jancovWhen looking for the most promising leads, or compounds worth developing, drug researchers always seek new tools and techniques that enhance the odds of finding a marketable compound as soon as possible. Likewise, these researchers want to eliminate—also as early as possible—any compounds that will be ineffective or, even worse, toxic. Often, lead identification involves various forms of high-throughput screening, but other approaches also exist, and these cover a range of levels from mathematical models and crystallography to protein fragments and cells. In fact, the most effective approach to identifying the best lead compounds should include a variety of tools and techniques.

The ’omics advances of the past decade or so give drug researchers more to explore, but these advances can also complicate the task. As Milind Deshpande, PhD, president of research and development and chief scientific officer at Achillion Pharmaceuticals (New Haven, Conn.), explains: "The targets are more complex. The biology is more complex."

Consequently, researchers need a better idea of how a specific target fits into healthy and diseased human physiology. To uncover this information, drug researchers use a collection of technologies to identify the most promising leads.

The key challenges
Part of the complexity comes from exploring a wider range of targets. "Some of them don’t even have any precedents," says Deshpande. Beyond the targets, researchers want to identify leads that will succeed downstream. That requires understanding the target’s biological and chemical features, as well as how the compound might affect those characteristics and downstream ones that could also be affected. Not surprisingly, Deshpande adds, "Understanding the downstream biological issues has become more complex."

Even at the target, biology can get complicated. For example, Brian Huber, PhD, vice president, drug development partnerships at Quintiles (Research Triangle Park, N.C.), says, "Some biological targets, like kinases, are very druggable, and you can find a lead series of molecules that would interact with that target." Yet, some target classes, like specific classes of proteases and nuclear receptors, are only somewhat druggable, according to Huber.

Huber also points out some of the chemical challenges in lead identification. When describing a library of potential compounds stored in multiwell plates, Huber says, "What you think is the chemistry in a given well is probably not the chemistry." He notes that the breakdown products and other factors change the chemistry in a well. So instead of managing large libraries that evolve into the unknown, Huber suggests turning to smaller, smarter libraries. "Maybe we should get more focused," he says.

To keep track of so much information, companies can apply tools like SAS Data Integration from SAS (Cary, N.C.). This software provides access to data across a company. Moreover, SAS also helps researchers visualize data, which can reveal more of the features of the compounds being considered as leads.

A two-pronged approach


click to enlarge
In this bond between a small molecule and the murine double minute (MDM2) protein, dotted lines show the molecular interactions. Such images help drug researchers identify promising lead compounds. (Image: Roche)  

At Achillion, Deshpande says that they’ve taken two approaches to lead identification. He calls the first chemical genomics. For that, scientists start with a selected set of compounds and test them against a virus or pathogen. "Starting off, you don’t know which target you will go after," he says. This work often involves cell-based assays.

Achillion might also start with a structure-based approach. "In this case," says Deshpande, "we would know the target we are pursuing." As an example, he describes work on an antibacterial where his company went after the enzymes involved in DNA replication.

As he explains: "We knew the enzymes that we wanted to target and we used three-dimensional information—from x-ray crystallography—about them to synthesize leads that displayed potency as well as the specificity that one likes to engineer in the early stages of lead discovery and optimization."

Building a toolbox
When asked what technologies Roche (Nutley, N.J.) uses for lead identification, Karen Lackey, vice president and head of medicinal chemistry, mentions several. First, she says that traditional high-throughput screening methods are "definitely fruitful." She adds, "We find fragment-based drug discovery very valuable." In addition, Roche scientists use structure-based drug design. "This is a very integrated approach that depends on the crystallographers and computational chemists," she says.


click to enlarge

This simulation shows an agonist bound to the adenosine A2A receptor, which is a member of the G protein-coupled receptor (GPCR) family. GPCR targets are becoming more amenable to structure-based virtual screening with the increasing number of available crystal structures. (Image: Schrödinger)

To develop even more tools, Roche sometimes reaches out to other companies. Roche uses a proprietary platform from X-Chem (Cambridge, Mass.) to search for leads against therapeutic targets. "This technology involves DNA-encoded libraries," Lackey explains.

Thinking over all of these lead-identification technologies used at Roche, Lackey says, "We have a toolbox of methods."

Developing such a toolbox generates options for every project. "We can customize each project so that we find leads in the most effective manner," Lackey says. For example, high-throughout methods help researchers to quickly explore many compounds, and computational methods can be used to explore fewer compounds, but in more detail.

