PodcastSystems Biology

Science Signaling Podcast: 30 June 2009

Science Signaling  30 Jun 2009:
Vol. 2, Issue 77, pp. pc12
DOI: 10.1126/scisignal.277pc12


This is a conversation with Ulrik Nielsen, author of a Research Article published in the 30 June 2009 issue of Science Signaling. He discusses the use of mathematical models to predict candidate drug targets in signaling pathways involved in cancer.

(Length: 14 min; file size: 6.5 MB; file format: mp3; location: http://podcasts.aaas.org/science_signaling/ScienceSignaling_090630.mp3)

Technical Details

Length: 14 min

File size: 6.5 MB

File Format: mp3

RSS Feed: http://stke.sciencemag.org/rss/podcast.xml

Download Podcast: http://podcasts.aaas.org/science_signaling/ScienceSignaling_090630.mp3

Educational Details

Learning Resource Type: Audio

Context: High school upper division 11-12, undergraduate lower division 13-14, undergraduate upper division 15-16, graduate, professional, general public and informal education

Intended Users: Teacher, learner

Intended Educational Use: Learn, teach

Discipline: Cell Biology, Molecular Biology

Keywords: Science Signaling, cancer, drug target, EGFR, ErbB, mathematical modeling, systems biology


Host – Annalisa VanHookWelcome to the Science Signaling Podcast for June 30th, 2009. I’m Annalisa VanHook. In this episode we’ll discuss recent research using mathematical modeling of signaling networks to predict viable targets for cancer therapies. I’ll be speaking with Ulrik Nielsen, corresponding author of a Research Article published in the June 30th issue of Science Signaling (1) and Chief Scientific Officer of Merrimack Pharmaceuticals. Nielsen spoke to me on the phone from his office in Cambridge, Massachusetts.

Interviewee – Ulrik NielsenSo, we’re particularly excited about this study because it’s our first example of going from really doing computational modeling of a network that we’re interested in in biology all the way to a therapy that we now put in the clinic, which is the vision that we set forth for Merrimack when we spun out of MIT and Harvard about 8 or 9 years ago.

Interviewer – Annalisa VanHookThis paper that you’ve just published in Science Signaling describes your group's efforts to use a systems biology approach to identify potential targets for developing drugs or other therapies to treat human diseases. Can you tell us a little bit about this systems biology approach?

Interviewee – Ulrik NielsenWe started doing this quite a while ago, trying to understand how the different components within networks interact with each other as a system, and the reason we were doing that was, we were sort of, the past decade or so has been the decade of genomics, where we identified all the parts of these systems, and, particularly in drug development, we started very much focusing on the individual target, making inhibitors of, say, a specific kinase, without truly understanding the system that they’re a part of. And I think what we’ve seen arising from that is a continuation of this trial-and-error approach, that we just didn’t understand these individual targets well enough to just straightaway go ahead and target them. So what we've been trying to do at Merrimack is to, through an integrative approach of experimentation and modeling, to try and understand how all these components work together in the network and use modeling and simulation as a way to identify the best way of perturbing a given network with, say, a drug. So, we are really trying to use this systems biology approach to change the trial-and-error paradigm that has been the sort of mode in drug discovery for a long time. The ability to simulate the activity of a drug before you make it would be incredibly powerful in terms of devising new therapies.

Interviewer – Annalisa VanHookSo you can use this systems biology approach to focus in on particular parts of a pathway that might be the best target for developing therapies.

Interviewee – Ulrik NielsenRight. Not just identify the most important target, but also think very carefully about what's the best mechanism of a drug targeting that target. And today there are many different ways to target, even if you’ve already identified that you want to hit a specific kinase, there are still many different ways of doing it. And we can use simulation to sort through that systematically. To us it’s not that much different than what has been done in the engineering field for, for decades, using modeling and simulation to figure out the best solution to a problem before you, say, build an airplane or whatever it is, that you would carefully simulate the different solutions to the problem. That’s really what we have been trying to do here for some time now.

Interviewer – Annalisa VanHookThe first step of the systems biology approach is to build models. So, how do you build these signaling pathway models?

