Therapy-Related Myeloid Neoplasms: Complex Interactions among Cytotoxic Therapies, Genetic Factors, and Aberrant Microenvironment

Cancer is the second leading cause of death worldwide, accounting for ∼9.6 million deaths in 2017 or nearly one in six deaths (1). Though the incidence of cancers continues to increase, the risk of death from cancer has decreased by a third over the last 20 years, translating into 3.5 million averted cancer deaths. The long-term cancer survivorship, although celebrated, presents a new set of challenges for survivors and their caregivers. One of the emerging challenges is the increasing incidence of a second primary malignancy. This can occur for a variety of reasons and includes a variety of cancer types, most commonly skin cancer. However, one of the most feared types of second primary cancer is therapy-related myeloid neoplasm (t-MN), which can present either as therapy-related acute myeloid leukemia (t-AML) or therapy-related myelodysplastic syndrome (t-MDS) and, by definition, occurs in patients previously exposed to cytotoxic therapies (2, 3). The prognosis of t-MN is universally dismal due to its aggressive nature and treatment refractoriness to conventional therapies, including stem cell transplantation (SCT). Encouragingly, new preclinical research identifies novel vulnerabilities in certain t-MN molecular subtypes that may lead to improved management. Meanwhile, however, it is critical for clinicians, molecular pathologists, and cancer biologists to understand the risk factors associated with t-MN in an effort to anticipate and even reduce the incidence in the long term. Finally, better understanding of the molecular pathogenesis of t-MN may also lead to preemptive screening and even intervention in high-risk patients. This review summarizes emerging data on t-MN risk factors, the impact of clonal hematopoiesis (CH) to t-MN, and an update on the molecular features of various t-MN clinical subtypes.

The risk of developing leukemia is 4.7-fold higher among cancer survivors than the population average, and this is most obvious in patients treated for non–Hodgkin lymphoma (NHL) and multiple myeloma (3–6). Alarmingly, over the last decade, the incidence of t-MN seems to have increased subsequent to most cancers, except for colon cancer (7, 8). For instance, the incidence of t-MN was 0.04/10,000 new cases between 2001 and 2007, which has increased to 0.20/100,000 new cases between 2008 and 2014 (incidence rate ratio, 4.87; P < 0.01; ref. 7).

Numerous factors influence the risk of t-MN, often operating in different directions. Breast cancer and lymphoma are the two most common primary diseases in patients with t-MN (Fig. 1A and B; Supplementary Table S1; refs. 3, 7, 9). Despite the lower incidence of t-MN in patients with breast cancer, ranging from 0.21% to 4.97%, the total number of t-MN cases following breast cancer is high, which is attributed to the sheer number of breast cancer cases in females and its association with long-term survival (3). Conversely, the higher incidence of t-MN in bone cancer and Hodgkin lymphoma is likely due to the multiple lines of intensive chemotherapies (CT) used.

The risk of t-MN is also influenced by the type of CT agents. Patients receiving alkylating agents and topoisomerase II inhibitors are at a higher risk than those receiving antimetabolites and taxanes (2). Additionally, certain treatments like radioiodine therapy for thyroid cancer and radiotherapy (RT) for early-stage prostate and uterine cancers are associated with three- to fivefold increased risk of t-MN (10, 11). The use of granulocyte colony–stimulating factor in patients with cancer is also linked to a higher risk of t-MN (12) and may reflect managing patients with regimes involving marrow toxicity, but a direct role for the proliferative myelopoietic cytokine cannot be excluded. Furthermore, repeated exposure to CT, particularly high dose, and autologous SCT are associated with an increased risk of t-MN. In patients with ovarian cancer, cumulative platinum exposure also demonstrates a dose-dependent relationship with t-MN (13).

Besides the type of cytotoxic therapy and cancer, the incidence of t-MN notably increases with advancing age, with reported incidence rates of t-MN of 0.09, 0.39, and 0.29 per 100,000 primary cancer diagnoses in individuals ages 40 to 59, 60 to 79, and >80 years when compared with 0.02 per 100,000 diagnosed in those between ages 20 and 39 years with the corresponding incidence rate ratios of 4.50, 19.18 and 12.89, respectively (14).

The number of solid organ transplantations is also steadily increasing, with 46,632 performed in 2023 in the United States, marking an 8.7% and 12.7% surge compared with that in 2022 and 2021, respectively. However, this upward trajectory has been met with a concerning increase in overall risk of secondary cancers, including NHL and myeloid malignancies. The estimated risk of secondary cancer is 2- to 20-fold higher in recipients of transplantation than in the general population, emerging as a leading cause of mortality after solid organ transplantation (15). Prolonged immunosuppression is likely a key contributing factor to this heightened risk.

Furthermore, 0.01% to 0.2% of cases developed donor-derived cancers, including donor-derived AML after organ transplantation. This is likely linked to the transmission of donor-derived CH into an immunosuppressed environment.

With the advent of targeted therapies and immunotherapies, and decreased reliance on the conventional cytotoxic therapies, it was hoped that the incidence of t-MN would decrease. Emerging data, however, describe specific types of noncytotoxic treatments that seem to be associated with an increased risk of t-MN.

