October 15, 2008 (Vol. 28, No. 18)

Predictive Modeling and Simulation Approach Could Result in More Effective Drugs

Systems biology is a rapidly evolving discipline. Even the word itself means different things to different people. In its most basic interpretation, systems biology describes approaches to characterize the effects of drugs, for example, on the activity of large, complex biological networks. At the recent CSB2 “Systems Biology of Human Disease” conference, scientists discussed different ways to model pathways, interpret data, and derive therapeutic targets from biological networks.

It would be a dream come true to organize and structure the explosion of biological data generated from high-throughput screening into a coherent cellular biological portrait. But to do that, researchers must have an accurate blueprint of the physical interactions that underlie all cellular responses. Massive data sets from genetics and all the omics can seem like mounds of bricks, mortar, and steel in which scientists must deduce what the building looks like, room by room. That is the task of biological reverse engineering.

“The complexities of biological systems are orchestrated by vast networks of interacting molecules that include DNA, RNA, proteins, and small molecules,” noted Gustavo Stolovitzky, Ph.D., manager of functional genomics and systems biology at the IBM Computational Biology Center.

“Currently, all we have is a partial understanding of how these systems interact. What we need is an accurate map of such molecular interactions. That’s the goal of the Dialogue for Reverse Engineering Assessments and Methods (DREAM) project established by the IBM Computational Biology Center and the MAGNet National Center for Biomedical Computing at Columbia University.”

According to Dr. Stolovitzky, the joint project hosts DREAM conferences that provide not only a forum for researchers to discuss the field but also a reverse-engineering challenge to those who wish to test their algorithms in order to discover if they can identify a gene-gene interaction, given only high-throughput data sets painstakingly assembled by the organizers with the help of contributing scientific partners.

“The reverse-engineering challenges are designed to verify algorithms. In other words, how can I massage the data given using statistical analyses and existing literature in an intelligent way to evaluate the data at hand? For example, from the 11 teams that participated in one of the DREAM2 (second conference) challenges, eight of them did no better than chance. However, those who achieved more did so with the simple premise that less is better. As you narrow the biological area in a more meaningful way, you get rid of data that is noise, and this provides the best route to understanding the pathways.”

Can we expect human-directed computing to be replaced by artificial intelligence? Dr. Stolovitzky thinks not, at least not yet. “I feel that the first and most important step in seeking to understand data is human insight. Artificial intelligence is a long way off from developing that. There is so much biology that still needs to be discovered that we cannot yet embed all knowledge in our algorithms. But, the trend is definitely there, and we are learning significant lessons from all of our challenges.”

The “DREAM3” conference, to be held later this month at the Broad Institute, will be a joint venture with the “Annual RECOMB Satellite on Regulatory Genomics” and the “Annual RECOMB Satellite on Systems Biology” conferences.

Docking Affinities and Signaling

Signaling pathways for receptor tyrosine kinases (RTKs) are often depicted as “on-off” wiring diagrams. Not so, said Gavin MacBeath, Ph.D., associate professor in the department of chemistry and chemical biology at Harvard University.

“RTKs encompass a large and diverse family of transmembrane proteins that elicit a wide range of cellular responses. Our understanding of these is largely based on wiring diagrams. But, RTKs have intrinsic differences that are not captured in such binary diagrams.”

Dr. MacBeath addressed a fundamental question about RTKs: Are intrinsic differences based upon the receptor itself or its cellular environment? “We placed six diverse receptors within the same cellular context and found that RTKs do indeed activate many of the same pathways, yet differ quantitatively. We also measured recruitment of proteins to the activated receptors using protein microarrays.

“In this case, we analyzed docking affinities of almost every member of two families of domains that interact with phosphorylated RTKs: the Src homology 2 (SH2) and phosphotyrosine binding (PTB) domains. Interestingly, we found that linear models that rely on combinations of docking affinities could predict the levels of phosphorylation of upstream signaling proteins.”

The microarray strategy involved studying virtually every SH2 and PTB domain encoded in the human genome in order to measure the equilibrium dissociation constants of each domain for 61 peptides representing all physiological sites of tyrosine phosphorylation on six different RTKs. The study also involved measuring the phosphorylation states of many different signaling proteins in six engineered cell lines. “We did lots and lots of Western blots,” Dr. MacBeath explained. “But the effort paid off by providing a new understanding of the intrinsic differences between RTKs.

“This information is applicable to disease. Since many RTKs are oncogenes, it is important to know which pathways get turned on and the impact of mutations in the RTKs. If we can parse this complex signaling network into segments, this could provide a means for prediction that would be invaluable for drug discovery and the rational design of therapeutics.”

Deciphering Systems Dynamics

Signaling networks contain multiple feedback loops and crosstalk. Making sense of them is hard when you only think linearly, suggested Ulrik Nielsen, Ph.D., svp of research at Merrimack Pharmaceuticals. “We need to understand systems dynamics and realize that all parts of the system may produce surprising effects if not taken into account.”

Dr. Nielsen illustrated the point with the insulin-like growth factor-1 (IGF-1) receptor pathway. The IGF-1 receptor plays crucial physiologic roles in development, differentiation, aging, and in the growth of malignant cells.

“Mechanistic modeling provides a foundation for developing targeted therapeutics. Stimulation of the IGF-1 receptor results in activation of multiple pathways that generate survival and proliferation cues with signaling models such as the ERK (extracellular receptor kinases) and AKT (serine/threonine kinase) pathways.”

