Researchers have had a tough time identifying, let alone developing, small molecule inhibitors of protein-protein interactions. But as scientists report an expanding repertoire of potential drug candidates, this nascent field may be on the verge of a growth spurt.

click the image to enlarge   A diagram of the p53-Hdm2 GRIP interaction assay. Treatment with RS25344, an anchor protein inhibitor, leads to development of fluorescent-labeled compact foci. When PR73401 is added, it disperses the foci formation, indicating it is a compound that inhibits protein-protein interaction in an intracellular environment. Nutlin-3 operates in a similar manner. (Source: BioImage A/S)

In March, researchers from Johns Hopkins University, Baltimore, and Harvard Medical School, Boston, reported in Nature [Rodgers et al., vol. 434, pp. 113-118 (2005)] that depriving mice of adequate food causes the SIRT1 protein to interact with another protein, PCG-1α, which in turn triggers glucose production in the liver. Because SIRT1 is a homolog of Sir2, which has been associated with increased longevity in some organisms, most press reports of the study highlighted the possibility of scientists somehow, some day, using SIRT1 to slow aging in humans.

More realistically, says graduate student Joseph Rodgers, lead author of the study, a better understanding of this protein-protein interaction will lead to novel treatment for diabetes. If the interaction could be blocked, the excess glucose production that is so damaging in diabetic patients could be reduced.

That is the kind of targeting that has led many biotech and pharmaceutical companies to invest in the study of protein-protein interactions, including the development of inhibitors that target those interactions. The global market for protein interaction technology and therapeutics is expected to reach $50 billion by 2010.

Perhaps the best known protein-protein interaction inhibitor is Herceptin (trastuzumab), an antibody that binds to the extracellular domain of the human epidermal growth factor receptor 2 protein, HER-2. Herceptin was approved by the FDA in 1998 for use against breast cancer, because clinical trial data showed a longer time to disease progression, a reduced risk of death, and other positive results in women with HER-2-positive metastatic breast cancer who received Herceptin in addition to standard chemotherapy, compared to women who received chemotherapy only.

Last year, Herceptin was the third biggest-selling drug for Genentech, San Francisco, bringing in close to half a billion dollars. Antibodies are the fastest-growing segment of the prescription drug market, but they have their problems. Antibodies are expensive and difficult to manufacture; they can't be taken in pill format (they can only be administered as injections or infusions); and they aren't cell-permeable, which means they can't be used to block intracellular interactions.

Enter small-molecule inhibitors of protein-protein interactions, which are not only cheaper and easier to manufacture than antibodies, but they also are cell-permeable. So far, scientists have only identified a couple dozen potential small-molecule protein-protein interaction inhibitors (SMPPIIs), almost all of which have been published within the last two years. Although few SMPPIIs have reached clinical trials and none have reached the marketplace, all promise novel therapeutic treatments for a range of human ailments, from cancer to HIV/AIDS.

When BioImage A/S, Soeborg, Denmark, started publicizing their novel, cell-based screening technology (i.e., for the identification of new therapeutics, including small-molecule inhibitors of protein-protein interactions), the company met with a lukewarm response. Len Pagliaro, PhD, vice president of business development at BioImage, says the disbelief was due in part to the fact that the company had yet to validate its assay prototype, which they have since done. More importantly, small organic molecules were not thought of as compounds that could inhibit, or bind to, protein-protein interaction interfaces. "That view has changed enormously in just the last few years," says Pagliaro. "I think they are going to be developed into fantastic drugs. Are they going to address all targets or work for every therapeutic area? No. But for the targets they address, they are going to be fantastic."

Ideas that in the recent past elicited disbelief, or almost derision, have sprouted into a bustling bud of activity. The growth has been fueled by the development of new screening technologies, new applications of common drug optimization technologies and, most importantly, new ways of thinking about protein-protein interactions and the small molecules that can inhibit them.

Crags and crevices
In the past, one of the major challenges to finding potent small-molecule inhibitors of protein-protein interactions was the perceived flatness of most protein interfaces. X-ray

SP-4206 is shown binding to the "hot spot" of the cytokine IL-2. SP-4206 is one of the most potent small-molecule protein-protein interaction inhibitors that has been developed using the fragment-based Tethering technology. (Source: Sunesis Pharmaceuticals)

structures have shown that a good chunk of the surface area of a typical protein-protein interface is buried, with the atoms closely packed together and not many small, deep cavities are available for small-molecule binding. Yet over the last seven or eight years, using scanning mutagenesis and other methods, researchers discovered that many protein-protein interfaces have a small, well-defined, compact area that plays an integral role in the affinity of the interaction. These high-affinity areas, which contain the nooks and protrusions necessary for the actual binding of the agent, have been termed "hot spots."

Additionally, many interacting proteins can shift their binding-site conformation in order to complement different protein partners (many proteins bind to multiple partners), which means that the cavities one would expect to see in an X-ray structure might not show up. The presence of these hot spots, combined with the plasticity of the protein interface, suggests that protein-protein interfaces may in fact be adept at binding small molecules, contrary to initial speculation that a typical protein interface isn't a very likely place for a small molecule to reside.