Even within one technology, Roche scientists keep looking for ways to improve it. For fragment-based technology, for instance, Roche recently gathered a group of the company’s experts as well as scientists from outside the company. "They reviewed all of the fragment-based drug discovery techniques and came up with a prioritized list for using them at Roche," says Lackey. "That helps us review lessons learned, challenges overcome, and it also broadens our opportunities to use these technologies."

Consequently, the most effective lead identification can take more than combining tools. It also helps to collect a group of expert opinions.

Smarter identification
When Huber suggested more focus—getting smarter about the compounds in a library and using less of them—he had some specific ideas in mind. "Before you pick up a test-tube, we can help you pick a target-product profile that will be embraced in the clinical arena and be a commercial winner." He adds that finding such a target-product profile depends on finding something that patients want, regulators will approve, physicians will prescribe, and payers will reimburse.

Predictive Software

Certara, a provider of software and services to improve productivity and decision-making from drug discovery through clinical development, has released the PredictFX. Developed by Chemotargets for predicting the off-target pharmacology of small molecules, and made available through exclusive distribution agreement with Certara, PredictFX provides improved opportunities to identify, monitor, and address safety issues earlier in the discovery process by predicting the off-target pharmacology and associated side-effect profile of a drug lead compound from its 2D chemical structure.

Chemotargets has spent the last several years developing and validating a ligand-based approach to predict the affinity profile of a small molecule across a list of thousands of protein targets. The program suite works as an open system that allows for creating customized ligand-based target models with all pharmacological data available by the user that can then process millions of small molecules, whether real or virtual. The predicted affinity profiles of small molecules can subsequently be used to anticipate the likely side-effect profile linked to off-target pharmacology.

The PredictFX software is designed to support the rapidly growing field of translational science, aimed at improving productivity in drug development by making informed decisions on drug optimization earlier in the development process.

Getting smarter in lead identification also involves a new look at numbers. As Woody Sherman, PhD, vice president, applications science at Schrödinger (New York, N.Y.), says, "Getting leads is generally not the biggest challenge. It’s getting suitable leads with promising potential for affinity optimization and good ADME/Tox properties that is hard."

To more efficiently sort the promising leads from the duds, an increasing number of drug researchers include computational tools in the mix. This includes structure-based approaches like docking, ligand-based approaches like pharmacophore modeling, shape-based screening, cheminformatics, and more. "By using a combination of these tools," says Sherman, "you’ll find the most diverse and interesting leads." He adds, "The computational results can help reduce the hit list to the most promising compounds to pursue."

To give researchers a collection of software tools, Schrödinger has a complete suite of structure-based, ligand-based, cheminformatic, and visualization tools. This suite includes Glide for docking, Prime for protein structure prediction, Phase for pharmacophore modeling, and Canvas for cheminformatics.

The computational approaches can also go beyond binding activity to assess what really makes a good lead, since off-target selectivity is often an issue. Sherman says, "There have been recent advances in structural modeling of important ADME/Tox targets, like hERG, P450s, P-glycoprotein, et cetera, so you can throw out compounds early in the game and focus on the most promising drug candidates as quickly as possible. Furthermore, structural insights from docking to off-targets can help guide the design process so potential liabilities in lead compounds can be fixed."

However, in filtering compounds, drug researchers must be careful. "It’s easy to throw away a lot of compounds," Sherman says, "but you want to make sure that you’re not throwing out the most promising ones." He believes that the best computational tools can effectively eliminate most of the compounds with liabilities while still retaining the good ones.

Tomorrow’s wish list
For Deshpande, he knows exactly what he wants to improve lead identification. He says, "I would like to have tools that help me understand the biology better."

Still, even today’s tools give reason for hope. "The good news is that we have opportunities to go after very novel approaches for treatments of very complicated diseases," says Deshpande, "and that’s a direct outcome of the ’omics era." He adds, "The challenge for us is: How do you use that information in designing compounds for lead identification?"

Lackey also holds high hopes for future capabilities in lead identification. For one thing, she’d like to see some breakthroughs in identifying targets from phenotypic screens. Second, she hopes that tomorrow’s technology will also allow multiplexed screening in more natural conditions. "What happens in high-throughout screening is not always relevant to what happens in the body," she says. "So we need primary assays that are closer to physiological conditions."

To find tomorrow’s most effective and safe drugs, researchers seek the most realistic approaches. Developing that overall perception, however, requires exploring compounds and targets from various perspectives.

About the Author
Mike May is a publishing consultant for science and technology based in Austin, Texas.