Drug Design in silico
OSUCCC – James researchers have used computers and computation to develop a new class of drugs that targets a key enzyme they discovered 10 years earlier in cancer cells.
BY DARRELL E. WARD
The enzyme PRMT5 is a key regulator of cell growth and proliferation during embryonic development. When the gene’s assigned task is over, the cell reduces its expression, and the enzyme’s levels become barely detectable.
As cells become cancerous, however, the low levels of mRNA of this dormant gene are translated more efficiently, and the enzyme is produced in abnormal abundance. In 2003, Saïd Sif, PhD, a molecular biologist and biochemist with The Ohio State University Comprehensive Cancer Center – James Cancer Hospital and Solove Research Institute (OSUCCC – James), showed that overexpression of the gene is key to the hyperproliferation of cancer cells.
PRMT5 is an enzyme that adds methyl groups to histones and other proteins. Specifically, Sif showed in 2004 that it adds methyl groups to histone proteins H3 and H4, and that this leads to the shutdown of important tumor-suppressor genes and promotes tumor growth.
About that time, Robert Baiocchi, MD, PhD, joined Sif ’s laboratory and began cell and animal studies of PRMT5. Baiocchi, who today has his own lab at the OSUCCC – James, and Sif found that overexpression of PRMT5 promotes survival and proliferation in high-grade lymphomas, glioblastomas and melanomas, and in a wide range of cancer cell lines.
They and others have since shown that PRMT5 is astonishingly versatile and a central enzyme in cell growth and cancer development. It regulates several pathways involved in cell growth and survival, epigenetic regulation of tumor-suppressor genes and even the synthesis of cell organelles.
They also have shown that overexpression of PRMT5 promotes invasion and metastasis, and that blocking the enzyme stops cancer-cell growth and prevents migration and invasion.
One summer day in 2008, Baiocchi walked briskly down 12th Avenue, the Medical Center’s research corridor, to the College of Pharmacy. There he met with computational biophysicist, in silico drug designer and OSUCCC – James researcher Chenglong Li, PhD, about developing a PRMT5 inhibitor using in silico drug design methods.
“PRMT5 plays a key role in regulating the cell cycle in cancer cells,” Baiocchi explains, “and it is expressed at very low levels in healthy adult cells. It’s an ideal target for a small-molecule inhibitor, and a PRMT5 inhibitor could be an effective cancer therapeutic for stopping tumor growth.”
Following that meeting, Li, Baiocchi, Sif and a large group of collaborators went on to develop a first-in-class PRMT5 inhibitor that is currently in preclinical testing, and they have a third-generation agent under development.
But it wasn’t easy.
In Silico Drug Design
In silico drug design uses the power of silicon-chip computer circuitry to design new targeted agents. Using computers, Li and his lab pull drug molecules apart, sort the fragments by chemical attribute and shower them onto the active site – the business-end – of target molecules. The fragments that stick are potential inhibitors.
It’s a dazzling process of computation, but hardly perfect. Nature is complicated. So OSUCCC – James medicinal chemists make small quantities of promising candidate agents and send them to Baiocchi, a biologist, who tests them for specificity and potency using enzyme and cell assays. If one is outstanding, Li and his colleagues computationally fine-tune the molecule to produce a second-generation agent that also undergoes biological evaluation. When all goes well, this process leads to phase I clinical studies and a new anticancer drug.
“Typically, about 5,000 compounds must be discovered and tested to get one FDA-approved drug when in silico drug design is not used,” Li says. “The use of in silico drug design can reduce the number of compounds to 500, or even 100, and produce three to five agents to choose from for preclinical evaluation and possible clinical trials testing.”
The accomplishments of Li and his lab include designing three inhibitors for STAT3, a central signaling protein that is overexpressed in cancers of the breast, prostate, lung and pancreas, and in myeloma, and devising a new docking simulation for modeling molecular binding. The STAT3 inhibitors are in preclinical testing.
Anticancer in silico drug development begins with a validated target molecule and its atomic structure (see sidebar), which normally is derived using X-ray crystallography or other technology. PRMT5 was a validated target, but the protein’s molecular structure was unknown. That problem had stymied attempts by others to develop an inhibitor for this enzyme.
