The term “systems biology” may mean different things to different people, but that’s not keeping pharmaceutical firms from adopting systems biology approaches to drug discovery and development.
At the IBC Systems Biology conference, held Oct. 29-30 in Boston, speakers detailed a range of approaches to describe, analyze, and model systems ranging from single pathways to entire organisms, categorized as various combinations of “top-down,” “bottom-up,” “data-driven,” or “hypothesis-driven.” Some speakers took great pains to pin a concrete meaning on the term, while others pointedly refused to define systems biology at all. Despite the identity crisis, however, one consistent theme ran throughout a meeting that one speaker described as a “lovefest”: Pharma is eagerly sampling from the systems biology toolkit — however its scientists choose to define it — in an effort to fill a dwindling drug development pipeline.
Pharmaceutical firms have realized that “we must transform the way we do drug discovery and development,” said Michael Luther, vice president of high-throughput biology at GlaxoSmithKline. The one-target/one-drug/one-effect “reductionist approach” to drug discovery “has not been successful,” Luther said, driving companies like GSK and others to recognize that “the target is only one component of a system” that must be understood in order to find new drugs.
But getting a handle on that system — whether it’s a metabolic network, a single cell, or a model organism — requires large amounts of data from various sources along with robust computational methods to bring that information into a single, understandable view. As Harry Harlow, senior research scientist in Eli Lilly’s systems biology group put it, systems biology is essentially “large-scale data integration.” Harlow’s team integrates information about DNA, RNA, proteins, and metabolites, using cluster correlations and correlation image mapping to generate web-based interactive reports that are easily understood by biologists. While genomics technologies are well-suited for target discovery, the network-based approach that characterizes systems biology is better suited for drug development, said Harlow, who added that he views genomics as “hunting and gathering” compared to the more civilized “farming” of systems biology.
Novartis is also adopting a systems biology strategy that relies heavily on effective data integration. Yong-Chuan Tao, senior scientist in the life science informatics unit at the Novartis Institute for Biomedical Research, described a database the company is building to integrate high-throughput screening data with genomic information. Using overexpression of cDNAs and inhibition with siRNAs, Novartis runs cell-based screens to “mimic disease states” in order to capture data on gene function. Screening data, gene expression data, proteomics data, and information from the scientific literature and the KEGG database are integrated by mapping to a single “meta-ID” for each gene, Tao said.
Novartis is outsourcing some of its systems biology work, as well. Robert McBurney, CSO of Beyond Genomics, presented some data from a research collaboration between BG and Novartis on coronary artery disease. Burney — who said the goal of systems biology is to “understand, describe, and manipulate a system” — discussed a “composite system descriptor” for 12 metabolites that the company used to detect biomarkers, along with a method for comparing fold changes in metabolite ratios for healthy vs. diseased and treated vs. untreated samples. Eric Neumann, VP of bioinformatics at BG, said the company is also drawing cross-species comparisons into its systems biology mix in order to design better mouse models. The company has found that metabolites are an effective way to map biological characteristics between mouse and human models, Neumann said, because metabolites are exactly the same across species, unlike DNA and proteins.
In another pharma partnership, Didier Scherrer of Entelos described the company’s asthma research collaboration with Pfizer’s R&D group in Paris. As one of the few projects to take a “top-down” approach to systems biology — based on clinical and phenotypic data rather than genomic data — Entelos used its PhysioLab disease modeling platform to create five separate “virtual patients” that exhibit the asthma phenotype in order to prioritize asthma targets for Pfizer. Scherrer said that it took only 18 months to build the models, evaluate three targets, and prioritize drug candidates against two of those targets in silico. The advantage of virtual patients, Scherrer said, is that “you can be very mean to them; they never complain.” Consequently, the research team was able to manipulate doses and dosing regimens using the models well before the compounds entered clinical trials.
Building the Systems Biology Parts List
Other conference speakers focused on efforts to generate the raw material of systems biology — data. Aaron Kantor of SurroMed, for example, described the company’s SurroScan technology — a laser-scanning cytometry platform for clinical characterization of blood and tissue samples. The company has performed over 90,000 clinical assays on its 16 instruments, Kantor said, and integrates the data with mass spec information on proteins and metabolites to identify biomarkers.
ExonHit, meanwhile, is generating data on alternative splice isoforms using its DATAS (Differential Analysis of Transcripts with Alternative Splicing) platform. Researchers often overlook the biological significance of alternative splicing, said Richard Einstein, VP of research at ExonHit, preferring to focus on the expression levels of genes associated with a disease rather than the isoforms that may also play a role. Einstein said that ExonHit currently has a compound in Phase II clinical trials for ALS that targets a splice isoform of PDE.
Those who want to follow ExonHit’s example may not have very long to wait for more data on splice variants to become available. Affymetrix is developing an “all-exon” array, with one probe set per exon, to discover all variations of alternatively spliced genes, said Melissa Cline, a senior scientist with the company. Cline said the first prototype for the all-exon chip, which contains data derived from 266,000 transcripts and more than one million exons, is “just out of manufacturing” and should be commercially available by early 2005. In the meantime, she said, Affy plans to run “a battery of human tissues” on the array, and will release the alternative splicing expression data publicly “to get people excited about other questions they can ask” once the chip is available.
Cline said that Affy is also developing “tiling arrays” that can span entire chromosomes to generate data for the ENCODE project. Commercial versions of those arrays are “in the early stages of release,” she said.