Clonal evolution of pediatric Acute Myeloid Leukemia Free

Relapse remains the major cause of mortality in pediatric acute myeloid leukemia (AML). Although initiating driver alterations are major determinants of clinical outcomes, the extent to which relapse emerges through clonal selection of pre-existing subpopulations or by the acquisition of de novo mutations remains unclear.

We established a cohort of 39 diagnosis-relapse (D-R) and 2 relapse-relapse (R-R) pairs of pediatric AML. We comprehensively profiled the cohort using whole-genome sequencing (WGS) and target-capture sequencing. WGS (median coverage: 49x) and capture sequencing (450x) were performed to detect single-nucleotide variants (SNVs), insertions and deletions (indels), structural variants (SVs), and copy number variants (CNVs). Somatic alterations with variant allele frequencies (VAFs) exceeding background levels were regarded as real (P<0.05 by binomial tests), and coding genes were classified according to pathogenicity to assign pathogenic or likely pathogenic (P/LP) labels. Enrichment was assessed by Fisher's exact test.

The study cohort represented major high-risk subtypes of pediatric AML, including KMT2A-rearranged (n=13, 32%) and NUP98-rearranged (n=6, 15%) AMLs. A total of 22,088 somatic alterations were identified across SNVs, indels, SVs, and CNVs. Among 142 P/LP mutations, WT1 (n=19) and FLT3 (ITD: n=9, TKD: n=7, other: n=3) were most commonly mutated, followed by NRAS (n=17) and SETD2 (n=9) mutations. The mutational burden was significantly higher at relapse (median 216 vs 376 mutations per sample at D-R, P=9.4×10⁻⁶), whereas the total number of P/LP mutations per sample did not significantly differ (median 4 at both, P=0.34). Among P/LP mutations, 39 had significantly higher VAF at relapse (P<0.05), with frequent involvement of FLT3 (n=7), WT1 (n=5), and SETD2, TP53, CREBBP, and NRAS (n=2 for each), indicating a growth advantage of clones with these alterations. Notably, 107 P/LP mutations were found at diagnosis, and only 35 P/LP mutations were specifically detected at relapse while undetectable at diagnosis, even by targeted sequencing, suggesting acquisition during or after therapy, or selection of rare pre-existing clones below the detection limit estimated to be 1%. Interestingly, clones with these mutations did not always expand at relapse, with different WT1 or FLT3 mutations depleted or enriched within an individual patient. In contrast, NRAS mutations were frequently depleted at relapse, suggesting that clones with NRAS mutations are sensitive to chemotherapy.

We further investigated changes in clonal structures of the disease during progression. VAF at diagnosis and relapse showed clustering of mutations with specific patterns in each patient, and all relapses were clonally related to the paired AML at diagnosis. For example, initiating and category-defining alterations were identified with high VAF at both time points. In most cases (n=35, 85.3%), we observed that non-initiating mutations with high VAF (>20%) at diagnosis were frequently depleted to undetectable levels in the later time points. Contrarily, subclones marked by somatic mutations that were undetectable or with low VAF (<5%) at diagnosis became dominant in 38 cases (92.7%), indicating that leukemic cells were depleted down to only a few persistent genetic clones during remission. One exception is a case with FUS::ERG AML, which showed concordant VAF of somatic mutations at both diagnosis and relapse, including mutations with low VAF, suggesting that the entire population was resistant to chemotherapy with no therapy-induced bottleneck. Notably, mutations that mark subclones fated for relapse were mostly non-pathogenic or non-coding mutations, and 16 cases did not have any P/LP coding cooperating mutations that mark the relapsed clone.

These data indicate that relapse of pediatric AML is driven by two mechanisms: one is a selection or expansion of clones with somatic alterations that confer chemo-resistance or growth advantage over the rest of the clones, and the other is a stochastic, non-genetic mechanism that likely confers a competitive advantage to certain clones, such as transcriptional heterogeneity. Bulk and single cell transcriptional analyses and clonal tracking using ultra-deep sequencing (~5000x) may reveal additional insights into the mechanisms of pediatric AML relapse.