Molecular Landscape of Therapy-related Myeloid Neoplasms in Patients Previously Treated for Gynecologic and Breast Cancers

Definition of therapy-related myeloid neoplasms (TRMN) is only based on clinical history of exposure to leukemogenic therapy. No specific molecular classification combining therapy-related acute myeloid leukemia and therapy-related myelodysplastic syndromes has been proposed. We aimed to describe the molecular landscape of TRMN at diagnosis, among 77 patients with previous gynecologic and breast cancer with a dedicated next-generation sequencing panel covering 74 genes. We investigated the impact of clonal hematopoiesis of indeterminate potential-associated mutations (CHIP-AMs defined as presence at TRMN stage of mutations described in CHIP with a frequency >1%) on overall survival (OS) and the clinical relevance of a modified genetic ontogeny-based classifier that categorized patients in 3 subgroups. The most frequently mutated genes were TP53 (31%), DNMT3A (19%), IDH1/2 (13%), NRAS (13%), TET2 (12%), NPM1 (10%), PPM1D (9%), and PTPN11 (9%). CHIP-AMs were detected in 66% of TRMN patients, with no impact on OS. Yet, patients with CHIP-AM were older and had a longer time interval between solid tumor diagnosis and TRMN. According to our modified ontogeny-based classifier, we observed that the patients with TP53 or PPM1D mutations had more treatment lines and complex karyotypes, the “MDS-like” patients were older with more gene mutations, while patients with “De novo/pan-AML” mutations were younger with more balanced chromosomal translocations. Median OS within each subgroup was 7.5, 14.5, and 25.2 months, respectively, with statistically significant difference in multivariate analysis. These results support the integration of cytogenetic and molecular markers into the future TRMN classification to reflect the biological diversity of TRMN and its impact on outcomes.


Introduction
Therapy-related myeloid neoplasms (TRMNs) arise after cytotoxic chemotherapy and/or radiotherapy administered for a prior neoplasm and include therapy-related acute myeloid leukemia (t-AML) and therapy-related myelodysplastic syndromes (t-MDS) as defined by 2016 WHO classification. 1 TRMN occur in up to 2% of patients with malignancies and represent 10%-20% of all cases of MDS/AML. 2,3 Different hypotheses have been proposed to explain the development of TRMN. 4,5 Inherited predisposition, a rare event, or direct induction of fusion transcripts, well described for KMT2Arearranged AML and acute promyelocytic leukemia (APL), can be responsible for TRMN. Recent evidence suggests that patient with clonal hematopoiesis of indeterminate potential (CHIP) at the time of treatment of their malignancy may have an increased risk of TRMN. 6 CHIP is an age-associated genetic event characterized by one or more somatic mutations in hematopoietic stem cells (HSCs), including mutations in genes such as DNMT3A, TET2, ASXL1, and TP53. CHIP occurs in 10% of healthy individuals over 65 years. In patients with solid tumors, the prevalence of CHIP can rise up to 25%, mainly after chemotherapy exposure, 7 and this is associated with a higher risk of primary hematological malignancies. It has been demonstrated in vitro and in vivo that TP53 and PPM1D mutations confer a clonal advantage to mutated HSC after exposure to chemotherapy. This suggest the potential clonal selection that leads to TRMN in this specific context. 8,9 Five-year overall survival (OS) rates of <10% are commonly reported in TRMN patients 10 ; however, prognosis is mainly driven by cytogenetic and molecular findings: complex karyotype and TP53 mutation-bearing TRMN are known to have a dismal prognosis, 11 whereas therapy-related APL with t(15;17) can be cured without intensive chemotherapy. 12 The current definition of TRMN is mainly based on chronological events and no molecular classification including t-AML and t-MDS together has been proposed, connecting physiopathology, patient characteristics and prognosis. Lindsley et al 13 proposed an ontogeny-based classification for AML, which allows distinction of 3 genetic subgroups, a "TP53 subgroup," an "AML with MDS mutations," and a "de novo/ pan-AML" subgroup which appeared to be relevant in de novo but also t-AML.
