YouTube Facebook LinkedIn Google+ Twitter Xingrss  

Hans Lehrach’s Predictive Biology Philosophy

Alacris Pharmaceuticals, a new venture with George Church, applies modeling to individualized medicine.

By Kevin Davies

May 19, 2009
| Hans Lehrach, one of Germany’s foremost molecular biologists and a director at the Max Planck Institute for Molecular Genetics in Berlin since 1994, has frequently indulged his entrepreneurial leanings. In the early 1990s, he co-founded Sequana Therapeutics with other genomics rock stars such as Peter Goodfellow (see, “Genomics Provides the Kick Inside,” Bio•IT World, Nov 2003), before its fortunes faded and its assets ended up at Celera. Another venture, GPC Biotech, did well until it suffered what the soft-spoken Lehrach calls “a somewhat catastrophic event” courtesy of the FDA in 2007.

Lehrach’s forte is his gift for developing molecular biology tools and automation, from new gene cloning techniques to chip-based platforms. His latest venture however, in collaboration with Harvard Medical School’s George Church, stems from modeling programs he began writing on transatlantic flights ten years ago. The new company, Alacris Pharmaceuticals, marries next-generation sequencing, systems biology, and computing horsepower, to turn complex biological data into a form that can be used by the researcher, the doctor, and the patient.

“Science is useful if it can predict things,” says Lehrach. “Meteorologists are useful if they can predict the weather! Most of the time, taxpayers aren’t interested in how many high-impact papers somebody publishes, but much more if he can predict something useful.”

Lehrach and colleagues have been working on systems modeling approaches to make predictions out of large amounts of data. Lehrach says there were three reasons to establish Alacris: modeling systems are improving, DNA sequencing is getting cheaper, and third, “GPC fired most of their researchers and many of their executives, and we had volunteers to set up a company!” Lehrach says jokingly.

Lehrach agrees that systems biology hasn’t had a good rap in pharma circles. “Academic systems biology is very much single genes, single pathways,” he says. “It’s a great way to get papers published but not to make decisions about complex situations. Many people are just using it as a buzz word. One group says systems biology is nonsense and it shouldn’t be funded, it’s detracting from cell biology. Another group says systems biology is cell biology because cells are whole systems, so systems biology should fund my work!” Lehrach laughs. “The hard core of systems biology is to make predictions and predictive models.”

Lehrach has known Church since George was a Harvard grad student in the mid 1980s. Lehrach admires Church’s many ventures, noting: “Predictions of cancer drug response is a very low hanging fruit compared to what George is trying with the Personal Genome Project!” The other scientific co-founder of Alacris is Bernhard Hermann, one of Lehrach’s former grad students, who is also a director at the Max Planck. “It’s an enormous advantage to have him involved,” says Lehrach. Hermann is a biologist who thinks about pathways, whereas Lehrach thinks more abstractly. On the business side, the company has a number of co-founding pharma executives and on the medical side is co-founded by top cancer doctors and researchers from the Charité Universitätsmedizin hospital in Berlin.

Computational Biology
Lehrach’s modeling mission is to duplicate the biology of a cell or pathway in the computer. “Each object in biology corresponds to an object in the computer, which basically interacts with all the other objects,” he explains. The starting point was a seminal cancer review paper published in Cell in 2000 by Doug Hanahan and the Whitehead Institute’s Bob Weinberg. Lehrach tried to set up a version of the cancer signaling pathways from the paper in the computer, setting up systems of differential equations that can be solved.

This is not a new idea; so why does it work? “It’s not a problem to model differential equations if you know which constants to put in,” he says. In biology, kinetic rate constants typically aren’t known. So Lehrach takes each unspecified kinetic constant, and performs hundreds of modeling runs using different starting values drawn from probability distributions. “If everything agrees—if one result comes up—that’s probably solid,” he says.

