Clonal architectures and ancestral relationships of early and late relapses. Clusters were obtained using Sciclone (variational Bayesian mixture model)³⁴  for patients 34 (A), 325 (B), and 62 (C) based on VAFs at PT (x-axis) and R1 (y-axis) or R(n) and R(n + 1). Each dot stands for a mutation (SNVs or indels, WES coverage ≥40×). Different clusters are depicted by different colors and dot shapes (6, 9, and 4 clusters identified for case 34, 325, and 62, respectively). Mutations of interest are annotated. Case 34: The ancestral clone (clone 1), predominant at PT, contained the driver mutation NRAS p.G12D. Clone 2, descendant of clone 1, harbored a nonsense mutation (p.W2006*) in the tumor suppressor gene MLL2/KMT2D and became predominant at relapse (VAF_PT = 0.31, VAF_R1 = 0.55). A low-frequency subclonal mutation (VAF ≤ 0.05) in the oncogene KRAS was detected at diagnosis and rose again at late relapse with a similar frequency (VAF_PT = 0.03, VAF_R1 = 0.05). Case 325: The PT architecture showed a very close structure to that observed at R1 with a persistent ancestral subclone (clone 4) and a similar distribution of subclones. Clone 4 harbored MAPK pathway–activating mutation p.A72D in the SH2 domain of the oncogene PTPN11 subclonal at diagnosis (VAF = 0.23), selected at R1 (VAF = 0.49), and dominant at the 3 subsequent time points (VAF = 0.52, 0.51, and 0.47 in R2, R3, and R4, respectively). Subclonal mutations in driver genes were counter-selected after treatment: KRAS p.K117N (VAF-PT = 0.16 and VAF-R1 = 0.003) and SETD2 p.R456* (VAF-PT = 0.24 and VAF-R1 = 0). A missense mutation in the glucocorticoid receptor NR3C1 (p.P626S) emerged at R1 (VAF = 0.22) and expanded at R2 to become predominant (VAF = 0.42). NR3C1 p.P626S was counter-selected at R3 (VAF = 0) during the most important changes in this tumor history and was replaced by 2 co-occurring mutations in NR3C1 (p.Y641H, VAF = 0.50 and p.D724E, VAF = 0.53) limited to the newly dominant clone 5. This latter clone, descendant of ancestral clone 4, acquired a homozygous loss-of-function mutation p.R134* in the DNA repair gene PMS2 and carried several mutated driver genes: CREBBP (p.Y1503H), IKZF1 (p.F173L), MLL3/KMT2C (p.R2609*), and SRSF2 (p.G12S). This led to the expansion of this clone at R3, which constituted the predominant population at both R3 and R4. Multiple minor subclones, descendant of clone 5 and harboring mutations in TP53 (p.R273H, p.R273C and p.Y236C), MSH6 (p.F573L), and WHSC1 (p.A457T), also emerged at R3 and were maintained as minor clones at R4. Case 62: The dominant population at relapse, probably pulled up by the driver mutation KRAS p.G12D (VAF-PT = 0.39, VAF-R1 = 0.55), harbored a series of newly acquired mutations partly belonging to cluster 3. A subclonal mutation (VAF-PT = 0.10), predicted to alter splicing in the HAT domain of the histone acetyl transferase CREBBP, was lost at relapse (VAF-R1 = 0). Clonal trees depicting the evolutionary history of tumors based on VAFs of somatic mutations predicted as putative drivers in PTs and/or the subsequent relapse(s) of cases 34 (D), 325 (E), and 62 (F). Ancestral relationships between clones were inferred using AncesTree³⁵  under the infinite sites (perfect phylogeny) assumption and are represented by black solid lines. Posterior probabilities of the ancestral relationships are indicated on the side of the black lines. AncesTree uses a probabilistic model for the observed read counts: if Xpj and Xpk are random variables describing the VAFs of mutation j and k in the sample p (PT or the subsequent relapse[s]), Pr[Xpj≥Xpk] denote the posterior probability that Xpj≥Xpk, and the sample with the smallest probability minpPr[Xpj≥Xpk] represents the weakest evidence that mutation j preceded mutation k (if close to 1, j is likely to be ancestral to k). Dashed lines show ancestral clones that existed at the time of sequencing. Each sample is indicated in a colored box at the bottom of the trees (A, B, C, D, and E stand for PT, R1, R2, R3, and R4, respectively). Colored lines indicate the inferred composition of clones and their fraction in each sample (only edges with usage of at least 0.05 are shown).