being

turned towards collecting very specic data on phosphopeptides for select proteins. What you really want to do is look at how all of those [phosphorylations] change as a function of some sort of perturbation, says John Yates of Scripps Research Institute. Phosphoproteomics researchers must also perform quantitative phenotypic measurements to ensure that the perturbation has the desired effect on the phenotype. Functional assays, such as measuring migration, proliferation, and apoptosis, can give information on the downstream biological response associated with a particular phosphorylation.4 THE USUAL SUSPECTS: MS AND MICROARRAYS The majority of current phosphoproteomic methods use MS or microarray technology. Over the past several years, the complementary nature of the two techniques has become apparent. MS is the best for cataloging and quantifying phosphorylations in discovery mode, when researchers dont know exactly which proteins are present. It is also the technique of choice for identifying exactly which amino acids in the proteins are modied with phosphate groups covalently attached to their side chains. Currently, bottom-up protein sequencing is the more popular MS method for studying protein phosphorylation, but with gains in instrumental performance, more labs are beginning to explore the top-down method. A few different stable-isotope-labeling methods have been developed for quantication of the phosphopeptides.1 High-throughput microarray technology is best for proling known phosphorylations in hypothesis-driven experiments.5 Antibodies, antigens, peptides, or complex mixtures can be immobilized on a microarray to capture and quantify specic proteins or antibodies. Arrays can be in a planar or suspension format, and uorescence is the traditional method of detection. Researchers have used planar microarrays not only for detecting phosphoproteins but also for predicting the substrates of various protein kinases.6 Suspension-based arrays harness uorescent beads of different colors that are labeled with different antibodies for a multiplexed assay; the beads are sorted by ow cytometry.7 In reverse-phase protein microarrays, cell lysates are spotted onto membranes and probed with phospho-specic antibodies to targeted proteins. Akhilesh Pandey of the Johns Hopkins School of Medicine says that MS can be high-throughput when it is working optimally but that microarray-type assays are high-throughput for the general public. Despite being high-throughput and direct, microarrays are in vitro systems where the results do not necessarily reect what occurs in vivo. Really, in the long run, you are going to want the combination of bothsin vitro plus the in vivosand I think together, the two will be incredibly powerful, Snyder says. RECENT MILESTONES I think that our ability to discover, on a large scale, sites of phosphorylation has really dramatically increased over the last 5 years, says Joshua Coon of the University of Wisconsin Madison. He and others attribute these gains to improvements in MS instrumentation and enrichment techniques. Most people now take for granted these hybrid instruments like ion trap/Orbitrap or ion trap/[ion cyclotron resonance], says Coon. Clearly, those types of instruments have been important in all proteomics, but
Analytical Chemistry, Vol. 81, No. 7, April 1, 2009 2419

in phosphoproteomics in particular, he says. Several researchers say that electron transfer dissociationsan alternative method to fragment peptidesswas another signicant advance for phosphoproteomics. In addition, Yates cites rened workows as another factor contributing to the recent ability to do such largescale analyses. In the past few years, researchers have begun to link results from their global and targeted phosphoproteomics studies back to cellular pathways. For example, Veenstra highlights the work of Forest White of the Massachusetts Institute of Technology and others who have focused on very specic sites within pathways and have started to put together how these networks of phosphorylated residues work in concert. REMAINING CHALLENGES Further improvements are necessary to achieve truly comprehensive experiments that can identify every single phosphorylation, independent of its concentration. I think getting comprehensive phosphoproteomics is still going to be a challenge, says Snyder. Were still only looking at [the] most abundant ones. Comprehensiveness, or lack thereof, also impacts reproducibility. Its very hard to reproduce experiments if youre not comprehensive, says Heck. Pandey points out that data-dependent acquisition is inherently irreproducible, so alternative ways of choosing ions for further fragmentation are needed. Although researchers give some idea of the reproducibility of their analyses, biological replicates tend to be overlooked. According to Roland Annan of GlaxoSmithKline, the majority of labs dont prepare sufcient replicate samples in their global proling experiments because the experiments are so time-consuming. For this reason, he advocates a paradigm shift toward targeted approaches that focus on specic phospho-epitopes derived from the initial global discovery experiment. Researchers also need to improve sensitivity so that they can work with smaller amounts of sample. We get huge numbers of phosphorylation sites, and we can do a pretty effective job with quantitation at large scales. But they require too much starting material, and they take way too long, says Annan. Sample requirements for MS-based phosphoproteomics need to be brought in line with the amounts of material available from clinical samples. One major challenge is that we are still not able to use limiting amounts of material, say, from small biopsies or even neneedle-aspiration biopsies ... and then be able to go deep in those samples and get the same kind of information, says Pandey. We must use the best possible instrumentation, because at some level, it is a statistical game. Better enrichment techniques for phosphopeptides and phosphoproteins also pose a signicant challenge to researchers. If you cant sh out the site of interest, it doesnt matter what your techniques are from then onsyoure not going to nd it, says Kristina Hkansson of the University of Michigan. According to Heck, the current enrichment methods are not comprehensive.