The system, he says, must be tolerant against a lack of knowledge. “The problem up to now wasn’t that we couldn’t solve differential equations, it was that we didn’t know how to handle uncertainty. Many people have tried to generate Boolean networks that pretend that the tumor is a Pentium chip with binary decisions, or basically tried to ignore the complexity and uncertainty.”

The most exciting model so far is that of the EGF receptor pathway, studied in the presence or absence of a downstream RAS mutation. “Using the model, we find that the RAS mutation increases cyclin levels just as much as EGF signaling. This predicts that EGFR antagonists will not have any effect in tumors with RAS mutations.” This is indeed what has been found over the past 12 months by companies such as Amgen and ImClone. (See “Amgen’s Personalized Medicine Story,” Bio•IT World, April 2008). 

Lehrach calls his model “a virtual human, which oncologists can use to try a specific drug for a specific patient if you have the sequencing information.” (See Box.) The first results are “extremely encouraging for predicting how patients react to existing therapies,” he says.

This model is just the first pass. Lehrach is encouraged by the results obtained with just a few hundred runs, letting the computer run overnight in a research institution with a small group. “Once you have 250 runs that give you the same qualitative result, then you can be confident. I’m sure we’ll find better mathematical methods as time goes on to cut down on the computing time.” Additional models are being developed for protein kinase inhibitors to explore how specific tumors will respond.

Lehrach is agnostic on how the sequence information is generated, although he hopes to use the Polonator, the low-price next-gen sequencer based on Church lab technology. When we spoke, however, he had shipped the instrument back to the U.S. for repairs and upgrades. For now, the focus is genomic data, but Lehrach will take any form of data: copy number variants (CNVs, low coverage sequencing of the genome), exome sequencing for the patient, tumor and if possible, the tumor stem cells; and transcriptome data to monitor epigenetic effects. “Proteomic data would be fantastic but it’s still a challenge. We’re working on it,” he says.

Safety Switch 
Ideally, Lehrach says, “we would tell patients not only which drugs won’t work but also, at some point, which drugs will work.” Funding for individualized medicine—sequencing patient tumor genomes, predicting drug response, and developing personalized drugs—could come from the patients themselves or insurance companies.

Another goal is to improve drug safety. Lehrach says the EGFR antagonist trials only worked because they identified a subgroup of non-responders with the RAS mutation. “If we’re able to predict the number of real responders, you can run a trial with ten patients! The same P value can be achieved at much lower cost and much shorter time.” Sponsors would see their market shrink, of course, but the reduced trial costs and increased lifetime of the patent (because of earlier approval) would more than compensate. In short, “the more work we can move to the computer, the better things are going to go.”

Alacris Pharmaceuticals would eventually like to engage with big pharma, but many pharma firms come from the single gene tradition, says Lehrach, so it’s hard to convince them of the value or validity of this approach. Some are interested, however, despite having “been burned by a lot of hot air.” It wouldn’t be difficult to incorporate pharmacogenomic information from the cytochrome P450 genes from the same patient, “so there’s a hell of a lot we can predict from sequence data.”

The name “Alacris” suggests promptness, but Lehrach is gearing up operations methodically. The company was founded last year—to say it has “launched” would be an overstatement, Lehrach admits. So how does he go about raising money in this economy?
Lehrach laughs: “Any suggestions?” 

CollabRx ONE, TREAT1000: the Cancer Connection

CollabRx was set up by Jay “Marty” Tenenbaum and Raphael Lehrer to slash the cost and time of developing drug therapies (see “Collaboration and the Long Tail of Disease,” Bio•IT World, March 2009). Its new service, CollabRx ONE, applies those resources to help identify bespoke therapies for late-stage cancer patients. The goal of the service, which costs between $50-100,000, is to provide a deep understanding of the individual patient’s disease by marrying genomic and computational analysis, and match the aberrant target or pathway with a potential therapy.

