A team of researchers at the Massachusetts Institute of Technology has developed a mathematical model that can predict the effect of certain stimuli on cellular phenotype. This approach, which is based on similarities in cellular signaling pathways, could eventually be used to test the effects of drugs in different cell types, according to the researchers, who published their work in the August 2 issue of Nature.
The authors note in their paper that while “the fundamental components of many signaling pathways are common to all cells,” stimulating or perturbing this network “often causes distinct phenotypes that are specific for a particular cell type.”
This phenomenon, called cell specificity, “presents a challenge in understanding how intracellular networks regulate cell behavior, and an obstacle to developing drugs that treat signaling dysfunctions,” they wrote.
The researchers applied a systems modeling approach to determine how cell-specific signaling events are integrated through effector proteins to cause cell-specific outcomes.
Using partial-least-squares regression, the team constructed mathematical models to predict how kinase signaling events are processed into phenotypes through effector substrates.
They found that accurate predictions of cell specificity are possible when different cell types share a common “effector processing” mechanism. The researchers also found that PLSR models based on common effector processing accurately predict cell-specific apoptosis, chemokine release, gene induction, and drug sensitivity across three different epithelial cell lines.
The investigators concluded that cell specificity originates from the differential activation of kinases and other upstream transducers, which act together to allow the use of common effectors by different cell types to generate diverse outcomes.
This week, Cell-Based Assay News spoke to Kathryn Miller-Jensen, the lead author on the study and currently a postdoctoral scholar at the University of California at Berkeley, about her work and possible areas for future research.
Could you give me a little background on the work that was published in this paper?
We were interested in the question of, “How do cells use the same signaling network to make different decisions?” We talk about cells having the same type of signaling network, yet we know that different types of cells respond differently to different environmental cues.
We were particularly interested in the chemical signaling that happens before the effectors or transcription factors start cells' genetic programs. We wanted to know if the information that is transmitted in those chemical signaling factors would be enough to provide the variety of responses seen in different cell types.
Can you discuss your results?
I should mention that one of the reasons this question has not really been asked before is that it is a difficult question to answer experimentally. We had previously used a modeling process called partial-least-squares regression, which can basically take the chemical signals, derive their bulk chemical properties, and translate those into a model that can predict different outcomes.
The outcome that we were looking at was cell death, or apoptosis. We took a kinase activity assay, measured the activity of five important known kinase signals, and measured them over time in one particular cell type [HT-29 colon adenocarcinoma cells].
We made a model of cell death using those signals and measured the same signals in HeLa cells. We were able to predict HeLa cell death in response to various environmental cues using the model we created with the HT-29 cells.
What that told us was that even though these cells responded differently to the same environmental cues, there was some “common effector processing,” which is what we called it in the paper, where the information and the upstream signals were sufficient, at least in this situation, to allow you to predict what the different cell death responses were going to be.
To our knowledge, this is the first time that anyone had been able to demonstrate this in a combined experimental and mathematical modeling way.
The paper notes that the model failed to correctly predict behavioral outcomes in Jurkat leukemic T-cells. Why is this the case?
As I said, this is the first time that anyone has tried to test this. We tested this model on three different cell types, and they were all epithelial cells: HT-29 colon adenocarcinoma cells, HeLa cervical carcinoma cells, and MCF-10A mammary epithelial cells.
What we wanted to do was test the model on a completely different cell type that wasn’t from the epithelial cell type lineage. We used the Jurkat T-cell line. We used a reduced model to do that testing, using only the most informative time points and signals to make that model.
We found that model failed to predict behavioral responses with the same level of certainty that it did for the cell types that were similar. We think there was some limitation in terms of how the common effector processing can work. We hypothesized that it is likely more robust within cell types of the same lineage, such as epithelial cells.
Perhaps, when you go far out to a type of cell that is very different, such as T-cells, the effector processing can be very different. In that case, you would need a lot more information to extend your model to work on those different cell types.
What do you feel would be the next step in this research?
In the paper, we started to look at responses to drugs. We used small-molecule inhibitors to knock down the chemical signaling of the different pathways that we measured to see if we could look at differences between the two cell types and their response to small molecule inhibitors.
One of the challenges in drug development is understanding why drugs that give you a certain response in one cell type give you a different response in another cell type. I think one of the more exciting areas to which to expand these models would be to try to look at the cell responses to different drug types, [which is] something that would be useful to the pharmaceutical industry.
I think another exciting area would be to see how far you could push this type of modeling outside of one cell lineage. We did a very early trial of that, but I think there is a lot more that could be investigated in terms of identifying the limits of how far such a model could go.
I think a lot of work that is clinically very important could come out of this down the road.
Do you feel that this mathematical model is something that could be commercialized?
I think it’s a tool, and this type of modeling is something that people developing drugs could eventually use to understand how different compounds could work. It has potential use in drug development applications, but I am not sure that it could be commercialized.