Twenty-five years ago all what was known about oncogenic signal transduction could be summarized in a single, simple, slide: nine oncogenes linked by a feeble functional and spatial topology that led from the cell surface to the nucleus.
This slide showing oncogenic signal transduction pathways in the late eighties was proudly used by one of the authors, Walter Kolch, when he was a postdoctoral fellow.
Today of course, we smile at such simplicity. We know signalling is organized in complicated networks, and modern omics methods have given us the tools to expand these networks beyond our comprehension. But when it comes to oncogenic mutations, we are still very much wedded to the simple view that a mutation changes the activity of a single protein and that’s that. Our approach to cancer drug discovery and development reflects this philosophy. We find the driver mutation and try to make a drug that targets the culprit. Some spectacular early successes justified this strategy, e.g. imatinib for inhibiting the BCR-ABL oncogene that drives chronic myelogenous leukaemia. However, more often than not this strategy has failed us. For instance, RAS is mutated and activated in ~30% of all human cancers1. To block oncogenic RAS signalling we have developed powerful inhibitors targeting its effectors, yet with overall disappointing results - likely because of intricate feedback mechanisms in the downstream pathways and the large number of effector pathways2.
The main known RAS effector pathways. Adapted from 3.
Although our knowledge of RAS effector pathways is impressive, the drug failures pose a pertinent question. Do we really understand RAS signalling, especially the effects of RAS mutations? This is the query underlying our study. To answer this question, we mapped the protein-protein interaction network (PPIN) downstream of the epidermal growth factor receptor (EGFR), which is the natural habitat where oncogenic RAS roams. Using quantitative mass spectrometry, we mapped two PPINs (together >6,000 interactions) in a KRAS mutated colorectal cancer cell line and a derivative line, where the dosage of mutant KRAS signalling was reduced threefold resulting in a non-oncogenic phenotype. The two PPINs had overall similar architectures sharing >70% of the ca. 3,000 nodes in each PPIN. However, elevated KRAS signalling was associated with a rewiring of ~25% of all edges affecting both the composition of known protein complexes as well as the connectivity between nodes in the networks.
Signal flow from the EGFR to nuclear transcription factors through the PPIN network in parental HCT116 cells (left) and derivative HKE3 cells (right), where the dosage of mutant KRAS signalling is reduced threefold4.
The mechanisms behind this rewiring appear to be multifactorial including changes in protein abundance, phosphorylation, and other as yet unknown mechanisms. Interestingly, genetic and expression alterations in the most strongly rewired nodes were significantly associated with the prognosis of colorectal cancer patients in the TCGA. This observation suggests that the proteins, which are most changed in their interacting partners in KRAS mutant CRC cell lines, are the proteins that strongly affect cancer progression in CRC patients. Using computational and mathematical modelling we also showed that PPIN rewiring profoundly changed the flow of information from the EGFR to nuclear transcription factors. This analysis allowed a global approach to validate the inferred PPINs and their differences by comparing predicted transcription factor regulation with results from genome-wide RNA sequencing.
What do these results mean for drug discovery and development? They reveal a surprising plasticity of PPINs in response to what can be perceived as a singular perturbation potentially explaining how they can outwit drug induced perturbations due to an enormous adaptive capacity. In a parallel study, we also showed that KRAS mutations similarly can change transcriptional and metabolic networks 5 indicating that adaptive processes affect diverse layers of cellular regulation. They also clearly demonstrate that drug targets need to be considered in a network context rather than as single isolated entities. This is difficult, tedious and slow when relying on experimentation alone, but can be substantially accelerated and streamlined with the help of computational modelling methods that can simulate network contexts and predict likely outcomes. Thus, a main message of our study is the call for including computational network modelling into screening strategies for identifying new drug targets.
Written by Walter Kolch [1,2] and David J. Lynn [3,4]
 Systems Biology Ireland, School of Medicine, University College Dublin, Dublin 4, Ireland
 Conway Institute, University College Dublin, Dublin 4, Ireland
 EMBL Australia Group, South Australian Health and Medical Research Institute, Adelaide, Australia
 College of Medicine and Public Health, Flinders University, Bedford Park, SA 5042, Australia
1 Hobbs, G. A., Der, C. J. & Rossman, K. L. RAS isoforms and mutations in cancer at a glance. J Cell Sci 129, 1287-1292 (2016).
2 Papke, B. & Der, C. J. Drugging RAS: Know the enemy. Science (New York, N.Y.) 355, 1158-1163 (2017).
3 Santra, T. et al. An Integrated Global Analysis of Compartmentalized HRAS Signaling. Cell reports 26, 3100-3115.e3107 (2019).
4 Fasterius, E. et al. A novel RNA sequencing data analysis method for cell line authentication. PloS one 12, e0171435 (2017).
5 Charitou, T. et al. Transcriptional and metabolic rewiring of colorectal cancer cells expressing the oncogenic KRAS(G13D) mutation. British journal of cancer 121, 37-50 (2019).