May 1, 2009 (Vol. 29, No. 9)
Richard A. A. Stein M.D., Ph.D.
Approach Provides New Perspectives for Biomedical Science Research Initiatives
While our ability to collect large amounts of experimental data in a timely and cost-effective manner represented, until recently, a top priority, a different challenge is currently taking shape.
Now the goal is to integrate the datasets generated by diverse approaches to allow the meaningful and accurate interpretation of complex biological phenomena. In this context, systems biology, which Hans Westerhoff and Lilia Alberghina, in their book Systems Biology Definitions and Perspectives (Springer 2007), refer to as being “new and old at the same time,” holds the key to understanding biological systems in their true complexity and dynamics.
“Over the next 10 years, a systems approach will dominate the landscape of understanding all simple and complex diseases,” said Leroy E. Hood, M.D., Ph.D., president of the Institute of Systems Biology.
In a recent study that integrated different levels of global information obtained from several inbred mouse strains with subtractive analysis, Dr. Hood and collaborators investigated the cellular perturbations during prion disease progression (see Molecular Systems Biology, published online April 7, 2009).
“The reason why global analysis is important is because it allowed us to demonstrate that we can explain virtually all of the known pathophysiology and it provided fundamental new insights into disease modules that people had no idea were associated with prion disease,” added Dr. Hood.
With the vast amounts of information that omics approaches generate, the challenges are now shifting toward finding the most robust methods for interpreting the data.
“The problem with any large-scale analysis, be it transcriptome, proteome, or genome-wide scans, is that the signal-to-noise challenge is enormous. While these omics approaches identify large lists of genes, I would argue that most of the genes on the list are noise and do not reflect pathological mechanisms. Appreciating the signal-to-noise ratio is absolutely critical,” explained Dr. Hood.
His group developed and implemented new statistical methods for dealing with noise and for integrating the datasets from different animals in new and powerful ways. Along with a thorough understanding of prion biology, this approach identified 333 genes that appear central to prion disease.
One remarkable aspect that emerged from this work is that many changes appear at the molecular level long before symptoms became apparent. Moreover, proteomics techniques identified several presymptomatic blood markers of potential diagnostic value.
These global approaches unveiled dynamic cellular networks that provide an important framework for drug discovery and design. “I think the future of drug target discovery is going to be understanding the dynamics of disease-perturbed networks,” predicted Dr. Hood. “In 10 years we will see a revolution in medicine like nothing that has ever been seen before.”
RNAi screens have recently emerged as particularly powerful approaches to survey host-pathogen interactions. Nevertheless, as genetic tools that have a propensity to reveal indirect associations, they are not very informative on whether a specific interaction is direct or indirect.
“We took an integrative approach,” noted Sumit Chanda, Ph.D., associate professor at the Burnham Institute for Medical Research. “We went in with the hypothesis that RNAi activity should not be the only criteria that one should chose when picking factors coming out from an RNAi screen.”
At a recent Cold Spring Harbor Laboratory conference on “Systems Biology: Networks,” Dr. Chanda revealed how his group integrated RNAi data with protein network analysis to obtain a spatial representation of how the factors that are identified interact with each other and with HIV-encoded proteins. The team identified 295 genes involved in early infection.
Around the same time, two other groups, led by Stephen J. Elledge from Harvard Medical School and Amy S. Espeseth from Merck & Co., used RNAi to identify host proteins involved in HIV infection. Interestingly, although each of the three screens identified approximately 300 genes, only 9 to 15 of the genes were shared between paired datasets.
“Our hypothesis is that many of these factors are going to be indirect regulators of the same pathways or biological processes, and that is why there is such a low degree of overlap,” continued Dr. Chanda.
“Most factors that are identified represent secondary or tertiary regulators of the process, such as a regulator of a regulator of a regulator. That is why this idea of integrating biochemical as well as genetic analysis, intersecting the two datasets, not only cleans up the data, but also provides a functional readout. It also offers a spatial and biochemical snapshot of how these host proteins mediate the phenotype through the interaction map.”
Microbial pathogens represent a major cause of morbidity and mortality worldwide. Their ability to develop resistance to antibiotics is thought to forecast what some investigators have called “the postantibiotic era.” As past decades demonstrated, resistant bacteria invariably start emerging after specific antibiotics become commercially available, sometimes as soon as within a few months.