In this setting, we aimed to define the molecular landscape of TRMN following treatment for gynecologic and breast cancers, and its impact on clinical outcome, as well as its relationship with the demographic, biological, and clinical features of the population studied. We searched for a suitable molecular classification of TRMN, especially focusing on "CHIP-associated mutations" (CHIP-AMs) and a modified genetic ontogeny-based classifier.

Patients
Within our large single-center database (data protection approval, CNIL GR-2018-01), we identified 113 patients previously treated for breast or gynecologic cancers (the latter including any ovarian, endometrial, or cervical cancer) diagnosed with TRMN  In total, 77 patients fulfilled all inclusion criteria and were retained for the present analysis. For all these patients, we performed Next-Generation Sequencing (NGS) analysis using a 74-gene panel and the Haloplex technique (Agilent), followed by sequencing on a MiSeq instrument (Illumina) (Details in Supplemental Digital materials ; http://links.lww.com/HS/ A189). In addition, CEBPA, NPM1, and FLT3-ITD mutations were screened by PCR and fragment analysis, as previously described. 14 Moreover, paired samples of diagnostic bone marrow aspiration at the time of TRMN and peripheral blood at the time of primary cancer were available in 12 patients for NGS.

Statistical analysis
Clinical, pathological, cytogenetic data, and information regarding treatment and outcomes were collected from the patient's medical records.
The type of first cancer treatment was categorized as chemotherapy alone, radiotherapy alone or chemo/radiotherapy. The type of TRMN treatment was categorized as best supportive care, low-intensity treatment (low-dose cytarabine and hypomethylating agents), intensive treatment (including induction chemotherapy and allogeneic hematopoietic stem cell transplantation [HSCT]).
Time interval between solid tumor diagnosis and TRMN was calculated as time from the date of primary cancer diagnosis to the date of TRMN diagnosis. OS was calculated from the date of TRMN diagnosis to the date of death from any cause or censored at the last follow-up. Event-free survival (EFS) was defined as time from diagnosis to induction failure, relapse, or death from any cause. Database cutoff December 31, 2019 (1 y after the last patient inclusion). Statistical analyses were performed with R software version 3.6.1. The comparison of percentages was carried out with a Pearson's Chi-square test or a Fisher's exact test. The distributions of a quantitative variable according to the modalities of a qualitative variable were compared with a Mann-Whitney test. The distributions of survival data were estimated using the Kaplan-Meier method, compared with the log-rank test and hazard ratios (HRs) with 95% confidence intervals (95% CIs). To identify variables associated with OS, a Cox proportional hazards regression analysis of candidate prognostic factors was performed. All the tests were two-sided and considered to be significant when P <0.05.
It has been suggested that TRMN could emerge by clonal selection. 4 Based on literature data, we classified patients as TRMN with CHIP-AMs or not (Table 1, Supplemental Digital Figure 1; http://links.lww.com/HS/A189) according to the presence of mutations usually found in patients with CHIP with a frequency >1%. CHIP-AMs were detected in 51 patients (66%).
To give substance to this classification we performed NGS on 12-paired samples. Nine of these patients had a CHIP-AM at TRMN stage (Supplemental Digital Table 4; http://links.lww. com/HS/A189). As expected, no CHIP-AM was detected at cancer stage for the 3 patients without CHIP-AM at TRMN stage. In 7 out of the others 9 patients (78%), at least one of the CHIP-AM was detected as preleukaemic clonal hemopoiesis at the cancer stage.

Discussion
In this article, we uncovered the molecular landscape of TRMN with a large NGS gene panel in a cohort of gynecological and breast cancer survivors. TRMN studies are generally very heterogeneous in term of primary cancer type, [19][20][21] and/or focus on a TRMN subtype such as t-AML 13 or t-MDS. 22 We think that our work is a good representation of TRMN in women with breast and gynecological cancers, which represent around one third of female cancer. 23 Moreover, we combine NGS data with detailed patient characteristics to deeply understand mechanisms underlying this secondary disease.