Interviewee – Ulrik NielsenThe models that we are focused on here are typically mechanistic models that represent all the molecular interactions that are in the network that we’re interested in. The way we do that is we very carefully read the literature looking for what protein interacts with which other proteins, if there’s any kinetic data out there, what's the strength of that interaction, the rate constant, et cetera. And then the other component that we need to build these models is, what are the concentrations of the different proteins that interact with each other in these pathways. What we often find is that a lot of the data on how these proteins interact in sort of a quantitative way is not out there, and we find ourselves generating a lot of data to support that. So it’s very much an iterative approach between carefully reading [of] the literature to populate our models and then extensive experimentation to generate data sets that support these models. And what we are particularly looking for when we generate biological data for these models is first, typically time courses of activation, and we show some of that in this paper, where you will stimulate a, say, a cancer cell line in culture for a period of time, and then at different time points take samples, lyse the cells, and analyze the state of phosphorylation of different nodes in the pathway. We can then use that data to, to fit the models to and through that, back out some of the kinetic rate constants that weren’t known in the literature. The other parameter that we’re looking to determine from our experiments is, what are the protein concentrations in the cell, of the different proteins in the network? And we can do that using quantitative biological methods such as ELISA, Western blots, et cetera to identify those. At the end of the day, we get a model that we can then go through a number of validation steps to see that it’s actually behaving the way that we see in real cells.

Interviewer – Annalisa VanHookWhat signaling pathway did you model?

Interviewee – Ulrik NielsenIn this paper we modeled a pathway known as the ErbB pathway. This is a pathway that encompasses EGF receptor, also known as Erb1, ErbB2, ErbB3, and ErbB4, and it signals very strongly downstream through a number of different pathways, but in particular MAP kinase pathway and the PI3 kinase-Akt sort of pro-survival pathway. The focus of this paper was looking at the interactions of many different ligands—there are about different 12 ligands that bind this pathway—their combinatorial interactions with the receptors, and the downstream signaling, particularly through the PI3 kinase pathway that is so important in mediating survival in cancer cells.

Interviewer – Annalisa VanHookSo this pathway, then, is involved in tumors. Is it involved in any specific type of cancer, or is it involved in multiple types?

Interviewee – Ulrik NielsenIt’s broadly involved in any tumor type that we’ve looked at. Particularly in solid tumors, we find this type of activation is very important, that occurs through what we call in the paper a combinatorial ligand-induced activation, when these receptors interact with one or multiple ligands to activate the pathway. And in this paper we show that that probably could be important in as many as 50% of all solid tumors in patients.

Interviewer – Annalisa VanHookSo, you built a computational model of ErbB signaling using everything that you knew about the kinetics of the pathway, then you used this computational model to make predictions about the pathway. What did that model teach you that you couldn’t learn from experiments in cells?

Interviewee – Ulrik NielsenOur focus was to understand what’s the most potent mechanism for inhibiting this ligand-induced activation of the PI3 kinase-Akt pathway. And that is something that’s been studied for a really long time experimentally. If you look in the literature, there are thousands of papers describing ErbB2 and EGF receptors, there are hundreds of papers describing ErbB3. But what we did not have a strong sense of, what is the relative importance of these different components of this network, the different receptors and some of the downstream mediators of the signaling. Can we use computational modeling to sort of rationally think about what are the key node within that pathway and how to best inhibit it.

This pathway was particularly interesting because there was a lot known about it, and initially it was almost an academic interest. We started building a model of this pathway because there was all this knowledge, there were lots of drugs that were even in the clinic or even approved in the market inhibiting this pathway. It was an interesting place to start running simulations to see how all of these things stacked up against each other. What we quickly started to understand was that once we had this type of view of the pathway from a sort of systems biology point of view, the kind of view an engineer would have if they were working on the system, we start to understand that the drugs that were already on the market or in development were not effectively inhibiting this type of ligand-induced signaling. And that actually motivated a whole drug development program at Merrimack.

Interviewer – Annalisa VanHookSo the model predicted good places to design drugs that were not the places in the pathway where current drugs were acting.

Interviewee – Ulrik NielsenExactly. There have been drugs that target all of the receptors in this pathway except ErbB3, which is the one that we found was the key node for this ligand-induced type activation of the pathway. And there are, there were good reasons why people hadn’t targeted it, I think. In one sense, ErbB3, which is known as a receptor to tyrosine kinase but is not kinase-active, the kinase domain of this protein is not active. So that means the chemists have had a hard time going after it. There hasn’t been a lot of interest form the chemistry side of things to make small molecules that inhibited it.

Interviewer – Annalisa VanHookSince it doesn’t have any catalytic activity…

Interviewee – Ulrik Nielsen…there’s no catalytic activity to inhibit.

Interviewer – Annalisa VanHookSo then were people going for trying to make inhibitors to the ligand binding domain?

