Mutation of Epigenetic Regulators at Diagnosis Is an Independent Predictor of Tyrosine Kinase Inhibitor Treatment Failure in Chronic Myeloid Leukemia: A Report From the RESIDIAG Study

Abstract

Additional mutations at chronic myeloid leukemia (CML) diagnosis have been shown to variably affect tyrosine kinase inhibitor (TKI) response [1], and were inconstantly detected at loss of response [2]. Contradictory observations may have resulted from difficulties in reliably inferring CML clonal architecture from mutations quantified by NGS, BCR::ABL1 by qRT-PCR, and ABL1-tyrosine kinase domain (ABL1-TKD) mutations by RNA-Seq. In the RESIDIAG (RESistance molecular markers at DIAGnosis) study, the mutational profile of 117 CML patients (n = 60 responders and n = 57 nonresponders) (Table S1) was analyzed at diagnosis (both groups) and at relapse (nonresponders only) by asymmetric capture sequencing (aCAP-Seq, Table S2) [3] to identify molecular events that predict TKI failure and decipher the clonal architecture and the order of acquisition of mutations relative to BCR and ABL1 fusion. This study complied with French regulations and was approved (no. 2019_048) by the Henri Mondor Institutional Review Board (No. 00011558). The study methodologies conformed to the standards set by the Declaration of Helsinki. All patient data were anonymized and de-identified before analysis. Informed consent was obtained from all participants.

The median time follow-up of responders was 7.1 years. There were no significant differences in terms of age, sex, CML stage, first-line treatment, additional chromosomal abnormalities (ACA), or first-line therapy between the two groups, while the proportions of patients with high Sokal or The EUTOS long-term survival (ELTS) scores were significantly increased among nonresponders (p < 0.001, Pearson's chi-square test, Table S3). Both ELTS and Sokal scores predicted failure-free survival (p < 0.001, Log-rank test, Figure S1). Patients in both groups were mainly treated first-line with Imatinib (61.5%), Nilotinib (25.6%), or Dasatinib (10.3%). TKI switch before failure analysis was mostly due to first-line intolerance. Blast crisis (BC) progression occurred in eight nonresponders, including four myeloid and four lymphoid BC, with a median time of transformation from diagnosis at 15 months [8.6–24.3 months].

At diagnosis, the number of additional mutations per patient was higher in nonresponders (p < 0.001, Pearson's chi-squared test, Table S4), especially in ASXL1, DNMT3A, and TET2 referred to as epigenetic genes hereafter (p < 0.001, p = 0.02, p = 0.02, respectively, Pearson's chi-squared test, Figure 1A, Figure S2A,B and Table S4). The average ASXL1 mutation VAF in nonresponders (23.6% ± 3.6%, n = 21) were significantly different from the BCR::ABL1 frequency (47.9% ± 0.8%, p < 0.0001, Dunnett's multiple comparison test, Figure S2C) suggesting that ASXL1 mutant were either CML subclones or clones driving an independent clonal hematopoiesis of indeterminate potential. In contrast, most DNMT3A and TET2 mutations were present in nearly all leukemic cells at diagnosis (Figure S2C).

FIGURE 1

Open in figure viewerPowerPoint

Mutational landscape of CML patients and its impact on TKI response. (A) Mutational profile at first TKI failure. MMEJ (microhomology end-joining, pink squares) includes microhomology domains ≥ 3 bp at BCR::ABL1, ABL1::BCR, and deletion breakpoints (B) Failure-free survival according to the presence of epigenetic regulatory gene mutations at diagnosis. (C) Cumulative incidence of ABL1-TKD mutations according to the presence (pink line) or the absence (black line) of epigenetic mutations at diagnosis. (D) Number of BCR::ABL1 (B::A) and/or ABL1::BCR (A::B) breakpoints identified among the total RESIDIAG Cohort (n = 117). (E) Genomic coordinates of BCR::ABL1 breakpoints in BCR (upper diagram, light blue vertical lines) or ABL1 (lower diagram, red vertical lines). Hg19 genomic coordinates are indicated. Exons (e) are represented as blue rectangles, UTR as thinner blue rectangles, and intron as horizontal blue lines with arrows. (F) Cumulative incidence of TKD mutations according to the presence of an MMEJ signature adjacent to genomic breakpoints of somatic genetic events including BCR::ABL1, ABL1::BCR, or somatic deletions (MMEJ all). ABL1-TKD: ABL1 tyrosine kinase domain. ACA: additional chromosomal abnormality; BP: breakpoint; Ph: Philadelphia chromosome; TKI: tyrosine kinase inhibitor; 2nd generation TKI: nilotinib, dasatinib, or bosutinib.

