Figure 2: Information gain from multi-region sequencing in patients with clear cell renal carcinoma.

(Panels a1 to j1): If from a set of n multi-region samples from a patient we consider different subsets of samples (n is between 5 and 11 per patient) with size i = 1,2,…n, we will identify different numbers of putatively clonal alterations, with great variation between different sets of the same size. The more samples we consider, the closer we get to the minimal identifiable set of clonal mutations, i.e. mutations that may have appeared clonal with one or few samples, turn out to be indeed sub-clonal in the whole tumour. (Panels a2 to j2): The probability to find the minimal set of clonal mutations falls onto the universal curve (8). Dots represent the data; lines correspond to best fits of f via Equation (8). In 2 cases (c2 and j2) we find a balanced left and right side (f = 0.5). One case (i) appears slightly unbalanced (f = 0.32) while all other cases are unbalanced (f < 0.01), supporting the presence of convergent evolution and on-going clonal selection. All patients but (i2) and (j2) developed metastasis. Only patients (h2 to j2) are treatment naïve. For balanced tumours, the information on the true set of clonal alterations quickly plateaus with few samples (for example 5 samples in patient (j)). (Panels a3 to j3): We repeat the inference of the balancing factor f on all available combinations of subsets of tumour samples with a minimum of 4 samples. The violin plots show the corresponding distributions of f values for each possible combination of i = 4,5,…n − 1 subsets. Most combinations of samples resemble the balancing inferred from the full data set. However, there is a trend towards a bimodal distribution for small i, which might be a direct consequence of the spatial evolution of tumours. Note that violin plots show the probability density distribution of the f-values. The actual f-values are never negative. Data from Gerlinger et al.²².