Combination strategies to promote sensitivity to cytarabine-induced replication stress in acute myeloid leukemia with and without DNMT3A mutations

Abstract

Cytarabine and other chain-terminating nucleoside analogs that damage replication forks in rapidly proliferating cells are a cornerstone of leukemia chemotherapy, yet the outcomes remain unsatisfactory due to resistance and toxicity. Better understanding of DNA damage repair and downstream effector mechanisms in different disease subtypes can guide combination strategies that sensitize leukemia cells to cytarabine without increasing side effects. We have previously found that mutations in DNMT3A, one of the most commonly mutated genes in acute myeloid leukemia and associated with poor prognosis, predisposed cells to DNA damage and cell killing by cytarabine, cladribine, and other nucleoside analogs, which coincided with PARP1 dysfunction and DNA repair defect (Venugopal et al, 2021). In this perspective piece, we first overview DNA repair mechanisms that remove aberrant chain-terminating nucleotides as determinants of sensitivity or resistance to cytarabine and other nucleoside analogs. Next, we discuss PARP inhibition as a rational strategy to increase cytarabine efficacy in cells without DNMT3A mutations, while considering the implications of PARP inhibitor resistance for promoting clonal hematopoiesis. Finally, we examine the utility of p53 potentiators to boost leukemia cell killing by cytarabine in the context of mutant DNMT3A. Systematic profiling of DNA damage repair proficiency has the potential to uncover subtype-specific therapeutic dependencies in AML.

Graphical Abstract

Acute myeloid leukemia (AML) is an aggressive malignancy of the blood system and the most common acute leukemia in adults (2,3). Improvements in chemotherapy combined with a limited number of targeted approaches thanks to better understanding of the disease mechanisms significantly extended survival in younger patients. Yet, the outcomes in advanced-age AML patients (65 and older, constituting more than half of all cases) remain dismal. Although studies demonstrated survival benefit of induction chemotherapy dose intensification (4,5), most patients are unable to tolerate aggressive treatment due to comorbidities and frailty (3,6). A recent clinical study evaluating a novel cladribine- and cytarabine-based low-intensity regimen pointed towards a potential survival benefit specifically in patients with DNMT3A mutations (7), which motivated our effort to identify a molecular mechanism underpinning this sensitivity (1).

Mutations in the DNA methyltransferase 3A (DNMT3A) gene are recurrent in de novo AML (20–35%) and are associated with poor prognosis due to relative resistance to anthracycline-based induction regimens, high rates of minimal residual disease (MRD) positivity, and increased risk of relapse (4,5,8–14). While in myelodysplastic syndromes (MDS) and in clonal hematopoiesis (CH) most DNMT3A mutations are consistent with a loss-of-function (LOF, represented by nonsense, splice, and frameshift alterations), in AML its mutational profile is dominated by hot-spot mutations at arginine 882 (R882, 50% and up to 75% of all DNMT3A mutations, with about a half of remaining mutations being LOF), which leads to decreased enzymatic activity and may have additional functions and confer differential therapeutic response (15–18). Although dose-dense anthracyclines can overcome resistance and improve outcomes in younger patients (<65 years of age), the survival advantage is lost in older patients due to unacceptable treatment-related mortality (3–5). This prompted exploration of lower-intensity approaches that often combine multiple nucleoside analogs such as cytarabine and cladribine (7,19,20), which work by inducing unrestrained replication stress while overwhelming DNA damage repair mechanisms.

The success of personalized cancer therapy depends on detailed understanding of the molecular mechanisms driving each disease subtype and treatment response. Today, cytarabine (ara-C) and similar replication chain-terminating nucleoside analogs remain a core of AML treatment, whether alone or in combination (21), with low-dose ara-C being a commonly-used regimen in advanced-age patients (3,22,23). However, combinatorial strategies to further increase cytarabine effect and/or decrease toxicity have not been investigated. Development of such approaches that synergize with this treatment strategy requires identification of tractable molecular mechanisms that execute cytarabine-mediated cell killing. We have zeroed in on impaired recruitment of poly(ADP-ribose) polymerase 1 (PARP1) to damaged DNA, which is necessary for DNA repair initiation as key to cytarabine efficacy in DNMT3A-mutant AML (1). The resultant DNA repair defect and ensuing accumulation of DNA damage triggers activation of the p53 pathway, which is further unleashed by p53-potentiating pharmacologics (1). Here, we outline DNA repair pathways responsible for the removal of chain-terminating aberrant nucleotides, discuss pharmacological inhibition of PARP as a mechanistically-informed strategy to augment cytarabine sensitivity in AML with wild-type DNMT3A, while weighing up the utility of p53 potentiators as a means to boost cytarabine efficacy in the presence of mutant DNMT3A. These combinations potentially offer lower toxicity approaches suitable for older/unfit patients, a subpopulation with limited treatment options. We also examine the link between PARP inhibitor therapy and emergence of clonal hematopoiesis (CH) in the context of DNMT3A mutations.