At best, they are complementary to each other, he says. We need to develop additional complementary techniques such as starting with proteases other than trypsin. But overall, he says, phosphoproteomics enrichment is still somewhat of an art. His declaration was conrmed by the Association of Biomolecular Resource Facilities 2007 Proteomics Standards Research Group (sPRG) Study, which concluded that varying enrichment results in its multilaboratory study pointed to the overwhelming importance of individual technique and experience.8 TO MAINSTREAM RESEARCH AND BEYOND An overarching goal for phosphoproteomics researchers is to create technology that can be implemented in mainstream labs and even in the clinic. Already, investigators have begun using phosphoproteomics to address real biological and clinical problems, such as unraveling important signaling networks and discovering how these networks malfunction in various types of cancer.9-13 Despite the success of some initial biological and clinical applications of phosphoproteomics, researchers want to see more progress in the methods. [Phosphoproteomics] should become routine, says Figeys, the same way as doing protein identication from a gel band 15 years ago was very difcult, and now any decent lab in the world can do it. And if the eld continues to develop at the same pace, it should not be too long before this goal is realized.
Christine Piggee is an associate editor of Analytical Chemistry.

REFERENCES
(1) Schreiber, T. B.; et al. Proteomics 2008, 8, 4416-4432; DOI 10.1002/ pmic.200800132. (2) Alberts, B.; et al. Molecular Biology of the Cell, 5th ed.; Garland Science: New York, 2007; p 175. (3) Alberts, B.; et al. Molecular Biology of the Cell, 5th ed.; Garland Science: New York, 2007; p 896. (4) Schmelzle, K.; White, F. M. Curr. Opin. Biotechnol. 2006, 17, 406-414; DOI 10.1016/j.copbio.2006.06.004. (5) Grifths, J. Anal. Chem. 2007, 79, 8833-8837; DOI 10.1021/ac071993n. (6) Ptacek, J.; Snyder, M. Trends Genet. 2006, 22, 545-554; DOI 10.1016/ j.tig.2006.08.005. (7) Sanchez-Carbayo, M. Dissecting Cancer Serum Protein Proles Using Antibody Arrays. In Clinical Proteomics: Methods and Protocols; Vlahou, A., Ed.; Methods in Molecular Biology, Vol. 428; Humana Press: New York, 2008; pp 263-290. (8) ABRF [Association of Biomolecular Resource Facilities]. sPRG Study 2007 Oral Presentation: Development and Evaluation of a Phosphoprotein Standard; www.abrf.org/ResearchGroups/ProteomicsStandardsResearchGroup/ EPosters/sPRG2007.pdf. (9) Olsen, J. V.; et al. Cell 2006, 127, 635-648; DOI 10.1016/j.cell.2006. 09.026. (10) Walters, D. K.; et al. Leuk. Res. 2006, 30, 1097-1104; DOI 10.1016/ j.leukres.2006.01.001. (11) Van Meter, A. J.; et al. Mol. Cell. Proteomics 2008, 7, 1902-1924; DOI 10.1074/mcp.M800204-MCP200. (12) Harsha, H. C.; et al. J. Proteome Res. 2008, 7, 4651-4658; DOI 10.1021/ pr800139r. (13) Bakal, C.; et al. Science 2008, 322, 453-456; DOI 10.1126/science.1158739.

AC802740T

2420

Analytical Chemistry, Vol. 81, No. 7, April 1, 2009