Combining proteomic, genetic, and functional data to understand cancer
Researchers identified common, dysregulated pathways among different cancers using a newly developed protein-protein interaction mapping technique. The results may inform treatment and lead to the development of more effective drugs for cancer.
Nevan Krogan, a quantitative biologist at the University of California, San Francisco, didn’t intend to publish three landmark papers in Science at once — and he never plans to again. But this September, he did just that (1-3). After a decade of research, Krogan’s team reported the first massive analysis of protein-protein interactions (PPIs) in cancer cells to pinpoint the mutations that cause cells to turn carcinogenic and the common pathways that spur tumor growth.
Since the discovery of the first cancer-causing mutation in 1970, nearly 700 genes have been implicated in cancer, according to the Catalogue of Somatic Mutations in Cancer (4). However, cancer rarely results from a single genetic mutation, so therapies targeted to a specific mutation help only a small fraction of patients.
In their recently published set of papers, Krogan’s team used mass spectrometry and publicly available databases to identify hundreds of PPIs — many previously unknown — across thirteen types of cancer, with studies focused on the unique protein network dysregulation in breast cancer and head and neck squamous cell carcinoma (HNSCC). By understanding dysregulated protein networks, Krogan hopes to facilitate personalized treatments for larger numbers of patients.
Why is it important to complement genetic data with PPI data?
If we sequence a tumor from ten different patients with cancer, we might find a few common genetic mutations, but there is a lot of heterogeneity. The variation makes us wonder if it’s even the same disease. But when we compare dysregulated proteins and pathways, there’s a great deal of overlap.
If we have five patients that each have a different component of the same protein complex mutated, the result is the same: cancer. The PPI maps we develop consolidate these mutations and make them interpretable. When looking at 100 mutations, we find that they only affect five pathways. Grouping patients by pathways rather than by individual mutations will help predict what drug may work best for that patient. If there’s no existing effective treatment, then cancer researchers have a new pathway to study. Our three studies published in Science don’t just generate new data about the pathways dysregulated in cancer, but they also demonstrate how others can use these maps.
How do you develop the PPI maps?
We express proteins of interest with an attached affinity tag in various cancer cells and non-cancer cells. Then we purify the proteins and use mass spectrometry to identify any proteins that co-purified with the protein of interest. We compare protein interactions in cancer and non-cancer cells to find interactions that occur only in the diseased cells. We also compare the interactions across cells carrying different mutations in a given gene to determine the effect a mutation has on interaction. And we examine how the interactions and protein networks change in different conditions, like in the presence of different drugs.
Just like genetics, proteomics alone has limited value. We put these pieces together. We combine genetics, proteomics, structural biology, and chemical biology to understand the functional relevance of our data.
How long does it take to make a PPI map?
It took many, many years to get to this point. I developed PPI maps in graduate school in budding yeast. We did that for a long time, but we pivoted when COVID-19 came along. We thought that if we knew what COVID-19 was doing, we could better predict future viral mutations and develop better vaccines and drugs.
We mapped the interactions between human and viral proteins for COVID-19 and other coronaviruses (5). We gathered a wealth of information in a matter of months, and, at the time, I wondered how the cancer project took so many years. Out of fear during the pandemic, we quickly forced these proteomic and analytic technologies together and realized that the data available to develop PPI maps could be integrated more than we previously thought.
Forcing things to fit together can be incredibly powerful, both experimentally and computationally. That effort informed our work with cancer. We developed a disease agnostic pipeline that we can point at anything if we have a cell to look at and a list of potential disease-related genes.
When did you start your work developing PPI maps for cancer?
Trey Ideker, a biochemist from the University of California, San Diego and co-senior author on our recent set of Science papers, and I founded the Cancer Cell Map Initiative in 2014. We realized that sequencing data provided virtually unlimited information about cancer, but the functional units of a cell are proteins, not genes. Sequencing defined the sets of genes linked to cancer, but it was time for the next step. We wanted to look at the genomic data in a new way by developing these PPI maps. The goal of the initiative is to generate functional protein networks dysregulated in cancer and develop new therapeutic strategies for patients with cancer.
What was the most exciting finding in your recent set of studies?
We mainly focused on breast cancer and HNSCC in these studies. We looked at the genes most often mutated in these two cancers and subjected them to PPI mapping. In HNSCC, we looked at three different mutations in phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA), a gene commonly mutated in cancer. The proteins interacting with PIK3CA were very different across mutations, but there was a set of point mutations that was prevalent in 5% of all cancers that resulted in a tighter connection with a protein called receptor tyrosine-protein kinase erbB-3 (HER3), which functions in the same pathway as PIK3CA. There is an antibody against HER3, so we wanted to know if HER3 inhibition could effectively treat cancer cells with mutations in PIK3CA. Cancer biologists told us that this inhibition would not work. This was true in most cases, but HER3 inhibition in mice with HNSCC tumors with PIK3CA mutations that caused a stronger interaction between PIK3CA and HER3 had a beautiful response. This has huge implications across many cancers.
Are you working on maps for other diseases?
We’re developing maps for heart disease, neurodegenerative disease, pathogens, and other types of cancer. The next big set of papers we plan to publish will focus on neuropsychiatric disorders such as autism.
Our discovery platform can be applied to many different diseases. There’s overlap between seemingly unrelated disorders at the biological level. The same genes mutated in breast cancer are being hijacked by COVID-19. The genes hijacked by Zika virus are mutated in Alzheimer’s disease. These are things we don’t see when looking at the individual gene level. Looking at complexes and pathways is where we will find the big discoveries. It will bring scientists from different disease areas together because a treatment for one disease could work for another. We’re not just making connections between genes and proteins. We’re making connections between scientists. That’s the most exciting aspect of all of this.
This interview has been edited and condensed for clarity.
- Kim, M. et al. A protein interaction landscape of breast cancer. Science 374 (2021).
- Swaney, D.L. et al. A protein network map of head and neck cancer reveals PIK3CA mutant drug sensitivity. Science 374 (2021).
- Zheng, F. et al. Interpretation of cancer mutations using a multiscale map of protein systems. Science 374 (2021).
- Sondka, et al. The COSMIC Cancer Gene Census: describing genetic dysfunction across all human cancers. Nat Rev Cancer 18, 696-705 (2018).
- Gordon, D.E. et al. Comparative host-coronavirus protein interaction networks reveal pan-viral disease mechanisms. Science 370 (2020).