Mechanisms of DNA repair after cytarabine-induced damage: PARP1 dysfunction in cells expressing mutant DNMT3A

Cytarabine (ara-C) has been the cornerstone of AML treatment since the 1960-s (24–26). The main antineoplastic mechanism of action of cytarabine and similar pyrimidine and purine nucleoside analogs gemcitabine, cladribine, and fludarabine is through incorporation of these non-extendable antimetabolites into the nascent DNA during DNA replication. As a result, these drugs interfere with DNA synthesis and inhibit replicative DNA polymerases, leading to replication fork stalling, strand breaks, and ultimately cell death (27–29). Genetic variation in the components of nucleoside analog cellular uptake, activation, and metabolism has been implicated in regulating intracellular drug concentrations (24), prompting development of dose-intensified regimens. Unfortunately, even with high-dose induction, a substantial subset of patients fail to respond, and many more experience relapse (3,22,25,26), suggesting additional mechanisms of primary resistance. Intensification of DNA repair capacity is appreciated as a key resistance mechanism to DNA damaging chemotherapy, yet the evidence of its contribution to the resolution of cytarabine-induced lesions is fragmented. More importantly, unlike repair of the damaged template strand, which has been comprehensively studied (29,30), the mechanisms of removal of aberrant chain-terminating nucleotides from the nascent strand during DNA replication remain speculative.

Incorporation of cytarabine or similar chain-terminating nucleoside analogs into nascent DNA effectively results in generation of single-stand breaks (SSBs) and activation of the ATR-CHK1-mediated intra-S-phase checkpoint; the cell attempts to repair the blockage and restore functional replication forks (Figure 1A, steps 1 and 2). One potential mechanism mediating removal of aberrant nucleotides is nucleotide excision repair (NER) coupled with repriming (29). In support of this, a small clinical study found that single-nucleotide polymorphisms in NER pathway genes XPC and XPD were associated with differences in overall survival in AML patients treated with standard cytarabine-based induction chemotherapy (31). Alternatively, the fork may undergo regression forming a “chicken foot” structure, followed by resection and partial fork degradation by exonucleases, most notably MRE11 and EXO1 (29). A critical role for MRE11 3’-to-5’ exonuclease activity in removing gemcitabine blockage from stalled replication forks was recently demonstrated in several eukaryotic systems (32). Should the rate of DNA damage repair provided by these mechanisms be inadequate (both of which are PARP-dependent), then SSBs are converted to double-strand breaks (DSBs) through fork scission and breakage, activating the ATM/CHK2 signaling node (Figure 1A, step 3). DNA ends are then processed in preparation for strand invasion mediated by RAD51 and homologous recombination repair (HR). Indeed, DSBs induced by another anti-leukemia nucleoside analog CNDAC were shown to be repaired exclusively through HR rather than non-homologous end joining (NHEJ) (33). Finally, failure to repair DSBs is highly lethal to cells through p53-dependent apoptosis or mitotic catastrophe (Figure 1A, step 4).

Figure 1. Cells with mutant DNMT3A exhibit specific DNA damage repair defects after cytarabine treatment.