Finding a foothold
Realizing that small molecules can bind to protein interfaces has been a significant step forward for the field, but it was only one step. Next, researchers had to find molecules that serve the purpose, which means they needed a set of potential candidates from which to choose. The search for potent SMPPIIs has proven to be a major challenge because of the scarcity of naturally occurring small molecules that are known to bind to protein-protein interfaces and the associated lack of small-molecule starting points for optimization.

Investigators successfully developed and used a variety of screening approaches to obtain chemical starting points for SMPPII discovery. These include functional biochemical screens such as traditional high-throughput screens, functional cellular screens, and fragment-based "binding screens."

Over the past couple of decades, high-throughput biochemical screens have become the focal point for many drug discovery programs in the pharmaceutical industry. It is not surprising that they may be finding a place in SMPPII discovery as well. In a review article in Current Opinion in Chemical Biology [L. Pagliaro et al., vol. 8, pp. 442-449 (2004)], Pagliaro and colleagues described 19 recently published examples of SMPPIIs.

In an effort to evaluate how helpful screening commercial diversity space is, or could be, with respect to SMPPII discovery, using a principal component analysis, Pagliaro and co-

Protein-Protein Interactions and New Antibiotics? Earlier this year, Andrew Emili, PhD, University of Toronto, Canada, and colleagues published a paper on protein interactions in Escherichia coli [G. Butland et al., Nature, vol. 433, pp. 531-537 (2005)]. The study was remarkable, say several leaders in the field, because it was the first large-scale analysis of protein complexes in a bacterium. As Tracy Palmer, PhD, of the John Innes Centre, Norwich, UK, wrote, "Expect a mad scramble across the globe as every microbial biochemist maniacally checks the supplementary information to see what their favorite protein interacts with . . ." Using a modified yeast-based tandem affinity purification (TAP) procedure for isolating protein complexes, the research team purified 648 different affinity-tagged E. coli proteins to homogeneity. They used mass spectrometry to identify the proteins' interacting partners. Eighty-five percent of the validated interactions were new (i.e., not described in select databases). Most of the proteins had just a few interacting partners, whereas others—the "hubs"—had far greater numbers of connections. Of course, scientists have been seeing the same in eukaryotes, but this was the first time a prokaryote's whole-cell protein network had been linked together. The data serve as a starting point for understanding not just how a bacterial cell is designed, but how it functions. "We don't understand how cells are designed," says Emili. "But if we understood the circuit board, the logical relationships of all the components, it wouldn't be so random. To figure out how cells work is a daunting and perhaps irrational task, but if we are ever going to look at the genome and make accurate predictions, that's the way we have to go." Even more exciting, Emili says, the map provides information that may be helpful in elucidating the evolutionary relationships of microbial protein interactions and networks, particularly if the data can be analyzed in combination with the extant wealth of comparative genomic data on prokaryotes. If the same interacting proteins are found in different species, what does that tell us about how these networks have evolved? If some of those pathways turn out to be broadly conserved, microbial protein mapping could facilitate the discovery of broad-range antibiotics. Currently, only about 30 bacterial proteins are targeted by prescription antibiotics, and antibiotic resistance has emerged as a major global public health problem. Although Emili is not involved with the pharmaceutical aspects of the research, he says that others are exploring the possibillity of using the information to find new bacterial drug targets. Whether these targets would ever be suitable for SMPPIIs remains to be seen.

authors found that about half the SMPPIIs and a handful of their protein-protein interaction targets were covered among three vendor databases: Chemical Diversity, Maybridge, and Asinex, which together cover more than 500,000 compounds. The analysis suggests that while it may be possible to access the chemical space of druggable active SMPPIIs for some protein-protein targets, using a traditional biochemical screen doesn't work for other targets, at least not yet. Pagliaro expects this to change, as available chemical diversity space is expanded.

"If you look at commercial chemical libraries and then at small molecules that are known protein interaction inhibitors, it is evident that most protein interaction inhibitors are not in today's commercial libraries," says Pagliaro. "People looked at libraries and haven't found anything, so they conclude that small-molecule protein interaction inhibitors don't exist."

Researchers are starting to realize there are indeed going to be small-molecule chemical classes that address protein-protein interactions. In the future, Pagliaro expects protein-protein interaction inhibitor libraries will be sold. "Then the real work can start."

Another type of functional screen being developed for use in SMPPII discovery is cell-based assays, also known as cell translocation assays, which can be used to monitor the intracellular behavior of targeted molecules (or interactions), as opposed to activity against purified proteins. These assays, like BioImage's GRIP technology, use proteins tagged with green fluorescent protein (GFP) to measure inhibition by potential SMPPIIs.

GRIP is a p53-HDM2 interaction assay that was recently validated through the use of Nutlin-3, a SMPPII that inhibits the MDM2-p53 interaction. Nutlin-3 was identified last year by Lyubomir Vassilev, PhD, and colleagues at Hoffmann-La Roche Inc., Nutley, N.J. (See figure on page 22) Overexpression of MDM2 impairs the tumor suppressor p53, so it is believed that SMPPIIs that restore p53 activity, such as Nutlin-3, could eventually be developed into novel cancer-fighting drugs.

