Feb 15, 2006 | Think Intel Inside, but instead of microprocessors, think industrial-strength experimental platforms, focused on all things omic, designed and even run by a third party, but living deep inside biopharmaceutical companies. Need high-volume, highly accurate proteomic or metabolomic data? No problem. Need to separate the signal from the noise? Can do. Worried about keeping pace with omic experimental technology? Don’t.
This is Pieter Muntendam’s embedded BGM concept. Muntendam is a Dutch-trained physician who’s logged many years at pharma and biotech (J&J, Glaxo Wellcome, Organon, and Millennium). He has also just finished his first year as president of BG Medicine
(BGM) and is presiding over another reinvention of the systems biology company. He cringes at the latter suggestion, but says stoically, “Companies need to be prepared to reinvent themselves every 90 days.”
BGM began life in 2000 amid much fanfare as Beyond Genomics, a pioneer in omics data generation and interpretation. Its founding leadership — Stephen Naylor (from the Mayo Clinic), Jan van de Greer (from Leiden University), and Eric Neumann (from 3rd Millennium) — was impressive and multidisciplinary. Flagship Ventures and Gilde Investment Management provided investment in excess of $26 million.
The combination of smart folks and funding quickly produced a powerful experimental platform, but the now-familiar refrain of weak demand for unproven technology prompted business model shifting and staff shuffling. In February 2005, the company was renamed BG Medicine, with Muntendam as a change agent at its helm.
“It had an instrument phase where they actually developed their own new mass spec. A number of people who were associated with the company came out of the instrument world. Then they said, no, we don’t want to be an instrument company, we want to build out the platform to become a drug company. And there was this service component in particular disease areas. That’s the time I came and said this is a broad-spectrum platform, you’re way ahead of where the industry is,” says Muntendam.
Companies come to BGM with biological specimens, such as plasma, a variety of tissues, and fluids, and “within 120 days, we get systems biology done — proteomics, metabolomics, integrated and extract information from it,” says Muntendam.
During his first year, Muntendam moved the company’s location, returned its focus to HT experimental services, and settled on three-prong go-to-market strategies.
“One [offering] is we’ll do a project for you. Bring us the samples, involve us in the design, and we’ll do it for you. The second is we can get the organization BG Medicine-ready. You want to do transcripts internally, maybe you want to do NMR internally, but let us help you with those platform requirements that you don’t have, including our computational efforts. The third is this concept of embedded BGM where companies realize this is a core competency they want to bring in-house.”
What’s different now, says Muntendam, is pharma’s readiness and need for robust omic experimental capacity. And transcripts alone won’t cut it. Citing the Vioxx debacle, he says, “If you listen to pharma and to the FDA, then by 2007 or 2008 most of the drug development will be with the black box open. People want to have characterized drugs’ on- and off-targets. If pharma wants to get there by 2008, they need a production-ready platform.”
The so-called black box problem — it’s not known how most drugs really work or what undesirable targets they hit — has proven intractable. Muntendam believes BGM’s brand of wet-lab-based systems biology will be able identify decisive biomarkers for toxicity and efficacy. He also thinks the BGM Inside concept will cut costs, risks, and time-to-ramp-up issues for biopharmas seeking crowbars to pry open the black box.
Others are listening. Last fall, BGM won a CRADA from FDA’s National Center for Toxicological Research (NCTR) to find preclinical markers of clinical drug liver toxicity.
Liver toxicity is generally the most important biological event in drug failures — about half of which is preclinical toxicity and half of which is clinical toxicity, meaning the preclinical model was entirely clean and went forward but showed LTL elevations in Phase II. “These obviously are late-stage failures. They’re expensive. We’re saying there’s a very high probability that with the right biomarkers we can detect these on the basis of a 28-day rat study,” says Muntendam.
The plan is to do five pairs of compounds. At this writing, expert discussions on which compounds to use hadn’t concluded. The goal is to choose some that were preclinically clean and later got into trouble. “There’s a good list of these. We’ve just got to make sure we use the ones that are most representative, carry the broadest value,” says BGM’s president.
BGM will perform the proteomics on individual samples, NCTR will do gel-based mass spec on pooled samples, FDA will handle NMR and transcripts. BGM will interpret the data using combined bioinformatics approaches including its own software and FDA ArrayTrack, the tool used by FDA’s voluntary genomics data submission program. Muntendam is actively soliciting pharma participation as they own the compounds and have the cash.