(A) DNA damage pathways involved in the repair of cytarabine-induced replication termination and fork stalling. Cytarabine and similar nucleoside analogs such as fludarabine, cladribine, and gemcitabine, are incorporated into nascent DNA leading to chain termination (1). Growing stretches of single-stranded DNA are bound by RPA, which triggers assembly and activation of the ATR/CHK1-dependent checkpoint (2) to facilitate repair mechanism(s), such as i) repriming and removal of the non-extendable nucleotides by nucleotide excision repair (NER), or ii) fork reversal and partial degradation by exonucleases EXO1 and/or MRE11. If single-stand break (SSB) repair is unsuccessful, fork breakage ensues (3). Resultant DNA double-stranded breaks (DSBs) activate ATM/CHK2-dependent checkpoint and DNA damage repair through homologous recombination (homologous recombination repair, HR). Excessive levels of DNA damage lead to activation of p53-dependent apoptosis (4). (B) Cells expressing mutant DNMT3A are proficient in engaging DSB repair pathways. U2OS cells lentivirally expressing wild-type and R882C mutant forms of DNMT3A, or empty vector control, demonstrate equal ability to form RAD51 (marker of HR) and 53BP1 (marker of NHEJ) foci after 24 hours of treatment with 10μM cytarabine, visualized by immunofluorescent staining. (C) Expression of mutant DNMT3A attenuates recruitment of XPA, a key player in the NER pathway, and of MRE11 exonuclease involved in processing of reversed forks, to the chromatin from soluble nuclear fraction after 12 hours of treatment with 10μM cytarabine, compared to cells overexpressing wild-type DNMT3A or empty vector control. All experimental procedures for (B) and (C) were as described

We have previously found accentuated sensitivity to cytarabine in cells expressing R882-mutant DNMT3A in vitro and in vivo, wherein high levels of replication stress and DNA damage markers coincided with impaired recruitment of PARP1 to chromatin and decreased DNA repair capacity (1). However, the specific DNA repair pathway differentially inactivated by expression of mutant DNMT3A was not identified. Examination of the DSB repair pathways showed that cells with mutant DNMT3A were proficient in engaging both HR and NHEJ as seen by efficient RAD51 and 53BP1 foci formation after cytarabine treatment compared to controls (Figure 1B). In contrast, chromatin recruitment of a critical NER pathway component XPA was strongly promoted by wild-type but not mutant DNMT3A, while chromatin binding of MRE11 was abrogated in the presence of DNMT3A^mut (Figure 1C). Thus, MRE11 is a double-edged sword which has been reported to excessively degrade replication forks in the context of HR deficiency such as in Fanconi Anemia (FA) or in cells with BRCA1/2 deficiency (34), yet plays a protective role in clearing up replication fork blockage (32). These observations point to impaired removal of chain-terminating nucleotides from stalled replication forks in DNMT3A^mut cells through PARP1 dysfunction. As a result, DNMT3A-mutant cells become disproportionately dependent on DSB repair mechanisms, which, when overwhelmed, constitute a therapeutic vulnerability. Further, this mechanism aligns with the favorable response to a nucleoside analog combination regimen consisting of cladribine and low-dose cytarabine alternating with decitabine observed in a group of elderly AML patients with DNMT3A-mutated disease (7). At the same time, cells expressing mutant DNMT3A are insensitive to PARP inhibitors.

Mutational profiles illuminate determinants of PARP inhibition sensitivity and resistance in myeloid malignancies

Since the FDA approval of olaparib in 2014, PARP inhibitors (PARPi) have gained a prominent role in the treatment of breast, ovarian, and prostate cancers with BRCA1 and 2 mutations that disrupt HR pathway. Later, clinical benefit of olaparib and other PARPis rucaparib, niraparib, and talazoparib was shown in tumors with HR deficiency yet without BRCA1/2 lesions, with clinical trials in multiple malignancies and more agents ongoing (35). Recruitment of PARP1 to sites of DNA damage leads to its activation and autoPARylation, which acts as a scaffold for other DNA repair proteins to bind and initiate DNA repair (36). Because of their mechanism of action that mainly relies on disrupting various pathways of SSB DNA repair (Figure 1A) and direct trapping of PARP1 on DNA lesions (37), PARP inhibitors have shown promise as a single agent and in combination with DNA damaging chemotherapy, beyond being synthetic lethal with BRCA1/2 mutations and other HR deficiencies (35,38).