Interviewee – Ulrik NielsenSo the other reason that I think there has been a lack of interest in this protein is that it’s not highly overexpressed in cancer. In fact, we find sometimes in cancer cells there’s as little as 5 or 10,000 copies of the receptor. This contrasts to the more traditional therapeutic targets in this pathway, EGF receptor and ErbB2, for which there can be up to 1 or 2 million copies per cell. So it hadn’t been, if you do a sort of genomic analysis of cancer, it hasn’t stood out as being this highly amplified, overexpressed potential target, and as such also escaped some of the attention.

But this turns out to be, when you start stimulating different therapeutic approaches to this pathway, these particular features of ErbB3 not being overexpressed, not necessarily having catalytic activity, also makes it a very interesting target when you look at it computationally. So it turns out that because ErbB3 has many, many phosphorylation sites that get activated by the other kinases in this family, it becomes the best substrate for these kinases and the most strong signaling component for activating downstream PI3 kinase signaling, which is the signaling that leads to survival in the cancer cell in the tumor. It also means, because there’s not a lot of it, from a sort of an engineer’s point of view, that it’s actually easier to inhibit it than when there’s a lot of the receptors, that just stoichiometrically it’s easier to inhibit it way down than it is to inhibit something that’s highly overexpressed. In some was these were the very reasons that I think people hadn’t been working on this as a therapeutic target, but were the same reasons we ended up developing an inhibitor to it.

Interviewer – Annalisa VanHookSo, your model had predicted that ErbB3 might be a good target for inhibiting the growth of cancer cells. Did you then take the predictions from the model and take it into cells to test these predictions, to see whether they held true in vivo as well?

Interviewee – Ulrik NielsenYes. So, we were able to, and we describe that in the paper – we were able to make a human antibody that antagonizes the interactions with the different ligands that can activate ErbB3 downstream. And using this human antibody, we were able to go into cell lines and predict the actual simulated activity, and we show that in this paper, that very much matches up with the simulated inhibition of an inhibitor with the characteristics of our inhibitor and the actual data that we obtained in cell culture. And we were able to take that a step further into animal models where we would grow tumor xenografts using human tumor cell lines that we would implant under the skin of a mouse, and then dose this MM-121, as we called this as we called this anti-ErbB3 monoclonal antibody, and show great efficacy in a range of different models, and what we show in this paper is in lung and ovarian cancer models that we can stop the growth of these tumors by administering MM-121. And these are not tumors that are traditionally sensitive to the drugs that target the type of oncogene activation that you sometimes see in this pathway either through mutation of EGF receptor or through amplification of ErbB2 or HER2. So it’s quite exciting that we're really going after a different type of activation of this pathway in that we can take it all the way from model to cell lines to showing it in vivo in a xenograft.

Interviewer – Annalisa VanHookSo, you’ve completed testing of this drug in mice and it does appear to have activity against cancer cells. I guess then the next step is to test in humans?

Interviewee – Ulrik NielsenWe already are. So, last spring we did safety studies in animals and showed this is quite a safe molecule to administer into animals. And last summer we started a human clinical trial in cancer patients that is wrapping up this summer and fall. We're hoping to start additional Phase II studies in various cancers this fall. So, it’s getting quite exciting.

Interviewer – Annalisa VanHookWell, thank you for speaking with me, Dr. Nielsen.

Interviewee – Ulrik NielsenThank you very much.

Host – Annalisa VanHookThat was Ulrik Nielsen talking about research from his group published in the June 30th issue of Science Signaling in a paper titled “Therapeutically Targeting ErbB3: A Key Node in Ligand-Induced Activation of the ErbB Receptor–PI3K Axis” (1).

As with many complex systems, the activity of a signaling pathway is rarely more than just the sum of its parts. Properties not apparent from studying the component reactions individually may emerge when the network is considered as a whole. Mathematical modeling is a way for scientists to integrate the vast amount of information that's known about the individual steps in a pathway to get a picture of how the network behaves as a whole.

Another example of researchers using modeling to learn more about signaling pathway dynamics is an article by Macia and colleagues in the Science Signaling archives (2). In this paper, the authors used a model of a yeast MAP kinase pathway to reveal properties of the pathway not readily apparent from the in vivo experiments alone. Also in our archives, a Review by Hlavacek and colleagues discusses the methods and utility of mathematical modeling of complex biological systems (3).


That wraps up this Science Signaling Podcast. If you have any questions or suggestions, please write to us at sciencesignalingeditors{at}aaas.org. This show is a production of Science Signaling and of AAAS—Advancing Science, Serving Society. I'm Annalisa VanHook. On behalf of Science Signaling and its publisher, the American Association for the Advancement of Science, thanks for listening.


View Abstract

Navigate This Article