Univariate Cox regression analysis was performed to identify factors associated with TKI failure-free survival (FFS) at diagnosis (Table S5). High-risk ACA, high-risk Sokal or ELTS scores, the total number of mutations, and the presence of mutations in epigenetic genes were significantly associated with an increased rate of TKI failure (Table S5). FFS was significantly lower in patients carrying mutations in epigenetic genes regardless of the genes involved (median FFS between 11.2 and 12.8 months as compared with median FFS unreached after 175 months for unmutated patients, p < 0.001, Log-rank test, Figure 1B).

A multivariate Cox regression model identified both high-risk ELTS score (Hazard ratio, HR = 3.75 [1.70–8.29], p = 0.001) and mutations in epigenetic genes (HR = 2.67 [1.33–5.35], p = 0.006) as independent predictors of TKI failure (Table S5). Integrating epigenetic mutations at diagnosis with ELTS scores into the Cox multivariate regression model increased the concordance index for failure prediction to 74% compared with 64% with ELTS alone. A conditional inference tree analysis identified the presence of mutations in epigenetic genes as the best classifier (p < 0.001, Fine & Gray test), followed by a high ELTS score (p = 0.001) (Figure S3) to predict TKI failure. Combining these two variables hierarchically allowed reclassification of ELTS intermediate-risk patients into either high-risk TKI failure (node 4 median FFS = 12.4 months [6.3-not reached and node 5], median FFS = 12.3 months [11.2–36.5], 19.4% [7/36] of patients with ELTS intermediate score) or low risk of TKI failure (median FFS not reached at 175 months, 80.6% [29/36] of CML patients with intermediate ELTS scores). Finally, 14.8% (8/54) of patients with low-risk ELTS scores were reassigned to the high-risk failure group due to the presence of epigenetic mutations at diagnosis.

At failure, patients could be categorized into three groups (Figure 1A): patients harboring ABL1-TKD mutations (n = 20, 36%), patients with mutations excluding ABL1-TKD (n = 18, 33%), and those without mutation (n = 17, 31%). In the first group, 75% of ABL1-TKD mutations (Figure S4A), co-occurred with mutations in epigenetic genes (Figure 1A and Figure S5A). In the second group, 100% had mutations in the epigenetic genes (ASXL1 = 55.6%, DNMT3A = 27.8%, and TET2 = 27.8%). Importantly, in all cases, at least one epigenetic mutation was already present at diagnosis (Figure 1A), and 3/18 patients harbored double mutants DNMT3A/TET2 at failure (Figure S5B).

Quantification of the epigenetic mutations VAF along with ABL1-TKD and BCR::ABL1 by aCAP-seq revealed the subclonal emergence of epigenetic mutations within the CML clone in 88% (n = 22/25 interpretable kinetics) followed by an ABL1-TKD mutation in 10 patients (Figures S4B,C and S6A, upper panels). Interestingly, mutations in epigenetic genes at diagnosis were associated with an increased cumulative incidence of ABL1-TKD mutations (p = 0.015, Figure 1C, Gray test). The second pattern of VAF kinetics was indicative of CML arising from clonal hematopoiesis (n = 3/25, 12%), driven by a DNMT3A mutant (UPN79), an ASXL1 mutant (UPN19) or a double DNMT3/TET2 mutant (UPN118) (Figure S6B,C).