In contrast to functional screening, where compounds are identified on the basis of their activity and then analyzed to determine their binding affinity and other biological features, fragment-based discovery involves identifying multiple molecules, or fragments, on the basis of binding affinity, and then combining the fragments to make a "key" that best fits the binding pocket. Fragment-based discovery is particularly valuable in cases where there is little known about how to identify novel compounds that productively bind to their targets, such as with small organic molecules and protein-protein interactions. "If binding is a rare event," says Michele Arkin, PhD, of Sunesis Pharmaceuticals, South San Francisco, "it's going to be hard to make the key right off the bat. But if you make it one pin at a time, you can make the pins first and then connect them afterward."

There are a couple different ways to screen the fragments. Nuclear magnetic resonance (NMR) is one. Another is what scientists at Sunesis Pharmaceuticals have termed "Tethering," which allows investigators to identify even weak-binding fragments that might otherwise be difficult to detect. Tethering involves screening the target protein against a collection of sulfur-containing fragments (i.e., so that a stabilizing, disulfide bond can form between the fragment and the protein surface, which is modified with a sulfur-containing amino acid); using mass spectrometry to detect bound fragments; and then combining multiple fragments that occupy neighboring sites into a single molecule.

The resulting compound is then ready for optimization. Although Sunesis Pharmaceuticals has since stopped developing it because the target is not very well validated, SP-4206, a small molecule that binds to the IL-2 interleukin receptor, is among the most potent SSPPIIs that have been identified thus far using Tethering technology, says Arkin. (See figure on page 23)

Other screening technologies being used to search for SMPPIIs include peptidomimetics and structure-based design, but there is debate on which is best. "Several methods have been shown to work, if the target is good," says Arkin. More important than technology, she says, researchers are having more success identifying novel SMPPIIs because they are choosing more tractable targets. The question is no longer whether protein-protein interactions can be blocked by small organic molecules, but rather which interactions are easiest to block.

Smart targeting
Most protein-protein interactions involve more than a single pair of proteins. In some cases, as many as 50 proteins interact as part of a large, multimeric protein complex. It is almost deliberately naive to think that inhibiting a single interaction will not have multiple effects, including counterproductive effects on other pathways. Thus, choosing a tractable target that actually has the desired biological effect poses yet another major challenge to SMPPII design, indeed, to the design of any drug. To some extent, only animal experimentation and clinical trials will test the field's capacity to overcome this hurdle, but protein-protein interaction maps provide a way to maximize the odds.

By identifying where and understanding how proteins interact with the myriad of other cellular proteins, which comprise more than 50% of cellular dry weight, drug hunters can more readily identify what Jacques Camonis, PhD, Institut Curie, Paris, and Laurent Daviet, PhD, Hybrigenics SA, Paris, both refer to as "smart targets."

Glimpsing the Future

"The hunt for inhibitors of protein-protein interactions has been going on in a low-level way for a long time. But as certain targets get to be better characterized and are shown to be effectively inhibited by small molecules, it is going to expand. There are a number of viable approaches to screening for inhibitors. Comfort with this target class is going to come more from biological validation than from technology. - Michelle Arkin, PhD, senior scientist, Sunesis Pharmaceuticals Inc.

Camonis and Daviet were senior authors on a recent Drosophila melanogaster mapping study, where they identified more than 2,000 protein-protein interactions and produced a map that had some surprising differences from a Drosophila map created in 2003 by researchers at CuraGen Corp., New Haven, Conn. Such differences highlight the need for cross-validation, say Camonis and Daviet.

The goal of producing a protein interaction map is to find new drug targets. "There are a limited number of genes that are 'drugged' [or acted upon by drugs] right now, only about 500," says Daviet. "But many more than that could be good targets. The point is to find them. Doing this kind of study is the first step to identifying new targets."

Ideally, with a map that had been cross-validated with good data, chosen targets are both tractable (i.e., capable of being inhibited) and smart. For example, if one sought to block the interaction between two proteins, A and B, it would be helpful to know how A and B interact with other neighboring proteins as well, so that the drug candidate doesn't also inadvertently block the interaction of A and C or D or E.

Although Drosophila may seem like a very distant cousin to Homo Sapiens, most cancer pathways are conserved between the two species. Because mapping a protein network of a fruit fly is a lot more manageable than doing the same for a human, a comparative proteomics approach should allow for a quicker validation of functional interactions.

As Andrew Emili, PhD, University of Toronto, says, "The pessimist would say, 'You'll learn nothing [from protein-protein interaction maps], because small-molecule drugs work on individual proteins.' But the optimist would say, 'If we know what some of the key hubs or key connection points are that link biological processes in cells, those might be great drug targets.' "

About the Author
Pray is a freelance writer based in Massachusetts.

This article was published in G & P magazine: Vol. 5, No. 4, May, 2005, pp. 22-25.