ERA-NET: A systems approach to preventing drug resistance in colon cancer

In the COLOSYS project, a multidisciplinary team of scientists from Norway, the Netherlands, Germany, France and Spain have worked together to develop computational approaches that can assist oncologists in the design of drug therapies for colon cancer patients. At NTNU we have focused on the use of computer models of the regulatory network that forces cancer cells to grow uninhibited and perform computer simulations to find pairs of cancer drugs that together show the strongest inhibition of cancer cells. Efficient approaches to build these computer models for different cancer cells have been developed, and these models show sometimes remarkable efficiency to find those drug pairs that work best for a specific cancer type. The team also has developed a large data set of experimentally observed drug responses that can be used to test the correctness of computational predictions. To build these computer models, building blocks are required that describe for instance how proteins in a cell interact and take part in the molecular decision process that tells a cell what to do. To standardize this, we have proposed the minimum requirements for how these building blocks should look: the MI2CAST standard, which helps life scientists to adopt a common way to describe causal statements so that they represent valuable information that can be used by computers for building logical models of biological processes. To apply the MI2CAST standard we have built the causalBuilder (https://vtoure.github.io/causalBuilder/index) software tool, for scientists to describe and archive these causal statements so that they can be shared easily with fellow scientists. Furthermore, we have designed a procedure to extract these building blocks from the signaling pathway database Reactome (www.reactome.org), and we have worked together with international consortia to propose to improve the way that computers can retrieve or exchange large sets of these building blocks easily from different databases (the CausalTAB file format and the PSICQUIC language that allows a computer to ask a database for these building blocks.

The COLOSYS project has delivered a drug effect prediction platform that proves to be valuable for identifying the best possible drug combinations for inhibiting specific cancer cells. The platform is now ready to be tested in a clinical setting, where specific models will be designed to represent tumours of specific cancer patients. Drug combinations that the computer model predicts to be most efficient for a specific patient can then be tested in vitro on patient derived cancer cells, after which a validated set can be used by clinicians to design drug therapy for the individual cancer patient. This approach will be applied in the NFR project RESORT (ES656054).

Colon cancer (CC) is a major cause of death. Current treatment involves chemotherapy combined with anti-EGFR or VEGF drugs and radiotherapy. Screening for biomarkers can indicate whether EGFR-inhibitors will be effective in patients, but otherwise biomarkers for personalizing patient treatment are scarce. First-line combinations of chemotherapy and EGFR-inhibitors for patients being RAS wild-type have led to an increase in overall survival to more than 30 months. However, most patients develop RAS mutations under anti-EGFR therapy, or do not respond to EGFRi for unknown reasons. The majority of patients develop resistance and succumb to the disease. We still have a poor understanding of how gene networks drive cancer, how they modulate response, and how they induce resistance to treatment. Good disease models that provide insight are simply lacking. With the availability of large public data resources, our unique collection of patient samples and patient-derived models and with computational and experimental approaches becoming mature, the COLOSYS consortium aims to develop in silico therapy response predictors. These will allow precision medicine, based on patient-specific driver and resistance mechanisms. We will identify new CC tumor driver genes by integration of multiple data types from large public tumor data repositories. A high quality, open repository of data and knowledge (knowledge commons) will be assembled and used to construct multi-scale computer models of the molecular networks that underlie cancerous cell proliferation. Logic model simulations will predict the effect of drugs in cancer cell lines and patient tumors. We will test these predictions on cell lines as well as patient-derived cell cultures, organoids and mouse xenografts, and perform preliminary testing in patients. The combined computational, experimental and clinical testing will provide a thorough understanding of resistance mechanisms, and allow personalised treatment of colon cancer.