Peptide receptor radionuclide therapy, which uses targeted radioisotopes for the treatment of inoperable gastroenteric and pancreatic neuroendocrine tumors, is associated with an increased risk of t-MN. Approximately 1.4% to 4.8% of the patients undergoing this form of treatment develop t-MN, with a median latency period of 3 to 4 years (16, 17).

Poly(ADP-ribose) polymerase inhibitors (PARPi) are effective in treating solid cancers (including breast, ovarian, and prostate) with defective DNA double-strand break repair mechanism, such as in patients with inherited BRCA1 or BRCA2 mutations. A comprehensive meta-analysis (18) confirmed a higher incidence of MDS/AML in the PARPi group (n = 5,693) than in controls (n = 3,406): 0.73% versus 0.47%, respectively, with an OR of 2.63 [95% confidence interval, 1.13–6.14; P = 0.026]. The median latency after initial exposure to PARPi was 17.8 months, which is less than 2 years on treatment (range, 8.4–29.2 months). Emerging literature demonstrates a strong association between hereditary risk factors and t-MN that is summarized in the section of “Hereditary risk factors for t-MN.”

Chimeric antigen receptor (CAR) T-cell therapies have emerged as groundbreaking therapies and, to date, have been approved for relapsed or refractory B-cell acute lymphoblastic leukemia (ALL), diffuse large B-cell NHL, mantle cell lymphoma, and multiple myeloma. Furthermore, recently there have been encouraging reports of safety and efficacy of CD19 CAR T-cell therapy in autoimmune disease cases including severe systemic lupus erythematosus (19). However, with longer follow-up and improved survivorship, an increased risk of secondary cancers, including t-MN, is emerging. For example, of 11,345 CAR T-cell recipients from the Center for International Blood and Marrow Transplant Research registry, 7.4% (n = 842) developed a second primary malignancy with a short median follow-up of 13 months (range, 0–69; ref. 20). Similarly, second primary cancers were reported in 536 (4.3%) of 12,394 adverse events reported to the FDA Adverse Event Reporting System (21). Importantly, 62% (n = 333) of these second primary malignancies (n = 536) were leukemias including MDS (n = 208), AML (n = 106), and T-cell large granular lymphocytic leukemia (n = 2; ref. 21). Furthermore, single-institute studies have also reported higher incidence (15%; ref. 22) of secondary cancers, including 5% to 6% of t-MN, after CAR T-cell infusion for lymphoma and ALL (22–24). Interestingly, the rate of t-MN was much lower among children and young adults with B-cell ALL who received CD19 CAR T-cell therapy, highlighting that factors related to non–CAR T-cell therapy may influence the development of t-MN among CAR T-cell recipients (25).

In contrast to the latency of 5 to 6 years after treatment with alkylating agents, the median latency of t-MN after CAR T-cell infusion is very short, ranging from 6 to 10 months (22–24). The short latency and relatively high incidence of t-MN after CAR T-cell therapy is striking and deserves an accelerated research effort. A number of variables could be involved in this high-risk patient group, including (i) a higher burden of CH, (ii) older age, (iii) often multiple lines of genotoxic therapies prior to CAR T-cell infusion, and (iv) further immunodepleting CT given prior to cell infusion.

The incidence of CH is 34% to 56% in patients receiving CAR T-cell therapy (26, 27), which is substantially higher than the incidence of 5% to 10% in a similarly aged healthy population, with mutations predominantly observed in PPM1D, DNMT3A, tumor factor protein 53 (TP53), and TET2. Notably, mutations in TP53 and PPM1D are not commonly seen in healthy populations but are enriched in patients that have been exposed to cytotoxic therapies. How the preexisting clone interacts with external selection pressures such as cytotoxic therapy and CAR T-cell–related inflammation is a subject of active investigation.

Considering the survival benefits offered by these approaches, additional data are needed to strike a balance between the competitive risk of death from primary malignancy and the risk of developing t-MN. Because a vast majority of patients receiving the above therapies will have received the conventional CT during the course, it is a challenge to segregate the relative impact of each therapy in t-MN development. Currently, the benefits of CAR T-cell therapies outweigh the potential risks in the vast majority of cases. However, accelerated research effort focusing on understanding pathogenesis and stratifying the risk of second primary cancers is crucial. In this context, the availability of risk stratification models, along the lines of the CAR-HEMATOTOX score for predicting prolonged cytopenia, is urgently needed. A recent single-institution analysis suggested that universally available baseline factors such as age and thrombocytopenia have stratified t-MN risk in CAR T-cell recipients into three distinct categories: older patients ( ≥65 years) with baseline thrombocytopenia (platelet count ≤140,000/μL) constituting 16% of the cohort had a t-MN rate of 27 per 100 person-years (PY). The presence of either factor was seen in 48% of the cohort and was associated with a t-MN rate of 4 per 100 PY. Reassuringly, in younger patients with intact hematopoiesis at the outset (43% of the cohort), no events after CAR T-cell MN were observed with >80 PY follow-up (24). Large international collaborative effort integrating CH of indeterminate potential, cytopenia, primary malignancies, and cytokine and chemokine profiles, along with a number of lines of prior therapies, will lead to further improvement of the current model and will guide appropriateness of CAR T-cell therapy and inform surveillance strategies.

Overall, therapy-related myeloid cancers are highly heterogeneous and unified principally by previous exposure to genotoxic therapies. The majority are t-MDS and t-AML, and also include cases of t-MDS/myeloproliferative neoplasm overlap syndromes including therapy-related chronic myelomonocytic leukemia.