“For some time we have been taking a systems biology approach to study how bacteria respond to antibiotics,” said James J. Collins, Ph.D., professor of biomedical engineering at Boston University and a Howard Hughes Medical Institute Investigator. At the Cold Spring Harbor meeting, Dr. Collins presented his group’s recent findings that all bactericidal antibiotics, regardless of their drug target, induce oxidative damage and cell death pathways that lead to the production of hydroxyl radicals and thus contribute to cell death.
Targeting bacterial protective pathways that are induced to remediate reactive oxygen species damage, and in particular manipulating the DNA damage repair pathways, becomes, therefore, one potential approach to potentiate the effect of these antibiotics.
“We believe that small molecules could be produced that would lead to the creation of super-Cipro, super-Gentamycin, or super-Ampicillin,” predicted Dr. Collins. Most recently, while examining the events following aminoglycoside interaction with ribosomes that lead to the formation of reactive oxygen species, Dr. Collins’ group revealed that these antibiotics lead to the mistranslation of membrane proteins and showed that the envelope stress response and two-component redox regulatory systems are involved in antibiotic-mediated oxidative stress and cell death. This provided additional insight into the common mechanism of killing induced by bactericidal antibiotics.
“Systems biology approaches can help provide insight into bacterial cell death pathways and the protective mechanisms induced by antibiotics,” said Dr. Collins. “These network-based analyses will lead to the development of novel, more effective antibiotics, as well as ways to enhance existing antibacterial drugs. These efforts will be critical in our ongoing fight against antibiotic resistance.”
In what represents the first effort of this kind involving a protozoan organism, Jason A. Papin, Ph.D., assistant professor of biomedical engineering at the University of Virginia, together with collaborators, reconstructed the first Leishmania major metabolic network that accounts for 560 genes, 1,112 reactions, 1,101 metabolites, and eight unique subcellular localizations.
Moreover, in collaboration with Vitor Martins do Santos, Ph.D., from the Helmholtz Center for Infection Research in Germany, his group used available genetic, biochemical, and physiological data to perform a genome-scale reconstruction and constraint-based model of the Pseudomonas aeruginosa strain PAO1, mapping 1,056 genes whose products correspond to 833 reactions and connect 879 cellular metabolites. “It is such a great system,” emphasized Dr. Papin, “since it is relatively well characterized and it is a pathogen.”
The model was validated with published genome-scale gene essentiality screens and with substrate-utilization assays that predict whether the organism can catabolize specific substrates.
Although these comparisons provide important validation, they also have significant implications for understanding what makes certain genes essential, or why an organism is capable of utilizing one substrate but not another.
“We think there is a real and unmet need. I think systems biology can have one of its earliest and largest successes in tackling infectious disease and in identifying and validating drug targets in these pathogens,” said Dr. Papin.
Systems biology promises to impact virtually every scientific field. For example, “angiogenesis is a complex multistage process that involves many molecular players with numerous cross-talks and interactions. In a system like this, the use of systems biology is absolutely necessary to understand the process. It also is an effective tool to design novel therapeutics,” noted Aleksander Popel, Ph.D., professor of biomedical engineering and director of the systems biology laboratory at Johns Hopkins University.
Angiogenesis is known to be involved in over 70 different diseases. Some, such as cancer and age-related macular degeneration, are characterized by excessive blood vessel sprouting while others (e.g., peripheral or coronary artery disease) are marked by insufficient angiogenesis.
Computational modeling, bioinformatics, and in vitro and in vivo experimental methods are the major approaches that converge in Dr. Popel’s lab to provide a better understanding of the mechanisms of angiogenesis. They also show promise for designing novel therapeutic agents. By using these approaches, Dr. Popel’s lab characterized the involvement of several key vascular endothelial growth factor (VEGF) family members in angiogenesis, described a number of models for different applications, and validated model predictions against in vitro experiments and in vivo animal models for conditions such as cancer and ischemic disease.
Investigators in the Popel lab are currently simulating administration of agents for pro- and anti-angiogenic VEGF therapies that can be used systemically or introduced by gene transfer to understand how they affect the balance of growth factors within the entire organism. The lab also has reconstructed the biochemical network that involves hypoxia-inducible factor 1α (HIF1α), a transcription factor acting upstream of VEGF and regulating over 200 genes involved in the hypoxic response. The group has developed a model to explain how in conditions such as ischemia and cancer, reactive oxygen species and antioxidants affect signaling through this pathway.
