The Relapse Mechanisms and Genomic Landscape Differ in KMT2A-r Pediatric Leukemia in Relation to Relapse Time

The genomic landscape and mechanisms driving relapse in KMT2A-rearranged (KMT2A-r) infant and childhood acute lymphoblastic (ALL) and acute myeloid leukemia (AML) are not completely understood. We therefore studied 36 KMT2A-r ALL (n=19) and AML (n=17) patients of which 25 relapsed and 11 remained in remission. Twenty diagnose-relapse-germline trios and 5 multiple relapse samples were analyzed by whole genome (WGS) and whole exome sequencing (WES) and 30 patients longitudinally by using patient-specific mutations identified by WGS/WES, including the KMT2A-r (average coverage 3300X).

The mutational burden increased from diagnosis to relapse and relapse evolved through branching evolution. Relapse was seeded by multiple diagnostic clones in 56%, by a single sweeping clone detected at diagnosis in 22%, and by a single sweeping clone not detected at diagnosis in 22%. Notably, the evolutionary patterns correlated to relapse time, where multiple diagnostic clones seeding relapse were connected to an earlier relapse with all very early relapse ALL (3/3, relapse <9 months from diagnosis) and half of the early AML relapse showing this pattern (2/4, relapse <1 year in complete remission, CR1). By contrast, later relapse was connected to a sweeping clone at relapse with 67% of early relapse ALL (>9 months from diagnosis) and 40% of late relapse AML (>1 year in CR1) showing this pattern.

Pathway analysis showed that cell cycle genes, glucocorticoid signalling, purine metabolism, mismatch repair, and B-cell differentiation, were enriched in early relapse ALL (83%, 5/6) and included TP53, CREBBP, NT5C2, PMS2, PRPS2, NR3C1, IKZF1, with none of the very early relapse infant ALL harboring such alterations (n=4). Further, TP53 and IKZF1 alterations co-occurred (n=4/4). These results were validated in public data sets of 98 KMT2A-r ALL infants (n=84) and children (n=14) at diagnosis and relapse (n=24) and showed that 50% of early relapse ALL, and none of the 8 very early relapse ALL, had such alterations. Ultra-deep sequencing did not detect the CREBBP, NT5C2, PRPS2 or TP53 mutations at diagnosis and manual inspection of the WGS reads failed to detect the PMS2 and NR3C1 deletions. In AML, TP53 and CCND3 alterations were maintained, and gain of WT1 was seen in late relapse AML.

Signalling mutations were the most common type of mutations at diagnosis (64%) and relapse (56%) and the frequency was similar in patients that remained in remission and in those that relapsed (55% versus 60%). One infant ALL and four AML patients had multiple relapses, allowing us to study how the genetic landscape evolved across consecutive relapses. This showed a stepwise replacement of clones during treatment in agreement with a fitter clone that evolves under chemotherapeutic selective pressure.

Longitudinal analysis allowed sensitive detection of residual leukemia cells and showed that the relapse clone could be detected at diagnosis in 64% of patients. Further, infants with >10% of molecularly detectable leukemia cells after induction therapy, had a high risk of a very early relapse. Ultra-deep sequencing allowed detection of the relapse clone up to 4 months before relapse. In 11 of the 30 patients (3 remission and 8 relapse), low-frequency KMT2A-fusion positive leukemic cells were found at remission outside of the MRD time points. Our longitudinal data also provided unique insights into clonal response to treatment by showing that 1) a change in therapy can favour the eradication of one clone and expansion of another, 2) a clone that initially was the most sensitive clone to therapy, was the one that eventually caused relapse, and 3), a diagnostic clone can be undetectable for a long time before expanding to cause relapse, suggesting that molecular monitoring with personal mutations is a powerful tool to follow response to therapy.

These results provide new biological insights into the relapse mechanisms in KMT2A-r leukemia. The data shows different clonal evolution patters depending on when in time the patient relapsed, with very early relapse ALL being seeded by multiple diagnostic subclones and a paucity of acquired genetic alterations at relapse. By contrast, early relapse ALL was characterized by a single diagnostic clone seeding relapse by a clonal sweep along with acquired mutations in chemoresistance-associated genes. To validate and extend these findings, we are currently analyzing 11 additional infant relapse samples with WGS.