Currently, t-MN makes up 5% to 15% of newly diagnosed MDS, chronic myelomonocytic leukemia, and AML (28–30). The proportion of t-AML and t-MDS in t-MN cohorts vary between studies: the Italian network (31) reported a higher proportion of t-AML (57%), whereas t-MDS (73%) predominated in Chicago series (32). Clinically, t-MN presents similar characteristics to their primary counterparts either with a cytopenic or leukemic phase.

More than 90% of t-MNs present with chromosomal abnormalities (30, 32–35). Deletion of chromosomes 5 and 7, 17p, and a complex karyotype (CK), as well as a monosomal karyotype (MK), are highly prevalent in t-MN compared with primary MN (p-MN; Fig. 2A; Supplementary Table S2A; refs. 30, 32, 33, 36). CK is detected in 30% to 40% of patients with t-MN compared with 5% to 10% of p-MN (31, 32, 37). Poor and very poor risk cytogenetic changes, according to the Revised International Prognostic Scoring System (38), are more prevalent in t-MDS than in primary MDS (p-MDS; Fig. 2B; Supplementary Table S2B). Similarly, intermediate and poor risk cytogenetic changes, as per The European LeukemiaNet recommendations, are more prevalent in t-AML than in p-AML (Fig. 2C; Supplementary Table S2C).

CKs and MKs are more frequently observed in t-MN after treatment with alkylating agents after a median latency period of 5 to 7 years. On the other hand, translocations such as t(15;17), inv(16), t(8;21), and KMT2A (MLL) rearrangements have been associated with treatment involving topoisomerase II inhibitors and a shorter latency of 1 to 3 years (reviewed in ref. 39) However, due to widespread use of combination CT, attributing specific cytogenetic changes to individual drugs or classes of drugs is increasingly challenging, although not unimportant.

Comprehensive mutation profiling has extensively covered p-MDS and p-AML in large patient cohorts (40–48). However, there is relatively limited literature available, and the reported somatic mutation frequency in t-MN varies widely (49–60), likely due to relatively small sample sizes, number of genes analyzed, and variability in bioinformatics pipelines used. We previously reported that 93% of t-MN cases harbor somatic mutations comparable with the proportion seen in p-MDS (61, 62). However, the mutation pattern was highly distinct. Comprehensive analysis of the published literature (33, 37, 40–45, 48–67) demonstrates gene-specific differences in the frequency of somatic mutations in t-MN compared with their p-MN counterparts (Fig. 3A–C; Supplementary Tables S3–S5). Unlike p-MDS, the number of mutations did not correlate with patient age at t-MN diagnosis, suggesting that treatment of the primary cancer may contribute to increased mutagenesis in t-MN.

TP53 mutation was more prevalent in t-MN than in p-MN (30.9% vs. 7.9%; P < 0.0001), t-MDS versus p-MDS (35.1% vs. 11.3%; P < 0.0001), and t-AML versus p-AML (26.2% vs. 11.3%; P < 0.00001). In contrast, somatic mutations in spliceosome complex gene SF3B1 were enriched in p-MN compared with t-MN (16.8% vs. 4.2%; P < 0.0001) and in p-MDS versus t-MDS (21.4% vs. 6.2%; P < 0.0001), but there was no significant difference in p-AML versus t-AML (0.6% vs. 1.9%; P = 0.2). Mutations in U2AF1, another spliceosome complex gene, were more frequent in p-MN versus t-MN (7.8% vs. 4.8%; P < 0.001) and in p-MDS versus t-MDS (8.7% vs. 5.0%; P < 0.01). However, the frequency of U2AF1 mutation was comparable in p-AML versus t-AML (4.4% vs. 4.5%; P = 0.6). Interestingly, the frequency of SRSF2 mutations was comparable between t-MN and p-MN (10.1% vs. 11.6%; P = 0.10) and between t-MDS and p-MDS (11.9% vs. 12.6%; P = 0.5). However, SRSF2 mutations were significantly higher in t-AML than in p-AML (9.5% vs. 0.6%; P < 0.0001; Fig. 3A–C; refs. 37, 39, 68).

Besides TP53, mutations in PPM1D, ASXL1, and RUNX1 were more prevalent in t-MDS than in p-MDS, and KRAS/NRAS and WT1 mutations were more frequent in t-AML than in p-AML (Fig. 3A–C).

The protein phosphatase Mg²⁺/^Mn2+-dependent 1D (PPM1D) gene, located on chromosome 17q, plays a crucial role as a homeostatic regulator of DNA damage response (DDR) pathways. In murine models, hematopoietic stem cells (HSC) carrying PPM1D mutations downstream to the catalytic domain (in exon 6) showed a proliferative advantage over wild-type (WT) cells during serial transplantation assays or after genotoxic (RT or cisplatin) exposure, suggesting that PPM1D regulates self-renewal capacity and proliferative capacity of HSC during the steady-state hematopoiesis and under genotoxic stress (69). PPM1D somatic mutations were enriched in t-MN compared with p-MDS (9%–20% vs. 0.5%–3%; refs. 65, 66), especially after exposure to RT, topoisomerase I and II inhibitors, and platinum-based therapy (Fig. 3A–C).