Several types of DNA damage are associated with replication stress. Replication fork stalling can be attributed to DNA-protein crosslinks forming bulky adducts, abnormal DNA secondary structures or topology, RNA-DNA hybrids (R-loops), or in rapidly proliferating cancer cells, scarcity of histones or deoxyribonucleotide triphosphates. Further, impediments to fork progression lead to uncoupling of DNA polymerase from helicase generating extended stretches of single-stranded DNA (ssDNA). If not adequately protected by RPA in case of depletion of RPA pool, this results in scission of naked ssDNA and formation of SSBs (39). Since repair of these DNA lesions is PARP1-dependent, it presents a therapeutic opportunity for the use of PARPis.

In AML and MDS specifically, current efforts are directed at identification of mutationally-defined disease subgroups that may benefit from addition of PARPis as part of their management (38). To this end, a tour-de-force study by Tothova et al. identified PARP1-mediated DNA damage repair mechanisms as a dependency in STAG2-mutant cells critical for balancing out elevated replication stress experienced by these cells. As a result, AML and MDS with cohesin mutations were uniquely sensitive to PARP1i talazoparib (40). Similarly, accumulation of R-loops induced by spliceosome mutations was a source of preferential sensitivity to PARP inhibitors in myeloid malignancies (41). Earlier studies have shown that, akin to solid tumors, myeloid malignancies with HR deficiencies may be more sensitive to PARP inhibition (42). For example, olaparib was effective against AMLs driven by AML1-ETO and PML-RARα as these fusion transcription factors repressed key HR genes (43). Neo-oncometabolite 2-hydroxyglutarate (2-HG) produced by mutant IDH1 and 2 led to increased DNA damage, which was further accentuated by PARPi. Yet, this therapeutic sensitivity was extinguished by pharmacologic inhibition of mutant IDH (44). Conversely, JAK2(V617F) mutation boosts HR pathway in addition to potently driving myeloproliferative neoplasms (MPNs). This can be reversed by JAK2 inhibitor ruxolitinib leading to re-sensitization of MPN cells to PARPis (45).

In circumstances where PARPis are ineffective as a single agent despite DNA repair defects, sensitivity to PARP inhibition can be unveiled by therapeutic interventions that induce or augment DNA damage. Thus, addition of cytotoxic drugs was shown to overcome resistance to PARP inhibition in AMLs driven by MLL fusions attributable to highly active HR (43,46). Hypomethylating agents (HMAs) covalently trap DNMTs on DNA forming bulky adducts, which requires PARP1 activity for repair and hence may synergize with PARPis (47). Vitamin C, an essential TET2 co-factor, promotes PARPi sensitivity likely through increased demand on PARP-dependent base excision repair (BER) that removes products of TET2-catalyzed methylcytosine oxidation during active DNA demethylation (48).

Venugopal et al. found that cells harboring R882 mutant DNMT3A were unable to effectively resolve replication fork damage induced by chain-terminating drugs like cytarabine and fludarabine. This DNA repair defect coincided with impaired chromatin recruitment of PARP1, rendering it functionally inactive (1). Consistently, DNMT3A-mutant cells derived no further benefit from pre-treatment with a PARP inhibitor olaparib suggesting PARPi resistance, while cells with wild-type DNMT3A were sensitized by the combination (1) (Figure 2). In support of this, clinical studies found that efficacy of PARP inhibition correlated with functional levels of PARP1 measured by a novel tracer for microPET imaging, while lack of PARP1 conferred resistance to all PARPis in vitro (49). Although effects of DNMT3A LOF was not investigated in this study, our finding of lack of PARPi synergism in cells with leukemia-associated DNMT3A R882 mutations aligns well with the joint work of Skorski and Challen groups demonstrating that AMLs with DNMT3A mutation or loss were highly resistant to olaparib due to repression of PARP1. On the other hand, TET2-deficient cells were sensitive to PARP inhibition, owing to downregulation of several HR genes (50). Genetic background and choice of primary drug are of paramount importance when devising synthetic lethal approaches for maximal clinical benefit. The possibility of promoting selection and expansion of PARPi-resistant pre-malignant and leukemic clones should be carefully considered given an emerging link between PARP inhibition and clonal hematopoiesis.

Figure 2.

Mechanism-based combinatorial targeting opportunities to augment cytarabine-induced replication stress in AMLs with and without DNMT3A mutations.

Clonal hematopoiesis driven by DNMT3A mutations: collateral damage of PARP inhibitor therapy?