Understanding Cellular Networks
The model organism S. cerevisiae fueled some of the most important advances in biology and helped scientists understand pivotal concepts in areas spanning signal transduction and DNA repair, cancer, neurodegenerative and cardiovascular pathology, and cholesterol metabolism.
An effective way to gain insight into the regulation of biochemical processes is by exploring protein-protein interactions. Researchers estimate that the total number of interactions in S. cerevisiae ranges between 10,000 and 40,000. While many methods examine protein-protein interactions in vitro, the extent to which such interactions represent an accurate reflection of the in vivo cellular context emerges as an important question.
Stephen Michnick, Ph.D., professor of biochemistry at the University of Montreal and the Canada Research Chair in Integrative Genomics, presented work at the Cold Spring Harbor venue that he performed together with several collaborators to characterize the S. cerevisiae protein interaction network in vivo. The team took advantage of a protein fragment complementation assay. Two proteins of interest, each fused to complementary fragments of a reporter protein, are brought together and reconstitute the reporter activity if an interaction is established between them.
A genome-wide screen of protein-protein interactions identified 2,770 interactions among 1,124 endogenously expressed proteins and established a protein topology map at 8 nm resolution that promises to provide a valuable framework for future studies. A comparison with previous reports revealed that most interactions unveiled by this survey were previously unknown, pointing toward yet unexplored features of the yeast protein interactome.
“The key is to utilize these facts because we now can study interactions in living cells, and we can manipulate live cells with drugs and changes in nutrients. We can ask how the network reorganizes itself and what these changes mean? We can do new things that we have not been able to do with other approaches and learn about the dynamics of the interactome,” said Dr. Michnick.
Robust computational tools represent a driving force behind all systems biology applications. “From my perspective, which is mainly computational, I am convinced that by using sophisticated algorithms we can really push the envelope of systems biology studies much further,” said Ron Shamir, Ph.D., professor of computer science and Sackler Chair in Bioinformatics at Tel Aviv University.
“I am a great believer in the power of using sophisticated ad-hoc tools based on approaches that combine good algorithmics and a solid understanding of the biology behind the question.”
At the CSHL meeting, Dr. Shamir discussed contributions his group made toward tool development. MATISSE, an algorithm that can analyze genome-scale interaction and expression data and detect functional modules, is already an important tool being used by research groups worldwide. Extending MATISSE and its capabilities, the Shamir lab most recently introduced CEZANNE, a new methodology to extract functionally co-expressed pathways.
One of the best demonstrations of the power these approaches hold is reflected by a recent collaborative study, conducted jointly with Jeanne Loring, Ph.D.’s group at Scripps, that categorized a collection of approximately 150 stem cell lines based on their expression profiles. That study found a protein-protein subnetwork, called Plurinet, which is strongly linked to pluripotency.
In addition, other algorithms developed in the Shamir lab were used in studies of patient data, and identified disease-specific dysregulated subnetworks. The analysis, which used case-control data along with various clinical parameters such as age at onset and time to metastasis, may lead to a better understanding of the disease process.
As opposed to previous studies that used predefined pathways, these systems biology tools rely on the rationale that sets of co-expressed genes forming small connected subnetworks should correspond to pathways or processes taking place in cells. Hence, the algorithms examine the full protein interaction network and look for subnetworks that are characterized by high response and are connected.
“An advantage of our approach is that you don’t need these predefined sets,” explained Dr. Shamir, “because there is always some ambiguity as to what constitutes a pathway. Our current knowledge on pathways is incomplete and, besides, it is difficult to set the boundary between a pathway and the rest of the network.
“Pathways overlap, there is always crosstalk, and we find that it is better to algorithmically identify the subnetwork based on data, using the whole protein interaction network as a source.”
As systems biology becomes an indispensable tool to interrogate dynamic cellular events, it promises to profoundly transform society in the years to come. While each approach is powerful in its own way, it is only by integrating multiple experimental tools spanning biology, physics, chemistry, mathematics, and computer sciences, or by “welding together the sum total of all that is known into a whole,” as Erwin Schrödinger wrote more than half a century ago, that the most accurate and meaningful picture can be obtained, regardless of the scientific question being pursued.