hide
Free keywords:
cancer models; applied mathematics
Abstract:
Modern cancer therapies aim at targeting tumour-specific alterations, such as mutations or neo-antigens, and maximal treatment efficacy requires that targeted alterations are present in all tumour cells. Currently, treatment decisions are based on one or a few samples per tumour, creating uncertainty on whether alterations found in those samples are actually present in all tumour cells. The probability of classifying clonal versus sub-clonal alterations from multi-region profiling of tumours depends on the earliest phylogenetic branching event during tumour growth. By analysing 181 samples from 10 renal carcinoma and 11 colorectal cancers we demonstrate that the information gain from additional sampling falls onto a simple universal curve. We found that in colorectal cancers, 30% of alterations identified as clonal with one biopsy proved sub-clonal when 8 samples were considered. The probability to overestimate clonal alterations fell below 1% in 7/11 patients with 8 samples per tumour. In renal cell carcinoma, 8 samples reduced the list of clonal alterations by 40% with respect to a single biopsy. The probability to overestimate clonal alterations remained as high as 92% in 7/10 renal cancer patients. Furthermore, treatment was associated with more unbalanced tumour phylogenetic trees, suggesting the need of denser sampling of tumours at relapse.