Somatic mutations accumulate in human cells over time, fueling clonal evolution. Each decade, a hematopoietic stem cell (HSC) acquires 100–200 random mutations in its genome, 1 or 2 of which affect protein-coding regions (51). It is estimated that with 100,000–200,000 HSCs present in the bone marrow in early adulthood, on average 600,000 to 1,200,000 protein-coding mutations are accumulated in the HSC pool by the age of 60 (52–55). Although most mutations are functionally inconsequential or even detrimental, select genetic changes may confer growth advantage allowing a handful of HSCs to outcompete their peers. Similar to other tissues such as esophagus and skin (56–60), age-related clonal expansion of HSCs is a common phenomenon, termed clonal hematopoiesis (CH) (61). While emergence of small clones defined by variant allele frequency (VAF) <2% without overt hematologic abnormalities is ubiquitous, presence of larger CH clones (VAF >10%) is associated with increased risk of blood malignancies (62–68) and other adverse outcomes ranging from atherosclerosis to cancer to chronic kidney disease (69–72).

More than 90% of CH mutations fall into one of three major categories: a) epigenetic modifier genes, including DNMT3A, TET2, and ASXL1 (“DTA mutations”), which together define a lion’s share of CH cases; b) splicing factors SF3B1, SRSF2, etc; and c) DNA damage response (DDR) genes such as PPM1D and TP53 (55). DNMT3A lesions dominate the mutation landscape, constituting up to a half of all genetic alterations identified in CH (66,73–75). Although the accumulation of mutations is a stochastic event, research to date has showed smoking, chronic inflammation, and previous exposure to chemotherapy or irradiation can promote context-dependent clonal expansion driven by specific genetic alterations (67,70,76–79). Notably, recent clinical studies demonstrated that previous anti-cancer therapy including with PARP inhibitors selectively expands clones with mutations in DDR related genes PPM1D, TP53, and CHK2 and confers elevated risk of therapy-related myeloid neoplasia (t-MN) (76,80,81).

To date, overwhelming majority of research efforts have been dedicated to investigating the prevalence and implications of CH in various disease settings. In contrast, our knowledge of potential clinical tools for CH prevention and management is still in its infancy. Venugopal et al. showed increased sensitivity to medicamentous replication stress in cells expressing DNMT3A^R882, due to impaired PARP1 recruitment and dampened DNA damage repair. Consistently, primary DNMT3A^WT AML cells were sensitive to pre-treatment with PARP inhibitor olaparib, while DNMT3A^mut samples remained unaffected (1). These data suggest that hematopoietic cells with DNMT3A mutations may exhibit PARPi resistance and warrant investigation of PARP inhibitors for managing non-DNMT3A-driven CH to reduce the risk of progression to hematological malignancies and of adverse outcomes in other diseases. However, considering recent studies linking CH and t-MN to PARPi exposure (76,80,81), careful patient stratification will be paramount for balancing potential beneficial and detrimental consequences of this intervention. Indeed, a meta-analysis of over 7,000 evaluable patients enrolled in 18 randomized placebo-controlled clinical trials of PARPis found significantly increased risk of MDS and AML during follow-up (82). This concerning finding raises an important question: what strategies can be devised to target PARPi-resistant pre-malignant and leukemic clones?

Combinations with p53 activators – an emerging therapeutic opportunity for DNMT3A-mutated malignancies

Despite being the most commonly inactivated tumor suppressor in human cancers (83,84), the TP53 gene which codes for the p53 protein is rarely mutated in de novo AML (85–87). Instead, many leukemias evade its tumor suppressor function by increased expression of p53 negative regulators MDM2 or MDMX, which inactivate p53 through direct protein binding and ubiquitin-mediated degradation (88–90). Retention of functional p53 in these neoplasms spurred efforts to develop targeted anti-cancer therapies that aim to reverse its negative regulation.

Interest in compounds that impede p53 binding to MDM2 and/or MDMX and thus restore p53 activity has led to an explosion of small molecule drug candidates, as well as stapled peptides and PROteolysis TArgeting Chimeras (PROTACs) (90–94). Additionally, taking aim at negative regulators of the p53 downstream effectors, such as BCL-2 inhibitor venetoclax which in combination with hypomethylating agents is now considered the standard of care for AML and MDS patients ineligible for intensive chemotherapy (95,96), has shown great promise. For tumors that express mutant p53 protein, including gain-of-function isoforms, there is a class of therapies that target aberrant p53 either via inhibiting its activity or by refolding it to the correct conformation (97). These and other strategies are reviewed in detail elsewhere (97–99).

