Skip to main content
Premium Trial:

Request an Annual Quote

Q&A: Team Uses Dynamic Proteomics to Study How Multi-Drug Combinations Affect Proteins


By Tony Fong


Name: Naama Geva-Zatorsky
Position: PhD student, biology, Weizmann Institute of Science, thesis subject, "System-level analysis of response-networks to drugs in human individual living cells," 2005 to present
Background: MSc, biology, 2002 to 2005

In the current issue of Cell, researchers from the Weizmann Institute of Science in Rehovot, Israel, describe a dynamic proteomics approach to predict how proteins react to multiple-combination drug therapies.

The approach is based on work they first started eight years ago to measure the levels and locations of endogenously tagged proteins in individual living cells at high temporal resolution. The first study using this approach was published in December 2008 in Science.

While that study explored protein interactions with one drug, the current study investigates protein interactions with up to four-drug combinations.

In the Cell article, the authors said that although drug effects on outcomes are well-studied, "the effect of drug combinations on protein dynamics in the cell is much less investigated. Understanding the impact of drugs on each protein is important in the context of a vision of a future medicine that controls protein dynamics precisely, using specific combinations of a large number of drugs."

Using 13 drugs, they studied how different combinations would affect 15 different proteins. "We find that the dynamics of each protein in the presence of a drug combination is described accurately by a linear superposition (weighted sum) of the dynamics in the presence of each drug alone," the researchers wrote. "The weights in this superposition are constant over time and depend on the drug dose."

They further add that their approach may represent a way "to bypass the combinatorial explosion problem of research on drug combinations." Rather than having to perform individual experiments for each drug dose in order to understand its effect on proteins, "a potential way to bypass this challenge lies in the finding that three- and four-drug dynamics can be predicted on the basis of two-drug dynamics," the authors said. "If this may be extended to other proteins, drugs, and cells, the present approach can reduce the combinatorial explosion problem inherent in understanding multi-drug combinations."

This week, ProteoMonitor spoke with Naama Geva-Zatorsky, a PhD student at the Weizmann Institute and the first author of the study, about the work she and her colleagues did. Below is an edited version of the conversation.

Can you describe this dynamic proteomics approach you used and how you adapted it for your study?

We created a unique system in which we can monitor the proteins inside living cells without any perturbation to the cell. It's like putting on glasses that enable us to look at the proteins inside the cells without any perturbation. And we did it by adding a fluorescent cell to the protein, but it's a special way to add the fluorescent tag.

We incorporated the fluorescent tag inside the coding gene, inside the chromosome, so that the protein is under its normal regulation, [and is ] expressed under its normal condition but with another marker that enables us to see it with our 'glasses.'

We created a library of 1,300 different clones amidst several different proteins [which were] fluorescently tagged and we can automatically monitor these proteins under a microscope with growth conditions. As the cells live, they don't feel that someone is looking at them, and we are monitoring their behavior, their movement, their decisions, their death events, and the proteins inside them.

In this project we applied several drugs and their combinations and monitored the protein responses to these drugs and their combinations.

Was the dynamic proteomics approach developed to study the components within the cell or to study different interactions?

Originally, it was created in our lab something like eight years ago. … Eight years ago, we can say we had the concept, but four years ago we started with actually starting the library that we worked with, and the idea was to look at the fundamental processes inside the cell without perturbing the cell.

[ pagebreak ]

The first big project was published [in 2008] in Science [and described] the response of 1,000 different proteins to [the cancer drug camptothecin] to see how the proteome responds to the drug. In that project we found several very interesting proteins that behave differently in cells that survived the drug action.

Let me take one step back: When applying an anti-cancer drug to cells, some cells survive the drug action and some cells die. And in that project we found several proteins that characterize the cell that died and the cells that survived the drug. These are interesting proteins for potential new drug targets.

And then we continued with the second project …to look at the drug combinations and their effects on proteins. We didn't know what would be the answer when we started. We just thought that this would be a good approach to look at the response of the proteins to drugs.

Another point is that usually when applying drugs and drug combinations to cells, people look at the outcome, [whether] the cell survives or dies in response to the drug, but in these cases, usually the effect of the drug combinations are either additive, where the effect of the combination is the sum of the effect of each of the drugs alone; or synergistic, where, when you add the two drugs, the effect is much higher than the sum of the two individual drugs; or antagonistic, where the effect is much lower than the sum of the two individual drugs.

This is when you look at the phenotypic level.

We looked at the proteins. The state of the cell — for example, healthy cell, sick cell, differentiated or undifferentiated cell — can be determined and described by its protein composition, the protein levels, activities or localizations. [That means] that one can describe a cell in a certain state by locating a point, the cell, on a graph with thousands of dimensions — say 5000D or at least 1000D — where the axes are protein-level activities or localizations.

By studying protein dynamic responses to drugs, we sought to control the state of the cell by using drugs to tune the proteins inside the cell.

This is one of the [reasons we] looked at the protein. The motivation was to control the state of the cell. This will be the next project now [that] we found that the response of the proteins to drug combinations is extremely simple.

We encountered a very surprising simplicity where we found that the proteins respond to combinations of drugs by weighted sum of the response to individual drugs. It simplified very much the response of the cells to drugs.

It means that you can measure each of the individual drugs alone and only several time points of the two-drug combination … is enough to predict the response of the proteins to two-drug combinations and to three- and four-drug combinations.

And perhaps this can be generalized to much [more] complex combinations.