“In every case we have actionable hypotheses that the doctors have not previously considered. It’s not cheap, but we’re working hard in order to be able to reduce costs… and we’re actively seeking collaborations with major medical institutions,” says Tenenbaum.
Lehrer, who is in charge of CollabRx ONE, has been a friend of Tenenbaum’s for 20 years. After getting his PhD in physics from Harvard, Lehrer left research and moved into consulting and biotech, spending five years at Gene Logic working on toxicogenomics platforms and drug repositioning efforts, before finally joining forces with Tenenbaum.

CollabRx ONE has an informatics platform that builds tools for integrating a variety of data streams—gene expression, SNP analysis, copy number variant (CNV) analysis, sequence data—and potentially provides decision analysis. The small informatics group in California is led by co-founder and chief technology officer Jeff Shrager (best known for writing an application called BioBike). Meanwhile, the wet-lab analysis is outsourced to CLIA-certified labs.

Lehrer says the innovations are chiefly in trying to distinguish signal from noise using so few samples. “Happily I’m a physicist,” he says. “Hopefully you create ideas you happen to have training for!” The informatics group handles tasks such as pathway analysis (using Ingenuity’s IPA) and data visualization, customizing ways of visualizing the data given the variety of data types.

Once the analysis is complete, CollabRx ONE staff meet with the patient’s oncologist and discuss their findings, hopefully to advise on potential drugs or drug combinations, based on information on drugs that are either FDA approved or in clinical trials. Tenenbaum says his group is applying selective use of proteomics and “trying to understand the values of whole-genomes sequencing.”

CollabRx has formed a joint project with Alacris Pharmaceuticals called TREAT1000, to add whole-genome sequencing to the CollabRx ONE offering. By sequencing 1000 genomes of cancer sufferers, TREAT1000 will not only provide potentially life-saving information about individual cancers, but create a compendium of cancer genome information that will inform future research and treatment.

What the Data Say
“We look for what the data are telling us, and how that matches the therapies,” says Lehrer. Leading the analysis is Bob Coopersmith, a former Gene Logic colleague. “We push back and forth conclusions and alternate explanations, discuss discrepant observations,” says Lehrer.

Although the project is in its early days, Lehrer says progress based on the first half-a-dozen patients is extremely promising. “It’s important for us to connect with the oncologist and that the oncologist buys into what we’re doing.” In nearly every case, “the oncologist has been pretty excited,” although it’s too early to predict how that might translate into clinical success.

The work bears an emotional toll. One patient, thought to have 12 months to live, died in a matter of weeks before the team could implement their findings. In other cases, there are signs of a partial, but only partial, response to drug. Nevertheless, for Lehrer and Tenenbaum, that’s encouraging. In the case of a lung cancer patient on Avastin, CollabRx analysis revealed the likely involvement of two different key pathways. “The second was going untreated, and that suggested a combination of drugs, or one drug that could hit both pathways.” But the oncologists must follow the standard of care, which typically means changing the drug regimen only after the patient becomes fully resistant. 

The recommended drugs are likely to be off label. “They may or may not be cancer drugs. They may or may not be generic. They may or may not be reimbursed,” says Lehrer. “Oncologists may have tried it, but the reaction is often, ‘I wouldn’t have considered using that kind of drug, but based on what you’re showing me, that makes sense.’”

Lehrer stresses that the results are always communicated to the patient’s oncologist, “because they are the ones who need to decide whether what we’ve found is something that should be tried, or if additional studies should be done. Depending on circumstances, it is valuable to have the patient there as well.”

CollabRx ONE is continuing to evaluate new technologies, studying whether a potential data source can add value to an actionable hypothesis. Lehrer says: “If it is adding value, are previous sources now redundant? Obviously it’s not worth doing SNP analysis if you’re doing whole-genome sequencing.”  K.D.

This article also appeared in the May-June 2009 issue of Bio-IT World Magazine.
Subscriptions are free for qualifying individuals. Apply today.

Click here to login and leave a comment.  


Add Comment

Text Only 2000 character limit

Page 1 of 1

For reprints and/or copyright permission, please contact  Jay Mulhern, (781) 972-1359,