Skip to main content
Premium Trial:

Request an Annual Quote

Stanford Scientists Using Phosphoprotein Profiling to Predict Oncogene Addiction, Drug Response


By Adam Bonislawski

A team led by Stanford University researchers has developed a method for predicting response to cancer therapies based on cellular signaling protein dynamics and changes in tumor size.

Their research, which was detailed in a paper published last week in Science Translational Medicine, represents a step toward better tools for predicting the effects of targeted cancer drugs, and suggests that these effects can be modeled using a smaller number of parameters than generally thought, Dean Felsher, a Stanford oncologist and leader of the study, told ProteoMonitor.

So-called "oncogene-addicted" cancers show dramatic tumor regression after treatment targeting specific oncogenes like epidermal growth factor receptor, or EGFR. As such, the ability to predict the emergence of this "addiction" could prove key to developing and prescribing oncogene-targeted therapies.

A wide variety of factors including angiogenesis, immune system activity, and genetic mutation are likely involved in oncogene addiction. The Stanford team's aim, Felsher said, was to find a set of relatively simple criteria onto which this large number of inputs might converge.

The researchers used a transgenic mouse model of K-ras G12D-induced lung adenocarcinoma that is known to exhibit oncogene addiction. The model featured the tetracycline-regulatory ON system, which allowed Felsher's team to turn the K-ras G12D oncogene on and off by either adding or withdrawing doxycycline in the mice's drinking water.

Using serial weekly microcomputed tomography imaging after activation and inactivation of K-ras G12D, they measured the kinetics of tumor formation and regression in the mice in vivo. Via immunohistochemistry they also measured over time in tumor cells the phosphorylation states of five pro-survival signaling proteins – Erk1, Erk2, Akt1, Stat3, and Stat5 – and one pro-death protein, p38.

The researchers then built a model using ordinary differential equations that incorporated these survival and death signals along with the tumor formation and regression data. With that model they demonstrated that K-ras G12D oncogene addiction could be explained almost entirely by the balance between aggregate survival and death protein signaling. Applying the model to a different cancer – lymphoma driven by the oncogene MYC – they found it could be generalized to this cancer as well.

Given its simplicity, the power of the protein signaling-based model came as something of a surprise, Felsher said, noting that prior to the study, he had "assumed you'd have to incorporate six or seven other variables."

He noted that his group has described "many different mechanisms of oncogene addiction" over the last 10 or 15 years — including studies showing that the immune system, angiogenesis, and senescence are involved — "so we were actually surprised that we could use growth and survival signaling alone" to make predictions.

"It turns out that growth and survival signaling are so critical that if you have a very quantitative measurement you can make a prediction," Felsher said.

The study doesn't represent an optimized method, he added, noting that "a few years from now we'll probably publish a paper saying we've improved the method and now we can predict [oncogene addiction] even better using other techniques."

One such technique is a nanofluidic immunoassay Felsher's lab has developed for measuring low-abundance phosphoisoforms involved in cancer signaling (PM 10/15/2011). The assay uses ProteinSimple's Nanopro1000 instrument to separate tumor cell proteins based on change via isoelectric focusing and then uses antibodies to detect and quantitate them.

Thus far Felsher has used the platform to investigate protein signaling in several hundred specimens, he said, including work identifying phosphoisoforms changes linked to clinical response in chronic myelogenous leukemia patients being treated with tyrosine kinase inhibitors.

"We're excited about that [platform], and we'd like to integrate both ideas – measuring proteins using highly sensitive technologies with the modeling that we published in this paper," he said.

While in the recent study the researchers were "able to make a pretty good model simply by measuring half a dozen proteins," Felsher said that it's unclear what the optimal amount of data for modeling oncogene addiction will be.

"I don't think we know exactly how much we need to know," he said. "Whether adding another 50 proteins will make it a more predictive model is hard to say. What seems to happen when you add in all that complexity is you start getting into a circumstance where it's easy to fit anything to anything – the overfitting issue."

Saying that he would "certainly be very eager to interact with pharmaceutical companies" to apply the modeling approach to drug development, Felsher noted that he has presented the work to pharma firms as part of basic research talks, but that the researchers are not currently collaborating with any industry players.

Moving forward, he said, they hope to do a small clinical study to validate the approach. "The experiment we did wasn't a clinical study, so we can't say for sure that this will help people, but we think we have enough preliminary data that we should test the model [clinically]."

Felsher added that an advantage of the model is that is can be put into clinical study with currently available imaging techniques. In collaboration with Stanford researcher David Paik, a senior co-author on the Science Translational Medicine paper, his team is also looking to incorporate new imaging methods in its work. In particular, they are interested in using position emission tomography for their measurements, given that such an approach could potentially provide data on both tumor size and protein signaling.

"Because PET is already available in the clinic, it makes sense that if we incorporate a PET-based approach [the model] could be even more useful," he said. "So we're looking at, for example, new ways of making PET-based measurements of in vivo signaling."

Thus far the modeling work has focused primarily on lung cancer and lymphoma, Felsher said. He noted that the fact that these are two very different kinds of cancers is encouraging for the method's broader applicability.

Have topics you'd like to see covered in ProteoMonitor? Contact the editor at abonislawski [at] genomeweb [.] com.