Therapy-induced mutagenesis in relapsed ALL is supported by mutational signature analysis

RESPONSE

We appreciate the interest of Gaynon et al¹  in our study on acute lymphoblastic leukemia (ALL) relapse.²  They argued against our finding that therapy-induced drug resistance mutations occur in ALL and stated that these mutations were potentially preexisting at diagnosis but undetectable. They focused primarily on our mathematical modeling of ALL doubling times, while ignoring other important biological evidence from our study.

First, our strongest evidence of therapy-induced drug resistance mutations came from mutational signature analysis, not the modeling of doubling times. We showed that >25% of relapsed ALLs bore 1 or 2 treatment-induced mutational signatures, referred as novel signatures A and B. We showed experimentally that novel signature B was induced by thiopurine treatment, and the therapy-induced mutations were clonal in most relapses (supplemental Figure 10); hence therapy does not simply induce minor subclonal variants.

Second, drug resistance mutations in relapsed ALL preferentially occurred at trinucleotide contexts mutated by therapy. For example, relapse-specific NT5C2 R367Q mutations were C>T mutations occurring at the center of TCG trinucleotides. We showed experimentally that thiopurines cause this type of mutation, and NT5C2 R367Q mutations were significantly enriched in relapses that bore the thiopurine signature (supplemental Figure 16). We further analyzed the probability that each drug resistance mutation was caused by individual signatures present in each leukemia sample, based on each signature’s preference to mutate at specific trinucleotides, using a published method.³  This indicated that NT5C2, TP53, NR3C1, and PRPS1 drug resistance mutations were likely induced by treatment in a subset of patients. A separate manuscript in preparation strongly reinforces this paradigm.

Gaynon et al also misinterpreted our conclusions; we did not dismiss the contribution of preexisting clones to relapse. They cited the study of Dobson et al,⁴  in which we participated as collaborators, including part of the genomic analysis, as evidence that a minor subclone is usually present at diagnosis, which leads to the eventual relapse. Similarly, we also showed in supplemental Figure 17 that in 80% of patients the relapse-fated clone was detectable at diagnosis. Indeed, our Visual Abstract and Figure 7 showed that even in cases where drug resistance mutations were likely therapy induced, this mutagenesis occurred within a minor clone present at diagnosis, which further evolved as a result of therapy (Figure 1). We referred to this minor-to-major clone as a “persistent clone” because it was not fully drug resistant but later acquired a bona fide drug resistance mutation, such as in NT5C2, which we never detected at diagnosis using 500× capture sequencing across the cohort nor with 4000 to 50 000× sequencing in 3 patients. Minor survivor subclones lacking drug resistance mutations can be readily detected at diagnosis, whereas the resistance mutations evolving from these subclones, such as in NT5C2 and PRPS1, are rarely if ever detected using high-depth sequencing at diagnosis (Figure 3 of original paper,²  and Figure 1), including in Dobson et al⁴  and other studies,^5,6  consistent with multistep acquisition of frank resistance. Furthermore, Dobson et al were not able to detect therapy-induced mutational signatures because they used exome sequencing of 14 patients, yielding an insufficient number of mutations to analyze mutational signatures (median of only 25 in diagnostic and 39.5 mutations in relapsed samples), compared with our whole-genome sequencing of >100 patients (median 332 at diagnosis and 810 mutations at relapse).

Finally, Gaynon et al misstated our conclusions from our mathematical modeling, when they noted: “With relapses outside these limits (later than 213 to 374 days), the authors conclude that the mutations could not have been present in even a single cell at diagnosis.” Our actual statement was: “Early relapses [9 to 36 months after diagnosis], by contrast, may have occurred through a 2-step process in which a “persister” clone survives treatment yet cannot proliferate until acquiring a bona fide resistance mutation during treatment ... which is also supported by the mutational signature analysis presented later. Alternatively, early relapses may have occurred through delayed proliferation (>9-day doubling time), perhaps during specific treatment regimens.”²  Thus, we clearly noted that relapses after 9 months could have been due to a preexisting resistant clone with slower proliferation. The statement of Gaynon et al that “[d]oubling times are not constant or uniform among the heterogeneous leukemic population” agrees with what we stated in the quotation above. Thus, we did not conclude that post-374-day relapses had to be from therapy-induced clones, but only claimed that at least 20% of relapses after 9 months arose from clones harboring therapy-induced mutations based on mutational signature data (see Visual Abstract and Figure 7). The only firm conclusion we made from the modeling was that very early relapses (<9 months) likely arose from ready-made resistant clones present at diagnosis (see Figure 3B).

We thank Gaynon et al for their provocative discussion and agree that more needs to be done to fully understand the biology of leukemia relapse.