The clinical features of our cohort are quite similar to other TRMN studies, except for a higher percentage of t-AML mainly due to the availability of genomic DNA at TRMN diagnosis (detailed Supplemental Digital Table 1; http://links.lww.com/HS/ A189). Molecular results are in line with previous reports 13,19,20,24 except for ASXL1 mutations (5%), described in 26% of t-MDS 24 and 17% of t-AML. 13 We found a higher frequency of TP53 (31%) mutations and lower frequency of NPM1 (10%) and FLT3 (9%) in TRMN compared to de novo AML/MDS as described by others. 13,19,21,24 More interestingly, we described 9% of PPM1D mutations, a gene usually not included in myeloid NGS panels. PPM1D is a Ser/Thr protein phosphatase that  www.hemaspherejournal.com negatively regulates TP53 and affects functional DNA damage response. The emergence of PPM1D mutations is associated with prior exposure to specific DNA-damaging agents as it has been shown for TP53. 25 Indeed, PPM1D mutations provide a survival advantage onto hematopoietic clones by rendering them resistant to apoptosis and confer to HSC resistance to chemotherapy leading to expansion during cancer treatment. The exact role in leukemogenesis is unknown, and doubt exists to know if this mutation is more a passenger or driver mutation. It has been shown in a large series of MDS patients 22 that PPM1D was more present in t-MDS (14%) than de novo MDS and often co-occurs with TP53 (44%) with a median variant allele frequency (VAF) of 5%. In our cohort, 7 patients had PPM1D mutations, 5 of them had a complex karyotype and the 2 others had monosomy seven. VAF of PPM1D mutation was low in patients with TP53 mutation (2%, 8%, and 9%) but high in patients without TP53 mutation (19%, 24%, 28%, and 41%) suggesting that his role in leukemogenesis may be considered depending on the presence of TP53 mutation or not. Although TRMN is recognized as a distinct entity in 2016 WHO classification of hematological malignancies, 1 TRMN remains a very heterogeneous disease. NGS could help distinguishing different entities that should be considered separately. We first evaluated impact of "CHIP-AM" mutations. The recent discovery of CHIP in healthy individuals suggested that myeloid neoplasms may have a premalignant condition characterized by clonal hematopoiesis. 7,26 As shown in this study and by others, 27,28 the existence of CHIP-AM at the cancer stage is detectable in 75% of the patients. Undetectable CHIP can be due to the detection limit of the NGS assay, not efficient under 0.1%, but it gives us some confidence to extrapolate that the majority of CHIP-AM we identified in the TRMN cohort was indeed present at the cancer stage in a minor clone. "CHIP-AM" patients (66% of our cohort) were older at TRMN diagnosis and the time interval between TRMN and first cancer was longer than the "no-CHIP" patients. Patients with CHIP-AM more frequently had an MDS phenotype, a complex karyotype and less commonly a balanced translocation. Two different peaks of incidence in TRMN have been well described. 4,29 The first one occurred with a short latency (2-3 y). This mechanism is mediated by topoisomerase 2 inhibitors, which induce a double-strand break during DNA replication and can link 2 DNA strands together after replication, leading to fusion oncogenes responsible for t-AML. 30 The second peak occurred with a long latency, usually described as following treatment with alkylating agents and/or radiation therapy, mimicking MDS features. Interestingly, our "CHIP-AM" and "no-CHIP" categories fit with this description, supporting the idea of a preexisting clone emerging under chemotherapy or radiotherapy in TRMN with CHIP-AM. In the healthy population, most common CHIP mutations are DNMT3A (52%), TET2 (9%) and ASXL1 (8%). TP53 and PPM1D are found in only 3% and 5% (Supplemental Digital Table 2; http://links.lww.com/HS/A189). 7,26 In cancer patients, PPM1D and TP53 CHIP mutations are overrepresented, 6,25,31 especially due to exposure to both chemotherapy and radiotherapy. Moreover, TP53 and PPM1D variant allele fraction rise under cancer treatment as opposed to TET2 and DNMT3A mutations. Recently, 2 large studies have shown that it was possible to predict the AML risk in healthy individuals years before diagnosis, based on the detection of CHIP. 32,33 Interestingly, TP53, IDH1/2 and spliceosomal mutations (including SRSF2 and U2AF1) are associated with a higher risk of subsequent AML, in contrast with other mutations such as DNMT3A and TET2 mutations. Larger studies will help to clearly distinct the role of each mutation in the development of TRMN, but we can extrapolate that the presence of DNA damage response gene mutations (ie, TP53, ATM, CHEK2, and PPM1D) could be considered as a preleukemic stage increasing the risk of TRMN. By contrast, the role of previous cancer treatment in TRMN emergence in TRMN with TET2/DNMT3A mutations is uncertain. These findings could be an explanation for the chemoresistance of TRMN with DNA damage response mutations.