PPM1D shares a complex relationship with TP53. When cells encounter DNA-damaging stimuli, the activated TP53 transcription factor triggers PPM1D mRNA expression. Subsequently, PPM1D dampens the stress response by dephosphorylating TP53 (particularly at Ser15) and other key DDR pathway proteins. As a result, PPM1D overexpression and mutations are believed to reflect at least partial inactivation of TP53 function (70). PPM1D-activating mutations confer cellular resistance to cytotoxic therapies (69), perhaps less selective than loss of TP53 function (69). Given their shared downstream signaling pathways, PPM1D aberrations in solid cancers tend to be either mutually exclusive with TP53 mutations or occur more frequently in TP53^wt cancer cells. Single-cell sequencing studies of AML also revealed mutually exclusive clones harboring PPM1D and TP53 mutations (71). In a study of azacitidine (AZA)–shaped clonal dynamics, PPM1D-mutated clones were more prevalent in posttherapy samples, including patients with TP53-mutated clones, prior to AZA. Furthermore, single-cell sequencing suggests that TP53 and PPM1D mutations affect distinct cell populations (72).

TP53 acts as a tumor suppressor by inhibiting growth and thwarting oncogenic transformation. Considering the role of p53 as the “guardian of the genome,” it is not surprising that TP53 mutations (TP53^mut) are observed in almost every type of cancer, including MN. The majority of TP53^mut in t-MN, p-MN, and other cancers are missense mutations in the DNA-binding domain, underscoring the critical importance of this domain for the tumor suppressor function of TP53. Although multiple mechanisms can lead to the loss of the second TP53 allele, this typically occurs through the “loss of heterozygosity” by segmental deletion, reflected by an increase in the variant allele frequency of the first allele (Fig. 3D; refs. 37, 67, 68).

TP53^mut are detected in 7% to 11% of p-MDS, 10% to 13% of AML, and 23% to 37% of t-MN cases (37, 39, 63, 67, 68, 73). The enrichment of TP53^mut in t-MN is probably driven by the selective advantage conferred to preexisting TP53-mutated clones under cytotoxic stress (74, 75). Indeed, TP53^mut clones are identified in the blood years before the development of MN (63). However, not all TP53^mut clones evolve into MN. In a recent study, only 10% of clonal cytopenia of uncertain significance harboring TP53^mut progressed to MN; however, the median follow-up period was relatively short (76). Thus, other cell-intrinsic and -extrinsic factors drive the progression of TP53^mut clones. Potential contenders promoting/selecting the clone leading to leukemic transformation include cytotoxic therapy (63) and inflammation. For example, a recent single-cell multiomics study demonstrated that a chronic inflammatory microenvironment confers a fitness advantage and promotes evolution of TP53^mut cells while suppressing TP53^wt hematopoietic stem/progenitor cells (HSPC; ref. 77).

TP53^mut MN exhibits characteristic genomic instability, including CK and MK, which are associated with adverse outcomes (37, 67, 68, 73, 78). TP53^mut are highly prevalent in cases with CK as compared with cases with normal karyotype and two chromosomal abnormalities (76.8% vs. 4.5% vs. 17.3%; P < 0.0001). In patients with CK, the frequency of TP53^mut increases with the number of chromosomal abnormalities. For example, TP53^mut were observed in 75%, 96.6%, and 94% of cases with 4 to 6, 7 to 9, and >9 chromosomal abnormalities, respectively, compared with 26.3% cases with three chromosomal abnormalities (37). This association is even more pronounced in t-MN versus p-MDS (37). Several biological explanations have been proposed for this association including defective DDR and repair of DNA breaks. Loss of TP53 function disrupts the spindle assembly checkpoint, leading to an increased rate of chromosome missegregation and tetraploidization. Polo-like kinase 4 (PLK4), a master regulator of centriole duplication and crucial for cell-cycle progression, mitosis, and cytokinesis (79), is compromised in TP53^mut cancers. PLK4 expression is suppressed by activated p53 signaling in TP53^wt MN, and high PLK4 expression in TP53^mut cancers is linked to increased centriole numbers, centrosome amplification, chromosomal instability, and aneuploidy (80), underscoring the pathogenic role of PLK4 in TP53^mut cancers. Aberrant expression of PLK4 can be targeted by PLK4 inhibitors, with encouraging preclinical and phase I clinical results (81).

Therapy-related MN was first recognized in the late 1960s. After three decades of first reports, the World Health Organization (WHO) formally incorporated t-MN as a separate subcategory in 2002 (82), which was maintained in subsequent classifications. However, the recent update of the WHO classification (fifth edition; WHO5; ref. 83) and International Consensus Classification (ICC) in 2022 (84) removed t-MN as a distinct entity. Although these two systems have differences, they share a common focus: increased emphasis on genetic changes driving disease pathogenesis, prognosis, and response to therapy. There has been a deliberate shift away from placing significant importance on morphologic factors, such as BM blast count and the presence of dysplasia, as well as clinical features like a history of MDS or exposure to DNA-damaging cytotoxic therapies before the diagnosis of MN.

In this evolution, both classification systems have reduced the significance of t-MN. WHO5 has integrated this category with secondary MN, whereas the ICC goes further by eliminating t-MN as a distinct subgroup and instead suggests the addition of a “diagnostic qualifier.” This shift in classification frameworks reflects a broader acknowledgment of the role of genetic factors in understanding and categorizing MN. However, this evolution raises the possibility of underestimating the poor prognosis of t-MN compared with p-MN with similar genetic changes.