The cell is such a complex [structure]. It's still an enigma … [there are] a lot of proteins, a lot of compartments, [and] protein-protein interactions. It's like a computer that computes the input that it's getting and then performs the output which are the protein responses, and then the outcome of the cell.

The simplicity [of our finding] is something that surprised us — that such a complex system computes its response to drugs in such a simple way.

That you can boil it down to a mathematical equation?

Right. We also could correlate these weights and the outcomes of the cells.

What did you mean by the 'linear superposition of the dynamics' described in your paper?

It's the weighted sum. The response of the protein to two drugs is based on the response of the same protein to the individual drug, each multiplied by a very simple weight, which is constant over time. You just need to find these weights by a very few experiments, only the individual drugs and the two-drug combinations, and with these weights you can predict the response of the proteins to two-, three-, four-drug combinations and, [presumably,] it can be generalized to more drug combinations.

Normally, without this principle, in order to study the response of cells or proteins to drug combinations, the number of experiments that you need to conduct grows exponentially with the number of drugs in the cocktail.

[ pagebreak ]

Is there a limit to this weighted sum approach to determining how proteins respond to multi-drug combinations? At some point would the protein response not be predictable because there are too many drug combinations?

We cannot expect it to be limited, and by our results, we tried 13 different drugs and 12 [of these] showed the same principle in two-drug, three-drug, and four-drug combinations.

The four-drug combination we didn't use on all the 13 drugs, but in all the experiments where we tried three drugs and four drugs, we received the same principle, so mathematically we do not expect a limit. But it's still a basic finding that still needs to be established in order to generalize it to more drugs in a cocktail.

In the study, the one drug for which the weighted sum approach didn't work was the P13K inhibitor Wortmannin. Since you submitted your manuscript have you been able to determine why?

No, we haven't, to our disappointment. We would love to receive comments from other scientists if they have any idea. It's a different drug since it targets a specific protein and signaling cascade, but we tried other drugs with this principle and they did show linear superposition.

You investigated 15 proteins for your study, but it didn't seem like you looked at membrane proteins.

No … we have metabolic proteins and DNA proteins, ribosomal proteins and some enzymes.

It's my understanding that membrane proteins play a significant role in drug interactions and efficacy. Are there any plans to include membrane proteins in further work you may be doing in this area?

Not specifically now, but in the Science paper, we analyzed 800 different proteins and there were some membrane proteins. We had some nice stories that came up in some groups of proteins but not any in membrane proteins.

Our next study is to try to control the state of the cell to predict theoretically how to move the cell from one state to the other and then to prove with experiments that we can control the movement of the cell state based on these findings.

We might also [include] membrane proteins. We didn't focus specifically on them [in the current paper] not for any specific reason. We had to limit the amount of proteins in this study since we couldn't use a large amount of drugs … but we took two or three proteins from each functional family of the cell.

Are there any classes of drugs that you didn't include in this study that you would especially like to investigate moving forward?

I think it would be important to enlarge the number of drugs. We took DNA-damaging agents from metabolic drugs. I think we need to expand metabolic drugs in future studies.

It appears that this paper would be able to open up new off-label uses of drugs in combination with other drugs. Is that correct?

I think the main application of these findings is the reduction of the complexities of drug cocktail experiments or studies. Hopefully, it's going to apply not only to lung cancer. I believe that it will apply to all kinds of diseases and different cancers [though] we worked on lung cancer.

It helps to navigate the space of drug combinations, to do [a minimal amount] of experiments, to find these weights, and with these weights to navigate the space of drug combinatorics, and then to do the correct experiment based on the theoretical navigation.

Does this method have any application for personalized medicine?

Of course, but the library is very unique. It can be very useful in personalized medicine, though we still need to find a way to look at the proteins in the person's cells.

If you could look at the protein responses in each person, find their responses for each of the drugs in the two-drug combination, then [you can] combine the correct cocktail of drugs to cure this specific person.

This study looks at chemical therapeutics. What about biologics? Can it be used to predict protein response to multiple combinations of biologics?

It could apply to them. For each drug you first have to do an experiment for each drug alone. In two-drug combinations, we could not predict a priori which drug will behave the same, or how it will behave with this simple principle.

We have more confidence now that families of drugs that we tested will behave similarly. Other DNA-damaging chemicals like small molecules or biological agents — we first wanted to test them and then only can we answer it.

It's a very simple principle, but we cannot generalize it to any drug before doing some basic experiments. But since proteins respond to biological agents as well with different dynamics, it could also be an avenue [for further research].

The Scan

Researchers Compare WGS, Exome Sequencing-Based Mendelian Disease Diagnosis

Investigators find a diagnostic edge for whole-genome sequencing, while highlighting the cost advantages and improving diagnostic rate of exome sequencing in EJHG.

Researchers Retrace Key Mutations in Reassorted H1N1 Swine Flu Virus With Avian-Like Features

Mutations in the acidic polymerase-coding gene boost the pathogenicity and transmissibility of Eurasian avian-like H1N1 swine influenza viruses, a PNAS paper finds.

Genome Sequences Reveal Evolutionary History of South America's Canids

An analysis in PNAS of South American canid species' genomes offers a look at their evolutionary history, as well as their relationships and adaptations.

Lung Cancer Response to Checkpoint Inhibitors Reflected in Circulating Tumor DNA

In non-small cell lung cancer patients, researchers find in JCO Precision Oncology that survival benefits after immune checkpoint blockade coincide with a dip in ctDNA levels.