Preclinical disease models are always flawed. In the case of patient-derived xenograft (PDX) mouse models of human cancer, a new study in Nature Genetics has catalogued some of the ways in which these 'avatars' lead scientists astray.

PDX models of cancer — in which cancer cells are harvested from patients and injected into mice — have become increasingly important in cancer research. Not only do researchers think that these in vivo models of cancer are more representative of human disease than in vitro cell line approximations, but they also offer the promise of enabling an empirical approach to precision medicine. By taking cells from a patient and building a personalized mouse avatar, researchers can, in theory, test targeted drugs and chemotherapies on an individualized basis before choosing a best option.

Given the growing importance of these models, cancer geneticist Todd Golub, at the Broad Institute in Massachusetts, and colleagues set out to study how well PDX mice mimic human disease. They assessed over 1,100 models, and found considerable genomic and phenotypic differences in terms of the evolution of PDX and human cancers. PDXs rapidly picked up copy number alterations as they were passaged, which led to different genomic architectures from those of the originating tumours. “The assumption is that what grows out in the PDX is reflective of the bulk of the tumour in the patient,” Golub told Nature News . “But there's quite dramatic resculpting of the tumour genome.” These changes may influence how PDXs respond to chemotherapy and targeted drugs, he cautions, raising new considerations for researchers that rely on these models.

Nevertheless, some groups hope that new approaches to PDX-model generation — including the use of mice that have been genetically engineered to carry human genes — could make these models less flawed.