WHO5 includes biallelic TP53^mut MDS but excludes monoallelic TP53^mut MDS (0%–19% blasts) from the newly defined distinct category of “TP53-mutated MDS.” Similarly, ICC excludes single-hit TP53^mut MDS (0%–9% blasts) from the “TP53-mutated myeloid neoplasm” group, as survival of single-hit TP53^mut MDS was considered comparable to TP53^wt MDS. However, in contrast to p-MDS, in which monoallelic TP53^mut MDS had survival comparable to TP53^wt MDS (67), monoallelic TP53^mut t-MDS was associated with poor survival comparable to biallelic TP53^mut MDS (37, 68). Poor survival of monoallelic TP53-mutated t-MDS is probably driven by enrichment of CK (37, 68); however, WHO5 does not recognize CK itself to be biallelic equivalent (83), whereas ICC considers it as multihit equivalent (84).

Collectively, both the classification systems underestimate the poor prognosis of monoallelic TP53^mut t-MDS by excluding these cases from TP53^mut MDS. Furthermore, prior therapy and TP53^mut were independently associated with inferior survival in MDS with isolated deletion 5q (5-year survival 27% vs. 69%), again highlighting the impact of prior cytotoxic therapies in genetically defined MN (85).

AML with specific chromosomal abnormalities such as t(8;21) (q22;22) and inv(16) (p13.1q22)/t(16;16) (p13.1;q22) result in fusion genes RUNX1::RUNX1T1 and CBFB::MYH11, respectively; although the majority of core-binding factor (CBF) AML cases are primary, 5% to 25% of AML with t(8;21) and 3% to 15% of AML with inv(16) are t-AML (86). Despite receiving similar therapy, the median overall survival (OS) for CBF t-AML was shorter than its primary counterpart (69 vs. 190 months; P = 0.038; ref. 86).

The prevalence of Fms-like tyrosine kinase 3 (FLT3) internal tandem duplication (12% vs. 24%), but not tyrosine kinase domain mutation (9% vs. 8%), was found to be lower in t-AML than in p-AML, but FLT3 internal tandem duplication was not associated with poorer survival in t-AML (87). In another study, after accounting for cytogenetic and molecular features, prior therapy remained an independent factor associated with shorter survival in patients >60 years (30).

Nucleophosmin 1 mutations (NPM1^mut) are prevalent in adults with AML, occurring in 30% to 35% of AML and 14% to 16% of t-AML (64, 88). Importantly, t-AML and p-AML with NPM1^mut share common disease characteristics, such as aberrant localization of nucleophosmin in the cytoplasm, normal karyotype, upregulation of HOX genes, co-mutation of DNMT3A and TET2, and downregulation of CD33 and CD34 expression. TP53 and PPM1D mutations, as well as chromosomal abnormalities, are less frequently observed in NPM1^mut t-AML. Importantly, the 3-year OS was found to be comparable in NPM1^mut t-AML and p-AML (54% vs. 60%; ref. 88).

Similarly, the prognosis of t-MDS with ring sideroblasts is poorer than that of p-MDS with ring sideroblasts (61). Furthermore, the strong association of SF3B1 mutation with ring sideroblasts observed in p-MDS was not reproduced in t-MDS with ring sideroblasts. Striking enrichment of TP53^mut was observed in SF3B1 WT MDS with ring sideroblasts (61, 89).

The prognostic scoring system developed for classifying patients with p-MDS performs poorly for t-MDS, and survival for t-MDS is poorer than that for p-MDS within each risk category of the Revised International Prognostic Scoring System. Second, the prognostic power of these scoring systems varies according to the type of primary cancer and treatment for the primary cancer (90). Collectively, these findings emphasize the need for careful consideration of clinical outcomes when applying these updated classification systems, as they have significant implications on patient management, such as consideration for allogeneic transplantation and clinical trials.

The prevailing hypothesis is that t-MNs are HSPC autonomous disorders, wherein cytotoxic therapies induce genome-wide DNA damage in HSPCs that drive the initiation and progression to t-MN. Evidence supporting this hypothesis includes the association between topoisomerase II inhibitor exposure and MLL rearranged t-MN (91), high frequency of CK, and deletion of chromosomes 5 and/or 7 in t-MN after exposure to alkylating agents and RT. Furthermore, somatic mutations in DDR genes, such as PPM1D, TP53, and CHEK2, are frequently observed in the peripheral blood of patients with ovarian, endometrial (92), or other solid cancers, who have undergone RT or CT with platinum compounds or topoisomerase inhibitors compared with those who have had surgery, immunotherapy, or targeted therapy. Moreover, longitudinal studies indicate that DDR-mutated clones are generally expanded in 61% of patients who receive cytotoxic CT or RT (92).

Only a small number of patients treated with CT will develop t-MN, suggesting the involvement of other factors such as genetic predisposition, BM microenvironment, age, and preexisting CH in t-MN pathogenesis (39, 63, 93–101). The prevalence of CH in blood samples of patients with cancer is 27% to 33% compared with 10% in the general population of a comparable age group (93–97). Strikingly, the prevalence of CH was reported to be 61% to 71% in t-MN compared with 27% to 31% in patients with cancer who did not (98). Moreover, individuals with mutations in leukemia driver genes at a higher variant allele frequency were associated with ∼10-fold higher risk of progression to hematologic malignancy than those without these mutations (3.2% vs. 0.3%; P < 0.001) during a short follow-up period (93).

