Why geneticists stole cancer research even though cancer is primarily a signaling disease

See allHide authors and affiliations

Science Signaling  22 Jan 2019:
Vol. 12, Issue 565, eaaw3483
DOI: 10.1126/scisignal.aaw3483


Genetic approaches to cancer research have dramatically advanced our understanding of the pathophysiology of this disease, leading to similar genetics-based approaches for precision therapy, which have been less successful. Reconfiguring and adapting the types of technologies that underlie genetic research to dissect tumor cell signaling in clinical samples may offer an alternative road forward.

The thrust of many current efforts in cancer research and personalized cancer medicine continues to overwhelmingly emphasize the importance of defining tumor cell genetics by DNA sequencing. Because cancer is a disease of abnormal signaling leading to excessive cell proliferation, why is cancer genetics considered to be the holy grail for defining patient-specific treatment?

The reason that geneticists can claim cancer research as their domain is very simple: Historically, they have been spectacularly successful in an area in which cell biology and biochemistry simply could not deliver anything close to the extraordinary insights and breakthroughs that genetics consistently provided (1). If anything, RNA interference (RNAi) and the subsequent explosion of CRISPR-based genome-editing techniques will even further hasten the pace at which genetic approaches will unravel the basic biology that underlies tumor initiation, progression, and response to treatment.

That being said, I believe that the unbridled success provided by these genetic tools has led many to conflate this approach with the basic pathophysiology of the disease. I refer here to the overwhelmingly accepted ism that “cancer is a genetic disease.” Although mutations, amplifications, and deletions in oncogenes and tumor suppressor genes are clearly direct contributors to various tumor types, most cancers are not passed down from generation to generation, and if the expression of aberrant genes is sufficient to label a disease “genetic,” then shouldn’t nearly all immune-mediated diseases, with autoreactive antibodies and T cell receptors (TCRs), equally be considered genetic diseases? I raise this point not merely to argue over semantics but rather to caution against the logical conclusions that result from thinking of cancer purely as a genetic disease. This logic underlies much of the basic foundation from which tumor genome-sequencing efforts have been championed. If we follow classical genetics principles, one would expect the frequency of specific cancer-associated mutations to directly reflect the evolutionary advantage that the mutations provide during tumor development. Therefore, “hot spot” mutations should naturally pinpoint the critical functions of the mutant molecules that were lost or gained, more so than rare and infrequently observed mutations.

There could hardly be a better example of this line of thinking than in the analysis of the tumor suppressor protein p53, the most frequently mutated gene in human tumors. It was therefore surprising when Giacomelli et al. (2), in a massively comprehensive screen that explored all possible mutations at every position in TP53, found that the five most commonly observed p53 mutations in human tumors were not the most effective at providing dominant-negative or loss-of-function properties. Instead, the researchers discovered that particularly common mutations are specifically observed because the mutagenic processes by which these mutations arise are preferentially enriched in the tissue types of the tumor (2). Thus, the final distribution of clinically detected missense mutations in p53 is a compromise between the various rates of different mutagenic events and the fitness advantage that those mutations provide.

In addition to what tumor genetics tells us, which is a lot about DNA damage and repair, cancer should primarily be thought of as a signaling disease. It is precisely the signaling behaviors involved in growth factor and nutrient responses; cell-cell and cell-substratum attachment; entry into and exit out of the cell cycle; responding to DNA mishaps during replication or exposure to mutagens; controlling the orderly, efficient, and accurate segregation of chromosomes during mitosis; and evading cell death that are aberrant in cancer and serve as the key drivers of tumorigenesis and tumor progression (3). This is why nearly all molecularly targeted therapeutics are directed against signaling molecules. Furthermore, the dysfunctional signaling in tumors arises not only from gene mutations but also from epigenetic alterations or by rewiring of signaling pathways, which probably explains why certain tumor types seem to lack dominant driver mutations.