“We have active dialogues. It’s clear that there’s no federal money for FDA modernization — that is the critical path. There is no critical path. There’s some talk earlier this year that there was going to be a critical path budget, but Congress and Katrina and whatever else cut it.”
The program seems like a clear win-win for participating drug companies, who would receive licenses to the biomarkers and all project data including any developed screens. But the data will be also submitted to FDA’s VGDS (voluntary genomics data submission program), and that may worry some potential collaborators.
Muntendam says simply, “Show me the number of compounds where the FDA advisory committee said you have great drug, it’s highly effective, highly safe, but you’ve done some transcripts, and we’ll hold up [approval based] on the transcripts. Companies should have gotten the message that the better you can characterize your drug, the better you understand why it works, or doesn’t work, will help you in the process.”
A strong CRADA result would boost confidence in BGM’s technology. The company has already worked with several major pharmas (AstraZeneca, GlaxoSmithKline, and Novartis are listed on its Web site). But convincing them to let BGM come in and design, set up, and potentially run their omics labs is a stiff challenge.
No customer has signed up yet for BGM Inside, but Muntendam says he’s in advanced discussions with several who’ve been happy with earlier BGM work.
BGM’s secret sauce is its rare combination of wet-lab expertise and massive
data-handling prowess. “We’ve made tremendous progress converting basic science into robust industry workflows,” notes Stephen Martin, BGM’s chief technology officer and former ABI proteomic veteran.
Variation is the chief gremlin plaguing high-throughput experiments, and BGM goes to great lengths to reduce it. For example, the company uses three different quality protocols to ensure results from its metabolomics platforms. It’s also developed powerful software. Raw LC/MS data files are peak picked and integrated using proprietary software, IMPRESS, and retention time alignment for analyte peaks is handled by EQUEST, more homegrown software.
Proteins and Profitability
In its early days, BGM went so far as to develop its own proprietary techniques for handling high-abundance proteins. Now it relies substantially on commercial instruments, which it tweaks as needed.
“Mass-spec-based technologies are inherently are sensitive to environmental exposures, machine issues, and operator behavior,” says Muntendam. “Most people use them qualitatively, but the moment you want to use them quantitatively and in group comparisons [such as] before and after, low dose versus high dose, you’ve got to get your coefficients of variation down in the teens. That requires a degree of diligence you [don’t] typically find in a discovery environment.”
Another challenge is integrating different platforms. “Even if the company has a perfect proteomic setup, that is the runs are high quality, you need to integrate that with LC/MS or NMR. In pharma these [activities] have different owners. They might reside in different countries, under different leadership, and now you’ve got to get the groups to work on one protocol in production-type environment,” Muntendam says.
A typical BGM project involves 200 primary samples and generates two to three terabytes of data. The FDA project is expected to produce 20 terabytes. Data storage isn’t that hard, says Muntendam, but the computational analysis is. “We very early on integrate all of that data, the transcripts, the proteins, the metabolites, the clinical data, and say which of these measurements — and there are thousands — carries a signal.”
What the signal investors wish to see is profitability. Muntendam says that’s coming in 2006. Currently BGM has a staff of 35 in the United States and 5 FTEs at a European nonprofit organization. The CRADA project is expected to wrap up by year’s end and could draw favorable industry attention and partners. Meanwhile, Muntendam is busy stumping the BGM Inside proposition and trying to change a paradigm.
State of the Art
While a lot of art remains in the science of high-throughput omic experiments, BG Medicine is working to industrialize the process. Here’s the palette of key instruments chosen by BGM to bring robustness to proteomic and metabolomic research:
Waters — QTOF MS/MS employed for metabolomics and proteomics
Thermo Finnigan — LTQ, LTQ FTMS (partner TNO Pharma)
Applied Biosystems — *4700 MALDI MS/MS, 4800 MALDI MS/MS
Applied Biosystems — Vision Multidimensional LC Systems Proteomics (abundant protein removal, strong cation exchange)
Dionex — LC Packings LCMALDI sample separation and deposition to MALDI plates
Waters — HPLC systems for ESI proteomics and metabolomics
Covaris — Tissue disruption system
* BGM has written software to handle the output of proteomics data from the ABI 4700. This software automates the process of selecting peaks from mass spectra that will be characterized by MS/MS.
Photo by Michael Manning