In addition to a higher frequency, the architecture of CH differs between patients with cancer who develop t-MN and those who do not. TP53, PPM1D, RUNX1, SRSF2, and TET2 are more frequently mutated in patients with cancer who develop t-MN than in those who do not (63, 65, 98, 99). Additionally, driver clones detected at the time of primary cancer diagnosis expand by the time of t-MN diagnosis (98). Similarly, 11% of patients with multiple myeloma display MDS-associated phenotypic abnormalities at diagnosis, with 50% of these patients having CH, which is associated with poor OS and progression-free survival (100).

In addition to somatic mutations, the mosaic chromosomal alterations, encompassing deletions, copy-neutral loss of heterozygosity, and amplifications, were observed in patients with hematologic and nonhematologic cancers, and co-occurrence of chromosomal alterations and somatic mutations further amplifies the risk of t-MN (102). Thus, emerging literature raises a fundamental question about the role of CT and RT in t-MN pathogenesis. Detection of CH at the time of cancer diagnosis suggests that driver mutations preexist in patients with cancer before exposure to CT, and DNA-damaging genotoxic therapies provide a selective advantage and facilitate clonal expansion of preexisting clones (39, 63), such as TP53- and PPM1D-mutated clones.

Hereditary risk factors also contribute to the pathogenesis of t-MN. A population-based analysis showed a higher risk of secondary acute leukemia in patients with breast cancer with a family history of breast or ovarian cancer (103). Familial and/or personal history of multiple neoplasms is present in 5% to 10% of patients with cancer. Furthermore, 10% to 15% of MN cases occur as second neoplasm in individuals treated solely with surgery for primary tumor.

With advances in gene sequencing platforms and improved awareness, the list of well-defined cancer-predisposing genes is increasing. Approximately 5% to 10% of pediatric, adolescent, and adult patients with cancer (including some hematologic malignancies) harbor pathogenic germline variants in cancer-predisposing genes (104, 105). Emerging literature, including our own research, suggests that 9% to 31% of t-MN cases harbor germline mutations, predominantly in the DDR pathway (11, 106–112), including CHEK2, BRCA1, TP53, and DDX41 (Fig. 4A and B; Supplementary Table S6). Collectively, these findings suggest a complex interaction of inherently increased susceptibility of the host with extreme and repetitive selection pressure by cytotoxic agents that culminates in t-MN development.

Identification of genetic predisposition has significant implications for patient management and can also guide surveillance recommendations for affected family members. The WHO has recognized the importance of genetic predisposition, incorporating MN with genetic predisposition as a distinct entity in the classification (113). In addition, recent update of The European LeukemiaNet guidelines recommends germline testing for individuals with a personal history of two or more malignancies, one of them being a hematologic malignancy (114).

BM fibrosis and an inflammatory BM microenvironment are well-known features of MN and are associated with poor prognosis. However, the current debate revolves around whether these manifestations are the cause or effect of malignant clone. Malignant clones can alter the BM microenvironment to favor their survival and expansion while inhibiting normal HSC (115). Conversely, recent studies have suggested that an aberrant BM microenvironment can also contribute to the development of MN. For example, the loss of function of genes, such as RAR-γ, SIPA1, SBDS, and DICER1, in nonhematopoietic cells has been shown to lead to the development of MN in murine models (116–118). Deletion of DICER1 in mouse osteoprogenitors disrupts hematopoiesis, leading to cytopenia, dysplasia, and progression to AML (118). Furthermore, downregulation of DICER1 and SBDS expression was reported in BM mesenchymal stromal cells (BM-MSC) isolated from patients with MDS, suggesting that impaired osteogenic differentiation of BM-MSCs also contributes to MDS pathogenesis in humans (119).

MSC are key regulators of the BM microenvironment and regulate HSPC retention, quiescence, and multilineage potential through direct cellular interactions and secretion of regulatory factors such as CXCL12, TGFβ, ANGPT1, and TPO (120). Cytotoxic therapies, whether short-term in vitro (121) or in vivo, have been shown to affect BM-MSCs by compromising their proliferative (121, 122), differentiation (121), and DDR abilities, in addition to inducing senescence and compromising HSPC support capacity. However, this is not universally agreed upon, as other studies have reported normal proliferative capacity, the ability to support HSPCs (122), and normal differentiation capacity of BM-MSCs after CT. Cytotoxic therapies induce long-term damage in BM-MSCs and induce senescence, which, in turn, leads to a highly proinflammatory microenvironment driven by paracrine and endocrine mechanisms mediated by senescence-associated secretory phenotype (SASP) in a t-MN murine model (123) and primary patient samples (124). BM-MSCs isolated from patients with t-MN were highly aberrant, lost spindle-shape morphology, and were significantly larger than BM-MSC from p-MN and healthy age-matched individuals. Importantly, t-MN MSC exhibited a high degree of senescence, defective proliferative and differentiation capacity, and increased secretion of highly proinflammatory SASP, which do not support normal hematopoiesis (124). This highly proinflammatory microenvironment provides a fitness advantage to clones with certain mutations, such as TP53 (77, 123).