Li and his colleagues derived the
3-D structure for the PRMT5 active site computationally, basing the model on the crystal structures of four homologous PRMT proteins: human PRMT3, mouse Carm1, and rat Prmt1 and Prmt4.
The active site is the all-important pocket in the enzyme where another molecule – the enzyme’s substrate – docks, or binds, with the enzyme and makes a chemical reaction happen. For PRMT5, that chemical reaction transfers methyl groups to arginine residues in histones H3 and H4.
Li’s ultimate goal was to design a small molecule that will readily occupy the PRMT5 active site, blocking the normal molecule from binding with the enzyme. This neutralizes the enzyme and triggers cancer-cell death.
This first “homology model” of PRMT5 was too crude to use for drug-development research, so they fine-tuned its atomic framework using a method called molecular dynamic simulation. In the end they showed – computationally – that their model pocket would bind both the normal human PRMT5 substrate and its cofactor. Their model of the human enzyme was accurate to its very atoms.
Now they could search for a small molecule to gum it up.
Li and his colleagues began the hunt by downloading the molecular structures of more than 1,500 Food and Drug Administration-approved drugs, and 7,000 to 8,000 experimental drugs in clinical trials testing from public databases. Then, Li says, “We used computers to chop these nearly ten thousand drugs into pieces and sort the pieces into subgroups according to structure, solubility, acidity and other properties.”
Then they flooded the active site with 200 to 300 of these fragments, along with nearly 1.5 million organic molecules from a bank of chemical structures that Li maintains. They did this using a computer program developed by Li called Multiple Ligand Simultaneous Docking (MLSD). It matched each molecule against the pocket according to three criteria: binding energy, binding mode and binding statistics. The Ohio Supercomputer Center crunched the numbers almost around the clock for nearly four weeks.
These high-throughput screenings identified eight molecules that fit the pocket.
Next, medicinal chemists in the College of Pharmacy and the OSUCCC – James Medicinal Chemistry Shared Resource formulated a small quantity of the eight compounds and sent them to Baiocchi and Sif. Sif studied the molecular interactions of the potential inhibitor, and Baiocchi screened them for specificity using an enzyme assay and glioma and lymphoma cell lines. One of the eight proved highly specific for human PRMT5.
That molecule consisted of two molecular fragments, and it became the starting point, the lead (rhymes with “seed”) molecule. The researchers called this first-generation PRMT5 inhibitor BLL1. “We did a lot of work in collaboration with other labs to develop this molecule,” Li says.
The Binding-Side Tango
Next, Li and his lab set out to optimize the crude inhibitor for specificity and potency, an inhibitor’s two most important qualities. That is, they wanted to strengthen the bond between the inhibitor and the PRMT5 active site. This interaction involves the six “weak chemical forces” that hold molecules together (see sidebar). These are noncovalent forces that individually are 10 to 100 times weaker than the covalent bonds that join atoms, but together they can hold molecules together.
“Much of the molecular biophysics in in silico drug design is devoted to simulating the nonbinding interactions between molecules,” Li says. “It is one of the most challenging, computationally involved and intriguing aspects of in silico drug design.”
Li uses molecular dynamic simulation to estimate the six forces, and this predicts how long the inhibitor molecule will stick to the active site. “If these forces match, the inhibitor will occupy the active site for a sufficient period; if they don’t, the drug will fall off quickly,” he says.
The six forces are influenced by molecular motion, Li says, noting that molecules can shoot straight ahead, rotate, flex and vibrate. “Molecular dynamic simulation accounts for these motions,” he says.
“Inside the cell, the target protein and the drug molecule bounce and weave around one another, but they won’t come together specifically if they don’t recognize each other,” Li says. “If you design a drug that is potent and specific, the drug will move around the protein and at some point the two will hook together.” The drug doesn’t sit there for long, though. Molecular motion causes it to repeatedly pop in and out of the active site. “The small molecule dances around the binding site,” Li says.
This molecular dance also relates to drug toxicity. “The ideal drug will stick to the target active site, but it could stick to other molecules, too. You don’t know,” Li explains. “The binding of the inhibitor to the target must also be reversible. If something stays there forever, it could be deadly. So we must computationally model this, too.”