If signaling in tumors is so important and determines both the tumor cell state and the drugs to which the tumor would likely respond, why then is signaling information not being regularly used in clinical decision-making? The obvious reason is that these types of measurements are difficult, tedious, and nonstandardized and need to be fashioned individually for each pathway that is being queried—not exactly conducive to the design of personalized cancer treatments. Is there any hope for some type of novel approach to lead us out of this backwater of low-throughput, nonmultiplexed subjective measurements in clinical samples and into the promised land? Here, I would argue, is where genetic technologies may again be our savior.

The difficulty with protein-based measurements is due to the difficulty with proteins themselves. Each protein has a unique sequence and structure and therefore requires different buffers, pH, salt conditions, detergents, etc. In contrast, the most remarkable thing about Watson and Crick’s structure of B-form DNA is that the structure is exactly the same no matter what the sequence is! Consequently, for nucleic acids, only a single set of reagents is needed to query any and all DNA sequences. So, if we want measurements of signaling to rival genomic sequencing for throughput and multiplexing, we need to convert the protein signals into something like nucleic acids, with all of their advantages for parallel amplification and detection. This could be accomplished by various techniques, depending on exactly what one wants to measure (Fig. 1). Importantly, not all of these techniques necessarily rely on nucleic acid chemistry.

Fig. 1 Example technologies that could enable signaling measurements to be multiplexed, quantified, and converted into medium- or high-throughput platforms for human precision medicine.

Nucleic acid aptamers enable direct quantification of their target abundance by simply counting the reads of each aptamer bound to immobilized cell lysates or tissue sections (5). It is less clear whether they can be broadly used to detect PTMs. Both cytometry by time of flight (CyTOF) and phospho-flow technologies enable single-cell analysis of multiple signaling circuits, ranging from a theoretical maximum of ~15 for fluorescently labeled antibodies to ~80 for heavy metal–tagged antibodies with mass spectrometry–based detection (6). Most success with large antibody multiplexing with CyTOF has been with cell surface markers; however, intracellular staining is increasingly being pursued. Examination of histological sections of tumors through either image-guided mass spectrometry (7) or various antibody-based multiplex epitope detection methods combined with image segmentation (8) can be used to determine the activation states of multiple pathways, both in the tumor and in the surrounding microenvironment. The ability to tag antibodies with oligonucleotide barcodes offers almost unlimited potential to multiplex signaling pathway activity readouts in a quantitative manner using high-throughput sequencing, including their application to reverse-phase protein arrays (RPPAs) (9), Western blots (WB), and immunohistochemistry (IHC). With the proper controls, these reagents also enable measurements of protein-protein interactions using proximity ligation assays (PLAs) and, when applied to single-cell sequencing analysis (SCSA), should eventually facilitate the simultaneous determination of signaling state, RNA expression, and genomic sequencing at the single-cell level (10).


Another important difference that complicates signaling measurements is that, although we have a fully annotated genome, we do not yet have a complete catalog of posttranslational modifications (PTMs) on all proteins that can serve as indicators of their activity, and for many of the known modifications, we lack knowledge of the enzymes responsible for them, as highlighted in the review by Needham et al. (4) on the dark phosphoproteome in this issue of Science Signaling. Nonetheless, combining the limited knowledge of signaling pathways that we currently have with a small future investment in some straightforward, nucleic acid–based imaging or flow technologies could enable us to transform signaling measurements in clinical samples into the types of rapid, quantitative, multiparallel expansive datasets that are necessary for precision cancer medicine but that currently lie only within the purview of genomics and gene expression studies.


Acknowledgments: I thank T. S. Netterfield, A. K. Shalek, P. C. Blainey, J. A. Lederer, and V. Gocheva for many helpful discussions. Funding: Work in my laboratory was funded by NIH grant ES015339, the Ovarian Cancer Research Fund Alliance, and the MIT Center for Precision Cancer Medicine.

Stay Connected to Science Signaling

Navigate This Article