Timely elimination of senescent cells within the tumor microenvironment may hold the potential to alleviate various short- and long-term effects of CT, including BM suppression. Senolytics, drugs that eliminate senescent cells and are capable of restoring cellular functionality (124, 125), are recognized for selectively targeting prosurvival mechanisms in senescent cells while sparing nonsenescent cells (125). A combination of dasatinib and quercetin, two of the most studied senolytics, has been shown to selectively eliminate senescent cells (124, 125). Furthermore, sufficient restoration of cellular dysfunction may be achieved without the need to eliminate all senescent cells (124, 125).

In summary, emerging literature, including our own data, suggests that genotoxic therapies used to treat primary cancers can induce DNA damage in BM-MSCs. Cells with defective DDR capacity that fail to undergo apoptosis become senescent. Although senescence can be protective in the short term, the persistence of senescent cells leads to a highly proinflammatory microenvironment conducive to the progression of mutated clones and progression to t-MN. Senolytics can target senescence and potentially reduce the risk of t-MN.

The lack of meaningful improvement in survival over decades (22.1% OS for 2001–2007 compared with 26.9% OS for 2008–2014; ref. 14) is the Achilles’ heel of the field. Virtually, all (∼95%) current trials exclude patients with t-MN (126), either explicitly or by excluding those with recent cancers. Hence, treatments used for p-MDS/AML are also used for t-MN. However, patients with t-MN have poor survival than those with p-MDS/AML (127). Thus, experience with novel therapies accumulates only after regulatory approval, leading to the availability of small, retrospective, typically single-center experiences.

Allogeneic SCT (allo-SCT) is the only potentially curative therapy for t-MN. However, only 17% to 29% of patients with t-MN receive allo-SCT. Survival after allo-SCT for t-MN remains poorer than that for p-MDS and p-AML (31, 128–132). A study by the Center for International Blood & Marrow Transplant Research reported a 5-year OS of 27% for t-MDS and 25% for t-AML (133), and the 1-year (24% vs. 25%) and 5-year (34% for both) nonrelapse mortality rates in patients with t-MDS and t-AML were comparable (Fig. 5; Supplementary Table S7). Similarly, relapse rates between patients with t-MDS and t-AML at 1 year (39% vs. 37%) and 5 years (46% vs. 43%) after allo-SCT were similar. Furthermore, there was no significant improvement in 5-year OS in patients with t-MN transplanted during 1990 to 2004 and 2000 to 2014, although transplantation-related mortality has improved over time (31, 134).

Factors such as adverse cytogenetics, high-risk disease, previous autologous SCT, persistent disease at the time of allo-SCT, older age, and mismatched unrelated donor all predict poor OS and disease-free survival (135). The failure to achieve full chimerism and the absence of chronic GVHD predict an increased risk of relapse.

Establishing an optimal pretransplantation algorithm for significant disease reduction without compromising the patient’s fitness is crucial. Newer therapies, such as CPX351, show promise in increasing the feasibility and outcomes after allo-SCT in patients with t-AML (136, 137). Three-year OS after allo-SCT was significantly higher in patients treated with CPX351 than in those treated with standard 7 + 3 induction (56% vs. 23%).

In addition to bridging therapy and appropriate conditioning therapy, optimal donor selection is crucial for improving transplantation outcomes. We and others have shown that 9% to 33% of t-MN cases harbor pathogenic germline variants (63, 106–110). Identification of genetic predisposition has significant implications for patient management, including donor selection for allo-SCT and for the selection of appropriate conditioning regimens. Avoiding HSPCs from an asymptomatic related donor with a pathogenic germline mutation can help prevent adverse outcomes such as donor-derived leukemia (138), graft failure, and prolonged unexplained cytopenia due to poor graft function. Special consideration is necessary when selecting conditioning regimens for patients with short telomere syndromes and Fanconi anemia, as these patients are at an increased risk of severe toxicities from treatment involving busulfan, radiation, and alkylating agents (139). Furthermore, pathogenic germline mutation in DDX41 is associated with a higher incidence of severe (grades 3–4) acute GVHD, even when transplantation was performed using WT donors (140). Importantly, after transplantation, cyclophosphamide reduced the risk of severe acute GVHD in patients receiving allo-SCT for DDX41-associated MN (0% vs. 53%; P = 0.03).

Following intensive CT, the median OS of t-AML was poor compared with that of p-AML (30). The median OS after intensive CT was significantly poorer in younger patients with t-AML than in patients with p-AML ages <55 years (14 vs. 158 months) and 55 to 74 years (16 vs. 9 months), whereas OS was poor in both patients with p-AML and those with t-AML ages >75 years (8 vs. 7 months; ref. 141). The unfavorable outcome in t-AML could be due to the high burden of poor-risk disease features associated with a poor response rate and high 90-day mortality (21% vs. 15%) in t-AML compared with that in p-AML.

Although treatment outcomes for the majority of patients with t-MN remain suboptimal compared with their primary counterparts, the remission and survival outcomes were similar for therapy-related acute promyelocytic leukemia and primary acute promyelocytic leukemia in both anthracycline-based and non–anthracycline-based (ATRA + ATO) regimens (142). The OS of CBF t-AML (n = 69) was poorer than it was in their p-AML (n = 282) counterparts treated with similar therapies (69 vs. 190 months, P = 0.038; ref. 86).

