Fig. 4: Tracking the metastatic cascade in the plasma of LEGACY patients.

Contribution of different tumour samples to ctDNA in LEGACY patient 1 (a) and patient 2 (b) using multiple linear regression analysis of somatic mutations. Similarly, contribution of tumour sites to ctDNA can be measured using methylation clock analysis (epimutations instead of nucleotide substitutions), as presented here for LEGACY patient 1 using two clock-like methylation regions that have been previously validated as subject to methylation drift, namely ZNF454 (c) and IRX2 (d). e With the same data we can also reconstruct the tumour phylogenetic tree using methylation clock haplotypes for ZNF454, corroborating the overall phylogenetic structure revealed by whole-genome sequencing. f This is because epimutations contain phylogenetic information about diverging cell lineages. g The same can be done for IRX2, leading to the construction of a consistent tree. Notably, methylation clock analysis can detect clones down to 1% prevalence and is >1000 less costly than whole-genome sequencing. h–k The same methylation clock analysis has been applied to an independent cohort of early breast cancer patients with matched primary, lymph node and ctDNA. High correlation (Pearson) between frequency of methylation haplotypes in the primary tumour of patient 6 was found for both clocks ZNF454 (p = 4.94 × 10⁻⁶³) (h) and IRX2 (p = 3.11 × 10⁻¹³) (i). Also Patient 16 demonstrated high correlation between methylation patterns in ctDNA and a specific lymph nodal lesion, again for both ZNF454 (p = 7.86 × 10⁻⁶⁰) (j) and IRX2 (p = 7.51 × 10⁻⁵⁹) (k). P values refer to tests of no correlation.