Our previous studies observed more efficient engagement of the p53 signaling cascade and increased apoptosis in cells with DNMT3A R882 mutations in response to cytarabine (1). Pharmacologic stabilization and potentiation of p53 by pre-treatment with MDM2 inhibitor nutlin-3a was effective in further augmenting cytarabine-induced cell killing. In agreement with this, ex vivo drug dose response studies conducted as part of the BeatAML clinical trial (100) found preferential sensitivity to the cytarabine/nutlin-3a combination specifically in AML specimens with DNMT3A R882 mutations. These findings are a strong indication that DNMT3A^R882 may be a predictive biomarker for the efficacy of p53-potentiating therapy (Figure 2). The pressing need to identify mechanistically-informed criteria for patient stratification is highlighted by the results of a recent double-blind placebo-controlled clinical trial evaluating the efficacy of idasanutlin, a derivative of nutlin-3a with improved drug-like properties, in combination with cytarabine in patients with relapsed/refractory AML (ClinicalTrials.gov identifier: NCT02545283). Although the experimental arm showed a favorable safety and toxicity profile (no patients were discontinued due to adverse events), the trial was terminated as meaningful superiority was not reached at time of interim analysis. The design of this study did not mandate molecular testing beyond the clinically-actionable mutation-defined subpopulations (FLT3 and IDH1/2). Yet, the IDH-mutated subtype tended to show superior rate of complete response and complete response with incomplete hematologic recovery (CR/CRi) at the end of induction. This indicates that more granular genetic characterization to separate disease subgroups wherein p53 potentiators are beneficial is warranted and feasible.

Further studies are necessary to clarify molecular mechanisms underlying enhanced cytarabine/nutlin-3a response in DNMT3A-mutant cells. It has been reported that p53 allosterically represses methyltransferase activity of wild-type DNMT3A, and this inhibition is blocked by R882 hotspot mutation (16). Whether p53 can promote the protective effect of wild-type but not mutant DNMT3A in stabilizing damaged replication forks, or if wild-type and mutant DNMT3A are differentially involved in p53 downstream responses (101), remains to be determined, along with potential implications of DNMT3A LOF mutations. Changes in p53 axis in the context of mutant DNMT3A may merit separating AML patients with this mutation into a distinct treatment category that may particularly benefit from a combination treatment with cytotoxic chemotherapy and MDM2/MDMX inhibitors.

Concluding remarks and translational outlook

Detailed mechanistic understanding of the dependencies and synthetic-lethal interactions in genetically defined disease subtypes is crucial to improving treatment response rates in AML. This is an exciting and rapidly growing field and many questions remain unanswered. How is DNA damage repaired depending on treatment and disease subtype? Which factors determine efficacy and should be used for patient stratification? Can mechanistically-guided combination approaches offer lower toxicity in elderly and/or frail patients while maintaining the “intent to treat” of high-intensity regimens? What intervention strategies can be devised to manage clonal hematopoiesis and to reduce the risk of malignant progression, and which patient subgroups will they apply to? We propose that AML patients with wild-type DNMT3A may benefit from a combination of PARP inhibitors with low intensity regimens of replication stalling nucleoside analogs in the absence of other negative predictive factors. Conversely, cases bearing DNMT3A hotspot R882 mutation may benefit from MDM2/MDMX inhibitors (Figure 2). Better understanding of the mechanisms of DNA damage repair, together with personalized therapy based on the genetic landscape of the disease, will be crucial to improving outcomes in this challenging group of leukemia patients.

HIGHLIGHTS.

• Chain-terminating nucleotides like cytarabine are removed by NER, resection of reversed replication forks, or HR

• DNMT3A^R882 predisposes to replication stress and sensitizes to replication-stalling chemotherapeutics like cytarabine

• PARP1 recruitment defect disrupts DNA damage repair in cells with mutant DNMT3A, renders cells insensitive to PARPi

• Combination with p53 potentiator/MDM2 inhibitor nutlin-3a augments cytarabine sensitivity with DNMT3A^R882