An algorithm developed by researchers at the University of Cambridge can help detect if tumors have DNA mismatch repair deficiencies and whether they will be vulnerable to therapies targeting these mutations.
DNA mismatch repair is a cellular system that, if functioning correctly, allows any errors occurring during the production of new DNA to be corrected. However, if this system fails then it can result in cancer. This can occur as a result of mutations or epigenetic changes in the DNA and also in response to environmental exposures to water, oxygen, sunlight, or a variety of chemicals.
The researchers first looked at the function of 43 genes known to be involved in mismatch repair by using CRISPR-Cas9 to create and study different cell cultures with genetic alterations in these genes.
The team found that nine genes in particular—OGG1, UNG, EXO1, RNF168, MLH1, MSH2, MSH6, PMS1 and PMS2—played a significant role in DNA damage repair.
“When we knock out different DNA repair genes, we find a kind of fingerprint of that gene or pathway being erased. We can then use those fingerprints to figure out which repair pathways have stopped working in each person’s tumour, and what treatments should be used specifically to treat their cancer,” explained Serena Nik-Zainal, M.D., Ph.D., a senior researcher at Cambridge University’s MRC Cancer Unit, who led the research.
Using the information collected by studying the different cell lines, Nik-Zainal and colleagues created an algorithm called MMRDetect, which they trained on whole-genome sequence data from the 100,000 genomes project. It is designed to pick up tumors that have mismatch repair mutations that should make them more vulnerable to treatment with immune checkpoint inhibitors.
In 7695 cancers with available genomic sequence, the algorithm successfully detected mismatch repair deficiencies. “MMRDetect has enhanced sensitivity, particularly at detecting mismatch repair deficient samples with lower mutation burdens, although it could miss cases where mismatch repair deficiency is present at a very low level,” write the authors in the journal Nature Cancer.
The team says their algorithm could be very useful for directing clinicians to therapies, such as immune checkpoint inhibitors, that might be more effective in patients with these kinds of mutations. However, they caution that more work is needed to refine the model. It was also primarily trained on highly proliferative colorectal cancers, so might be less accurate for other tumor types at present.
“As a community, we are at the early stages of seeking experimental validation of mutational signatures. However, we hope that our approach, which leans on experimental data, provides a template for improving biological understanding of how mutational patterns arise, and this, in turn, could help propose improved tools for tumor characterization going forward,” the authors conclude.