An Emerald in the Rough

By Kevin Davies

January 19, 2010 | Late last year, scientists at Emerald BioStructures, located on Bainbridge Island just across the Puget Sound from downtown Seattle, experienced the bittersweet ride that so often characterizes the drug discovery business. In August 2009, a team led by CEO Lance Stewart published a major paper outlining the application of a Fragments of Life (FOL) drug discovery approach to help identify an exciting small-molecule drug candidate, DG051, in the Journal of Medicinal Chemistry. But publication came just a few months before Emerald severed ties with deCODE genetics after the Icelandic firm filed for bankruptcy in November (the company was formerly deCODE biostructures) and was acquired by Beryllium LLC.

Newly independent, Stewart has the goal of expanding its drug discovery services in the field of gene-to-structure research. Emerald bills itself as “the largest gene-to-structure CRO in the U.S.” One of its success stories was the development of DG051, one of two drug candidates it helped deCODE to identify. DG051 has completed Phase IIa studies, and the second drug candidate, DG071, is approved for entry into Phase I safety studies.

The DG051 story is remarkable for the speed with which deCODE’s biologists and chemists and Emerald’s structural biologists sped from target identification to IND in just 2.5 years, which might be a record.  Yes, there were some fortuitous bounces along the way, but Stewart believes his high-throughput crystallography approach provides an attractive alternative to conventional high-throughput screening (HTS) techniques. It is, to his knowledge, “the first example of a human genomics target discovery  that was followed up by a gene-to-clinic paradigm enabled by a fragment-based drug discovery effort.”

deCODE’s strength was using genetic mapping studies to identify at-risk haplotypes for a host of common diseases. Several years ago, studies identified two genes in the leukotriene biosynthetic pathway that were incriminated in myocardial infarction. In January 2006, deCODE published a report that markers in the gene for LTA4H (leukotriene A4 hydrolase) were associated with increased risk of myocardial infarction, particularly in African Americans. After deCODE confirmed that those at-risk haplotypes were associated with higher levels of leukotriene B4, it targeted LTA4H as a viable drug target, bringing to bear the full capabilities of structure-based drug design (SBDD) and fragment lead identification, in a full-force effort to identify leads as quickly as possible.

By August 2006, DG051 was in Phase I clinical trials. “As far as we know, this molecule is the furthest along in a true genetic-underpinning, SBDD approach,” says Stewart. As of the end of 2009, DG051 had finished Phase 2A. Presently, the DG051 asset is tied up in deCODE’s bankruptcy proceedings.

Among the keys to identifying and developing DG051 so quickly was a fully integrated organization that saved a lot of time and money with integrated process chemistry, analytical chemistry and ADME-tox efforts. It wasn’t exactly under one roof—deCODE’s facilities were split between Seattle (now Emerald BioStructures) and Chicago (deCODE chemistry)—but that was a minor inconvenience. Stewart’s high-throughput crystallography operation dovetailed neatly with deCODE chemistry’s structure-based drug design, managed by president Alex Kiselyov.

When work began, there was little crystal structure data on LTA4H and only a few known inhibitors (from Searle). In all, Stewart’s group solved about 50 crystal structures, focusing on higher resolution, improved crystallization conditions, and structures with bound ligands. The goal, said Stewart, was “to make 100% sure we understood how fragments would interact with the target. We wanted to identify new chemotypes that could inspire chemistry on LTA4H.”

An important in-house advantage for the crystallization work was software called Gene Composer, which is used to design and optimize the codon usage in the gene depending on the nucleotide bias of the expression system. “When you’re doing X-ray crystallography, you’re doing a lot of protein engineering, and so you’re better off synthesizing the gene,” says Stewart. “So you may as well tweak and engineer the overall molecule.”

Synthesize This
While Stewart’s team was crystallizing the target, Kiselyov was leading the effort to identify the lead. “We were living the paradigm of moving therapeutic discoveries from gene to clinic,” says Kiselyov. Seeing DG051 finally reach the market would be “the key validation of this approach.” DG051 is actually 4-[(2S)-2-{[4-(4-Chlorophenoxy)phenoxy]methyl}-1-pyrrolidinyl]butanoic acid (See, Journal of Medicinal Chemistry. 2009 Dec 1).

Emerald used its FOL library of ~1,500 small molecules to initiate discovery of the DG051 drug candidate using high-throughput crystallography, rather than use the traditional funnel of HTS to screen large multi-million compound libraries.  “It’s a bit like a Lego kit; every component makes sense and fits this selection.”

Emerald gets “the biggest bang by doing structural biology and in parallel, doing the small-molecule chemistry. If they’re timed right, then like a Reese’s peanut butter cup, they come together.” The key to successful crystallography, he adds, is to feed the protein a well designed ligand, which aids the visualization of the enzyme-ligand complex.