CPX351 (dual-drug liposomal encapsulation of daunorubicin and cytarabine combined in a 1:5 mol/L ratio, respectively) has shown an improved overall remission (47.7% vs. 33.3%; P = 0.016) and median survival (9.56 vs. 5.95 months; P = 0.003) and similar safety to a traditional intensive chemotherapy regimen such as 7 + 3 in patients with t-AML ages ≥60 years and is approved by the FDA based on a randomized phase III trial.

Hypomethylating agents (HMA) have long been considered a cornerstone for the treatment of MDS; however, patients with t-MDS were not included in the pivotal AZA001 clinical trial (143). Furthermore, published experience is largely based on retrospective analysis of relatively small numbers of t-MN cases (127, 144), and the median number of completed HMA cycles ranges from three to six cycles (127, 144), lower than that reported for p-MDS (143, 145). Although the overall response rate of 38% to 43%, including a complete remission rate of 11% to 42%, is similar in t-MDS and p-MDS cases, the OS is poor in t-MN, with a median OS of 9.2 to 21 months (127, 144). The median OS of t-MN was significantly poor compared with that of p-MN (10.4 vs. 18.1 months; P = 0.0069) after HMA therapy (our unpublished data).

In a prospective trial of AZA with or without histone deacetylase inhibitor entinostat in t-MN, the median OS was significantly poorer in the combination arm than that for AZA alone (6 vs. 13 months; P = 0.008; ref. 146). A 10-day decitabine regimen showed an improved response in patients with AML/MDS with unfavorable risk cytogenetic profile and TP53^mut, both of which are enriched in patients with t-MN (147). However, this could not be validated in a prospective randomized trial of decitabine given on either a 5- or 10-day schedule (148).

Venetoclax to AZA is the new standard of care for older patients with AML; however, only 8% of cases in the landmark trial were with t-AML (149). In a retrospective study, the composite response rate with venetoclax-based therapy in t-MN was 39.1%, which is lower than that in p-MN (54%–66%; ref. 150). Despite the initial response, the majority of patients progressed, with a median progression-free survival and OS of 4.9 and 7 months, respectively.

Given the high frequency of TP53^mut in t-AML and the associated poor prognosis, there is considerable interest in identifying novel agents targeting the TP53 pathway. However, despite encouraging preclinical and early clinical trials, phase III/II trials of eprenetapopt (APR246), humanized IgG4 mAb-targeting CD47 (magrolimab; ref. 151), and T-cell immunoglobulin and mucin domain–containing protein 3 (152) failed to demonstrate efficacy. Emerging preclinical studies suggest TP53^mut AML may be particularly susceptible to other arms of the innate immune signaling, including STING agonists (153).

Recent studies have elucidated that t-MN pathogenesis arises from complex interactions between HSC-intrinsic and -extrinsic factors (Fig. 6A). Within HSCs, the emerging role of inherited variants, especially those in DDR genes, is reshaping our understanding of the contribution of DNA-damaging therapy to t-MN pathogenesis. Therapies can shape the clonal architecture by providing preferential survival advantage. Moreover, the proliferative advantage of the clone may be further influenced by senescent BM-MSCs and proinflammatory SASP while being regulated by the functional immune system. Consequently, a combination of factors may account for the relative rarity of t-MN despite the widespread use of cytotoxic therapies. The association of novel agents, beyond traditional cytotoxic agents, with t-MN development offers an opportunity to advance our understanding of the factors that lead to t-MN.

Furthermore, the disparate recommendations provided by the two classification systems emphasize the lack of consensus among experts about the relative contributions of genetics/genomics and prior therapies to t-MN. Addressing these complex questions will require the analyses of large, fully annotated databases that account for the rapidly changing landscape of MN.

Ultimately, these efforts aim to pave the way for personalized therapies that consider the wide heterogeneity within the clinical entity of t-MN (Fig. 6B). An important strategy is the prevention of t-MN by avoiding or minimizing genotoxic therapies in patients at higher risk of developing t-MN. This can be achieved by developing a risk model that integrates hereditary and acquired genetic factors along with environmental factors to predict to the risk of t-MN versus survival benefit after cytotoxic therapy.

C.N. Hahn reports grants from the National Health and Medical Research Council during the conduct of the study. M.V. Shah reports grants from AbbVie, KURA Oncology, Astellas, MRKR Therapeutics, and Celgene outside the submitted work. No disclosures were reported by the other authors.

D. Singhal and M.M. Kutyna: Literature search, data extraction, data analysis, writing–original draft. C.N. Hahn and M.V. Shah: Writing–review and editing. D.K. Hiwase: Conceptualization, project supervision and analysis, writing–original draft. All authors read and approved the final manuscript.

This work is supported by the Medical Research Future Fund—Early to Mid-Career Researchers Grant Opportunity (#2030689) by the National Health and Medical Research Council (to M.M. Kutyna); Mayo Clinic Research Pipeline K2R Transition Award and Bridget Kiely Clinician Career Development in Transplant Research and Mayo Clinic (to M.V. Shah); National Health and Medical Research Council/Medical Research Future Fund Investigator grant MRF1195517, Cancer Australia grant #2013617, and Leukaemia Foundation Australia grant #22007 (to D.K. Hiwase). The authors thank Drs. Daniel Thomas and Carla Toop for editing the manuscript and providing feedback.