We next thought that the ontogeny-based classification proposed by Lindsley et al 13 for AML could allow a perfectly understandable distinction of genetic subgroups. We proposed some adjustments given that MDS patients and PPM1D mutations were not taken into account in the study by Lindsley et al. Based on the close interaction with TP53 in DNA damage response, we decided to consider PPM1D in the TP53 group more than "MDS-like" or "de novo/pan-AML"group in our Lindsley's modified classifier, but larger series will help to clarify "prognosis role of PPM1D mutations" in the future. This classification segregates TRMN with clinical, biological, and survival differences. A "TP53/PPM1D" subgroup including patients with long history of cancer treatment and complex cytogenetics, a "MDSlike" subgroup with older patients, similar to standard high-risk MDS or secondary AML, and a "de novo/pan-AML" subgroup in which most patients have a balanced chromosomal translocation. These genetic subgroups correlate with OS and appear to be more efficient than morphologic distinction between t-AML and t-MDS, suggesting that the next TRMN classification would benefit from the incorporation of cytogenetic and molecular markers.
In conclusion, our study highlights the importance of genomic characterization of TRMN for prognosis as well as a proper understanding of oncogenic mechanisms. The integration of genetic features into the future TRMN classification could improve our understanding of the biological diversity of TRMN and our ability to predict clinical outcome. The most important challenge is now to improve the OS of TRMN patients. Development of news drugs such as VYXEOS, a liposomal formulation of cytarabine and daunorubicin has shown very impressive results in patients with t-AML fit for intensive chemotherapy. 34,35 In phase 1b/2 in combination with 5-AZACYTINE, APR-246, a small molecule that selectively induces apoptosis in TP53-mutated cancer cells, showed promising results in unfit TP53 mutated AML/MDS patients. 36,37 However, much therapeutic progress has still to be made for TRMN patients.

Disclosures
JBM received honoraria from Abbvie, Jazz Pharmaceuticals, and Astellas. SDB received honoraria from Agios, Celgene, Forma Therapeutics, Abbvie, Astellas, Daichi, Novartis, Pfizer, and Jazz Pharmaceuticals and has received research funding from Agios and Forma Therapeutics. CM received honoraria from Astellas. SD received honoraria from Pfizer, AstraZeneca, Roche Genentech and has received research funding from Novartis, Pfizer, AstraZeneca, Roche Genentech, Lilly, Puma, Myriad, Orion, Amgen, Sanofi, Genomic Health, GE, Servier, MSD, BMS, and Pierre Fabre. AL reports grants, personal fees and nonfinancial support from AZ, grants, personal fees and nonfinancial support from Tesaro, grants, personal fees and nonfinancial support from clovis, grants and personal fees from MSD, personal fees from biocad, grants and personal fees from ability, other from merck serono, personal fees from seattle genetics, grants, nonfinancial support and other from GSK, personal fees from Zentalis, outside the submitted work. ER received honoraria from BMS, Clovis, Astra Zeneca. The other authors have no conflicts of interest to disclose.