Stewart’s iterative crystallization process produced a much more dynamic picture of the protein target in the presence of candidate small molecules. That enabled Kiselyov to do both docking and optimization studies. “Small molecule fragments bound to the target provided us with critical structural information for the design. Think of them as tiny pieces of clay, a molecular 3-D imprint of a specific protein sites. Previous art in the field, for example research done at Berlex and Searle, was also considered in a final assembly of these fragments into the ultimate clinical candidate, DG051.”

“Instead of making thousands of compounds guiding us through the optimization process, synthesizing a matrix of about 500 molecules across four chemotypes was sufficient to identify a winner,” says Kiselyov. Notably, the deCODE team identified multiple chemotypes that occupied the buried active site of the target LTA4H enzyme. Fragments had to be lined up within three critical areas within the binding pocket—the catalytic zinc atom, a hydrophobic region, and a transitional kink or linker. “Since we identified the mode of binding and simultaneously optimized physiochemical and PKPD/tox parameters for our lead candidate, we didn’t even have to go to our backup. DG051 was good enough to push instantly from bench to a GMP production and clinic.” It took about 12 months from the initial structural biology efforts to finally settle on DG051.

One of the key observations was that the catalytic zinc atom liked to bind acetate, thus providing a hint regarding pharmacophore requirements for this portion of the enzyme. A rigid, prolinol derivative provided for a linker to combine a biaryl fragment with the acetate fragment culminating in DG051. Kiselyov’s team attempted various medicinal chemistry modifications, but most of them worsened the toxicology or metabolic properties of the molecule. “It’s almost like nature itself guided us and told us do not do silly things with this molecule,” he said. DG051 had outstanding specificity, with no HERG triggered toxicity. The molecule has very favorable drug-like properties including molecular weight, and production scaled nicely. Bioavailability was excellent in animal models and in human. The half-life was 9 hours, favoring a single daily clinical administration. There was also good evidence that DG051 affected levels of the key biomarker associated with the acute myocardial infarction and stroke.

From Genes 2 Drugs
DG051 is currently held up in Phase II trials. “We don’t know too many people who get molecules from scratch into the clinic in a safe way. Of course, efficacy is to be shown still, but we’d love to see them tried in phase 3,” said Stewart.

Another drug, DG071, did not progress as swiftly through development because it required some “heavy lifting” in structural biology, chemistry and enzymology. Additional crystallography information was required to picture the regulatory domains.

Stewart has seen upheavals in the market before. The predecessor to Emerald was a company called Medichem which acquired
Emerald in 2000 and completed the last life science IPO as Medichem Life Sciences just before the 2001 market crash.  In 2002, deCODE acquired Medichem and Emerald to run a hybrid business model, running contract research for multiple clients while executing a handful of drug discovery projects for itself.

So why was the development of DG051 so successful? Part of the answer was in the way the project was managed. Stewart says his group was driven by milestones and specified timelines. He says deCODE had a “gene to IND capability. We have goals and set timelines. We wanted an IND in two years... We wanted to meet those goals. When you put a number of shots on goal, in an integrated way, you don’t know when but you’ll catch a break somewhere.” Emerald did indeed catch a break—the acetate binding to catalytic zinc was deemed too good. “We still did more and more chemistry, thinking we could do better, but we didn’t need to do all that work.”

Another critical factor at deCODE chemistry, says Kiselyov, was having the process chemists working closely with the medicinal chemists to jump into large-scale production. That helps rule out building blocks that are too expensive or commercially unavailable, and avoid the possibility of toxic impurities. In the event, deCODE chemistry was able to do just a 4-step synthesis from commercially available building blocks, and scale up production before the IND.

Stewart says the FOL collection creates new IP and the ability to identify new druggable sites, which in turn leads to selectivity. And with so many constructs, he could look at the same lead molecule in different protein environments as he considered the medicinal chemistry.

“I’m way too humble to say we’re the best in the field,” says Kiselyov. He quotes a phrase from Fiddler on the Roof: “We’re just trying to scratch a simple tune here.” It’s still the early stages of using fragment based crystallography for SBDD. “Whatever our fate, we’re somewhat monolithic,” says Kiselyov. “If we can take an IND, that would be ideal. deCODE has a drug candidate in DG051 that has completed Phase2A. That is what pharma is or will be looking for.” 

Further Reading:
Davies, D.R. et al. “Discovery of leukotriene A4 hydrolase inhibitors using metabolomics biased fragment crystallography.” J. Med. Chem. 52, 4694-4715 (2009)
Sandanayaka, V. et al. “Discovery of 4-[(2S)-2-{[4-(4-Chlorophenoxy)phenoxy]methyl}-1-pyrrolidinyl]butanoic Acid (DG-051) as a novel leukotriene A4 hydrolase inhibitor of leukotriene B4 biosynthesis.” J. Med. Chem. December 2009.