Using a unique approach, Dr. Nielsen developed a model using ordinary differential equations. “We are modeling the actual interactions of biomolecules. This differs from the usual thinking because we try to represent biochemistry of the pathways in our models, whereas others look mostly for patterns or correlations. This doesn’t give a good picture of the signaling dynamics of the pathway.”

Dr. Nielsen noted that “three types of information helped build Merrimack’s equations: 1) Who interacts with whom? 2) What is the rate constant of the interactions? and 3) What’s the concentration of the reactants?

“After taking these into account, we generated experimental and simulated dose-response and time-dependent behavior of ERK and AKT in the presence of inhibitors targeting different positions within the IGF pathway including inhibitors that disrupt feedback and crosstalk mechanisms.”

According to Dr. Nielsen, this type of approach has paid off big-time. “In one case, we walked into a pathway (the EGFR/ErbB pathway) that big pharma had been developing drugs in for more than a decade using traditional approaches. Our modeling found that the most important molecule which was overlooked since the paradigm, was that the target had to be overexpressed in cancer to be important.

“We built a computational model of the system, and it quickly stood out that a key node in the network, ErbB3, was overlooked. We created a fully human monoclonal antibody designed to block signaling of the ErbB3 receptor. This is now in Phase I trials. To our knowledge, this is the first systems biology product as well as the first selective ErbB3 antagonist to enter human clinical development.”

Dr. Nielsen’s take-home message was that understanding network dynamics is crucial for predicting the best mechanism of targeting the pathway, as well as the unintended or counterproductive effects of an inhibitor. Also, performing drug optimization in silico provides a way to assess effects prior to actually creating the therapeutic.

Bioengineering Toolkit

Integrating bioengineering approaches with basic immunology seems like an unusual mix. Pioneering this approach is Darrell J. Irvine, Ph.D., Howard Hughes investigator in the David H. Koch Institute for Integrative Cancer Research at MIT. Dr. Irvine is pursuing the design of materials that interface with the immune system to create model systems for understanding lymphocyte biology and for developing therapeutic immune approaches to direct immunity against cancer and infectious disease.

“The interaction of T cells and antigen-presenting cells (APCs) forms an immunological synapse and provides a stunning example of the complexity involved in T-cell activation,” said Dr. Irvine. Binding by T-cell receptors to antigenic peptides displayed on the APC surface sets in motion T-cell activation. Dr. Irvine performed studies to address the question as to whether the unique physical structure of the synapse engages T-cell responses. To do that he created multicomponent protein patterned surfaces containing immobilized T-cell receptor ligands in defined geometries surrounded by immobilized adhesion proteins.

According to Dr. Irvine, “T cells seeded on the surface responded in a manner characteristic of live APC-T cell interactions. That is, T cells halted migration, increased intracellular calcium, and clustered receptors and signaling molecules at the interface. This suggested the response was dependent on the pattern of ligands presented. Using this model system allows both molecular recognition events and cell motility/dynamics to be tracked for hundreds of single cells simultaneously.”

Dr. Irvine’s model system has a broad range of applications. The ability to pattern commercially available proteins into discrete regions while retaining activity provides a patterning strategy immediately applicable to a broad range of available protein ligands. Additionally, the assembly of T-cell synapses can be modulated simply by altering the organization of stimulating molecules. This has ramifications for the priming of an improved immune response that could be valuable for enhancing immunotherapies.


Researchers in a consortium initiated by Pfizer are utilizing a systems biology approach to target identification in type 2 diabetes. The company has established a large-scale research consortium to study signal transduction in insulin-sensitive and insulin-resistant cells.

There are two ultimate goals, according to collaborator Frank Doyle, Ph.D., professor of chemical engineering and associate director of the University of California-Santa Barbara (UCSB)-MIT-Caltech Institute for Collaborative Biotechnology. “First, to create detailed knowledge of the biological networks underlying insulin signaling and glucose uptake, and second, to use this knowledge to identify novel targets for intervention.”

Researchers at four universities (UCSB, Caltech, MIT, and the University of Massachusetts) and Entelos, a physiological modeling company, are collaborating on the project. “We are bringing in an array of modeling capabilities (e.g., deterministic and stochastic) to this problem,” Dr. Doyle reported.

The driving force behind the effort stems from the problem that most of the type 2 diabetes medicines do not meet the needs of many patients. Nearly 60% of patients do not respond to current therapeutics.

Initially, biological data will be collected using time-resolved phosphoproteomics and gene expression, combined with chemical genetics and reverse engineering of transcription-factor binding events. Mathematical models of signaling events will help determine which targets can be modulated to best restore insulin sensitivity.

The strength of the project, according to Dr. Doyle is that, “No single lab can perform all of the work required to accomplish these goals, hence a collaborative endeavor is required to bring expertise from proteomics, transcriptomics, biophysical modeling, etc. together to solve the problem.”

The burgeoning field of systems biology is still seeking to establish itself as a viable tool. The challenge will be to prove this predictive modeling and simulation approach can accurately interpret complex data and translate those insights into therapeutics.

Previous articleColorectal and Breast Tumor Genome Analysis Reveals Prolific Copy Number Changes
Next articleCharles River Opens Preclinical Site in Shanghai