ASXL1 mutations were the only additional mutation at first TKI failure in eight patients, and VAF kinetics were consistent with their presence in the clone, driving TKI failure in six of them (UPN7, UPN15, UPN19, UPN81, UPN82, and UPN117, Figure S6A,B). Another pattern of clonal evolution was the acquisition of transcription factor mutations such as WT1, CEPBA, RUNX1, and IKZF1. These were associated with CML progression (Figure S5C) but were also present in responders (IKZF1, WT1, and RUNX1) at diagnosis (Figure S2A).

Breakpoint (BP) sequences were identified for 112/117 patients (95.7%, Figure 1D) in BCR and ABL1. BCR::ABL1 BP were mostly located in intron 13 or 14 of BCR (NM_004327.4), except for seven patients whose BP were in exon 14 (n = 5) or exon 15 (n = 2) (Figure 1E). ABL1 breakpoints were evenly distributed between the 5′UTR of ABL1 (NM_007373.3) and ABL1 exon 2 except for three patients (UPN1, UPN70, and UPN120) were found up to 3.8 kb upstream of the ABL1 transcription start site. One BP was located in ABL1 exon 1a, 88 were located between exons 1a and 1b, and eight were found between exons 1b and exon 2.

The prevalence of BCR::ABL1 BP location in specific DNA domains as defined by the RepeatMasker track of the hg19 UCSC genome browser (last update 2020-02-20) in BCR (n = 28) and ABL1 (n = 58) was not statistically different between responders and nonresponders (p = 0.0547; Pearson's chi-squared test, Figure S7). Nevertheless, among all detected fusion BP from BCR::ABL1 and ABL1::BCR, 19.1% (21/110 patients analyzed) of CML patients had adjacent microhomology sequences ≥ 3 bp that could not be present by chance (3.52%, p < 0.0001, one sample proportion test). This was, therefore, highly suggestive of microhomology end-joining (MMEJ) repair machinery involvement in BCR and ABL1 fusion formation. This MMEJ signature was significantly more frequent (p = 0.012, Chi-square test) in nonresponders (16/55, 29.1%) than in responders (5/55, 9.1%, Table S4). Considering all microhomology sequence domains next to deletions and fusion breakpoints, their presence predicted TKI failure (p = 0.019, HR = [1.12–3.55], univariate Cox regression analysis, Table S5) and was associated with an increased cumulative incidence of TKD mutations with a median time of 8.57 months [1.87–41.26] as compared with a median time of 17.56 months [7.75–88.08] for other CML patients (p = 0.009, Log-rank test, Figure 1F).

In conclusion, these findings support a model in which epigenetic gene mutations emerge mainly within the CML clone, leading to subclonal outgrowth, but a single mutation in CML cells cannot solely drive TKI failure. Instead, such mutations promote additional genetic events such as ABL1-TKD mutations that contribute to therapeutic failure. In line with this model, Skorski's group demonstrated that TET2 mutations can impact DNA double-strand break repair mechanisms by favoring the mutagenic MMEJ repair mechanism over homologous recombination or c-NHEJ [4]. Interestingly, microhomology domain sequences at BCR::ABL1 or somatic deletion breakpoints were also associated with the emergence of ABL1-TKD mutations providing a useful biomarker for targeted therapeutic interventions, such as PARP inhibitors, which have demonstrated promising in vitro activity in CML [5].

Finally, the results of the RESIDIAG study demonstrate that combining high throughput sequencing with the ELTS score at diagnosis allows TKI failure prediction with a concordance of 74%. These findings underscore the potential benefit of implementing high-throughput sequencing analysis for CML patients upon diagnosis [1]. Based on RESIDIAG study results, high-throughput sequencing analysis at CML diagnosis was mostly informative in refining TKI failure prognostication for CML patients with intermediate ELTS scores. From a cost-efficiency perspective, it could thus be restricted to this population of CML patients. However, the best therapeutic options for CML patients with additional mutations at diagnosis remain to be identified.