If an agent is potent and specific but too toxic, Li and his colleagues will tweak the molecule further. “Changing an atom here or there will sometimes make the agent less toxic or more stable,” Li says, “but then we have to check that it is still potent and specific. If not, we have to start over and redesign the molecule.
“An in silico molecule is always an approximation,” Li adds. “This is not like designing a building or a bridge because the nonbinding interactions are extremely hard to predict.”
The biological and biochemical evaluations by Baiocchi and Sif showed that the second-generation PRMT5 inhibitor, called BLL54, was stable, water soluble and had low toxicity.
BLL1 and BLL54 are in preclinical testing, and Li is developing a more potent third-generation agent. Their studies so far have shown:
- That PRMT5 is overexpressed in mantle cell lymphoma (MCL), Hodgkin’s lymphoma, diffuse large B cell lymphoma, Burkett’s lymphoma, glioblastoma multiforme, lung (small and non-small cell) and melanoma, suggesting that the target is relevant to aggressive solid and hematologic tumors;
- That PRMT5 partners with other proteins to silence multiple regulatory and tumor-suppressor genes;
- That BLL1 inhibited PRMT5 activity in cancer cells without affecting normal cells, and that this slowed or stopped cancer cell growth;
- That both inhibitors are safe, metabolically stable and effective in an animal model;
They also have developed a transgenic mouse that overexpresses PRMT5 and develops lymphoma, providing a preclinical model for evaluating inhibitors.
At the 2011 American Society of Hematology meeting, Baiocchi presented findings about the use of BLL1 as a novel experimental therapeutic in MCL. Their study, selected as one of the meeting’s top oral presentations, showed that BLL1 greatly reduced the expression of molecules involved in MCL development, and that 46 patient samples examined showed “abundant PRMT5 expression” in both the cytoplasm and nucleus.
In addition, Baiocchi says, “We’re using these novel inhibitors as tools to tease out the biology of lymphomagenesis.” For example, the Ohio State researchers have shown that PRMT5 is essential for Epstein-Barr virus to transform B cells, and that blocking PRMT5 with BLL1 inhibits that transformation.
Baiocchi, Sif and Li are collaborating with other OSUCCC – James investigators to study the inhibitor in melanoma, breast cancer and glioblastoma. “We’re looking at the role of this enzyme and these inhibitors in a range of solid and hematologic cancers,” Baiocchi says. “We hope to begin a phase I clinical trial for one of their PRMT5 inhibitors within five years.”
Last but not least, Li and his lab are working with Sif ’s lab to generate crystals from PRMT5. “We’re close to having the first crystal structure of human PRMT5, which would be a huge accomplishment,” Baiocchi says. “This will help validate what we’ve done to date, and it will help us find more potent inhibitors.
“We have a great multidisciplinary team working on this inhibitor,” he says, “and I think it’s going to help push the OSUCCC – James drug-development program to a new level.”
In Silico Drug Design in Seven Steps
1. Identify a validated target molecule such as a tyrosine kinase, cell receptor or a signaling pathway.
2. Generate the 3-D structure of the target molecule’s active site using data from a protein data bank or X-ray crystallography, or derive it computationally.
3. Identify a lead inhibitor molecule by computationally plying the target active site with fragments of known molecules.
4. Evaluate the putative inhibitor for specificity and potency in bioassays.
5. Optimize the lead molecule for specificity and potency to produce a second-generation agent.
6. Evaluate the second-generation inhibitor forspecificity and potency in bioassays.
7. Optimize the molecule further to develop third- and fourth-generation.
The Six Weak Forces Considered in In Silico Modeling
While covalent bonds hold atoms together to make a molecule, nonbinding interactions, or weak chemical interactions, hold two molecules together. There are six weak chemical interactions, or forces, that in silico drug designers simulate to determine the specificity and potency of a potential targeted inhibitor.
1. Van der Waals attractions – This force relates to molecular shape.
2. Electrostatic force – These are ionic bonds between atoms.
3. Hydrogen bonding – This bond occurs when two electronegative atoms share a hydrogen atom.
4. Polarization force – A force produced by shifting densities of moving electons.
5. Desalvation – A layer of water surrounds a target molecule. Desalvation refers to the force required to displace this layer of water when a drug binds to a target molecule. “This is one of the most difficult forces to model,” Li says.
6. Entropy – This force relates to molecular vibration, rotation and translation.