Clinical Trials Assessing Hypomethylating Agents Combined with Other Therapies: Causes for Failure and Potential Solutions

The small-molecule drugs azacitidine (5-azacytidine) and decitabine (5-aza-2′-deoxycytidine) are DNA hypomethylating agents (HMA). That is, both agents inhibit and deplete the key epigenetic regulator DNA methyltransferase enzyme 1 (DNMT1). DNMT1 is the “maintenance methyltransferase” that copies methylation marks of the parental DNA strand onto the newly synthesized strand during cell-cycle S-phase, and is also a corepressor (a protein mediating gene repression) recruited to gene loci by lineage master transcription factors. The HMAs are the most effective single agents for treating myelodysplastic syndromes (MDS) and acute myelogenous leukemias (AML), with response rates of approximately 35%–60%. Numerous attempts to further improve outcomes by combining HMAs with investigational drugs, however, have failed in randomized controlled clinical trials (RCT), except for combinations with the BCL2 inhibitor venetoclax. We extract from this hard-fought history: (i) properties of HMAs that can explain their success versus numerous other single agents, so that future trials can seek to preserve or enhance these properties; (ii) principles for HMA dose and schedule selection, taught by serial HMA single-agent studies; (iii) classes of agents to consider for combination therapy, suggested by successful RCTs; (iv) lessons learned from unsuccessful RCTs, to avoid repeating previous miscalculations. In this way, we can hopefully increase likelihoods of future success.

Azacitidine and decitabine are the only drugs approved, and routinely used as single agents, to treat all subtypes of MDS, succeeding where cytarabine, hydroxyurea, topotecan, fosteabine, gemcitabine, irinotecan, daunorubicin, etoposide, and many other drugs failed over several decades of clinical trials. These other drugs are cytotoxic, that is, their intended mechanism is DNA damage that upregulates the transcription factor p53 (TP53) and its downstream apoptosis transcriptional programs (reviewed in ref. 1). Attenuation of p53 or its major cofactors by mutations/deletion of TP53, amplifications of MDM2/4, and/or by other means (1–3), is, however, characteristic of malignant transformation of most tissue lineages, including myeloid. Thus, there can be upfront resistance to cytotoxic treatments (primary refractory), or ready selection by first-line cytotoxic treatments for the most apoptosis-attenuated malignant sub-clones that then resist next-line cytotoxic treatments also (relapsed-refractory; ref. 1). TP53 mutations are hence negative-prediction biomarkers for cytarabine-based cytotoxic therapy, the historical standard for treating AML (4). Meanwhile, normal dividing hematopoietic stem and progenitor cells (HSPC) do undergo apoptosis, because these have intact p53, thereby exacerbating or causing low blood counts (3, 5). This is problematic because morbidity and death in MDS/AML is usually from low normal blood counts to begin with, and clinical response/benefit requires not just cytoreduction of malignant clones, but blood count recoveries mounted by functional HSPC. Amplifying this problem, reserves of functional HSPC diminish markedly with age (reviewed in ref. 6) and the majority of patients with MDS/AML are >60 years of age.

By contrast, HMA efficacy does not depend on p53-dependent apoptosis: DNMT1 targeting prompts MDS/AML cells to terminal differentiation, exits from cell cycle that do not require p53 (refs. 1–3;Figs. 1 and 2). In normal HSPCs, DNMT1 targeting is not inherently cytotoxic but preserves lineage maturation in the case of committed progenitors, or self-renewal in the case of stem cells, thus sparing these vital cells (refs. 5, 7–10; Fig. 1). That is, DNMT1 targeting can terminate malignant but not normal hematopoietic stem cell self-replications (good therapeutic index). Accordingly, TP53 mutation and/or deletion is not a negative predictive biomarker for HMA clinical activity. In a randomized comparison of azacitidine 75 mg/m²/d for 7 days every 28 days versus conventional care regimens, including standard cytotoxic induction therapy with cytarabine and daunorubicin, in patients with newly diagnosed AML, subset analysis of patients with chromosome 17p (TP53) abnormalities (n = 46) found median overall survival (OS) of 5 versus 2.8 months (log-rank P value = 0.07; ref. 11). Although this analysis was limited by the small patient number, single-arm studies from more than one other institution have also documented meaningful responses to HMA in patients with TP53-mutated disease, and more generally, in the very elderly with high-risk disease and diminished myelopoietic reserves (2, 3, 12, 13). Having said this, TP53 abnormalities are still very much an adverse prognostic biomarker in HMA-treated patients; although clinically meaningful hematologic responses occur, these responses are short-lived (11). Even though HMAs have been shown scientifically to not require p53 to antagonize MYC and terminate malignant replication, p53 inactivation upregulates the endogenous de novo pyrimidine synthesis pathway that directly antagonizes the DNMT1-depleting nucleotide (Fig. 2E and F; ref. 14; reviewed in ref. 15). In sum, there remains a need to enhance therapy. Mechanism and clinical trial data thus far suggest that such efforts should seek to preserve or enhance a normal HSPC-sparing, p53-independent mechanism-of-action of HMAs, properties that distinguish HMAs from numerous less or non-beneficial evaluated drugs (Figs. 1 and 2; Table 1).

The earliest trials of azacitidine were based on conventional oncotherapy objectives of apoptosis/cytotoxicity, because high concentrations of azacitidine or decitabine do cause nucleotide imbalances, damage DNA, and can thereby activate p53/apoptosis (Fig. 1; Table 1). Doses were therefore based on MTDs (16; reviewed in ref. 17). These trials were unsuccessful. DNMT1 targeting, however, is achieved and saturated at relatively low azacitidine concentrations (Fig. 1), and 10-fold reductions from initially evaluated doses produced clinical success. Standard azacitidine regimens today administer 50–75 mg/m²/d (17). Decitabine clinical development was similar. Initial, unsuccessful trials administered daily doses as high as 600 mg/m² (17); the first approval used a >10-fold lower dose of 45 mg/m², and a second-approved regimen uses an even lower dose of 20 mg/m²/d (18, 19). These dose reductions were arrived at empirically, therefore, we conducted studies to formally identify minimum decitabine doses needed to deplete DNMT1 without cytotoxicity to normal bone marrow/HSPC in non-human primates and humans: 0.1–0.2 mg/kg/d (∼5 mg/m²/d; refs. 2, 20, 21).

In the earliest HMA trials, cytotoxic daily doses could only be tolerably administered for a few days, for example, 3 days, followed by intervals of approximately 6 weeks needed to recover from the cytotoxic side-effects (schedules were necessarily pulse-cycled). One benefit of dose reductions that decreased cytotoxicity was that intervals between treatment pulses could be shortened. Standard azacitidine regimens administer 50–75 mg/m²/d for 7–10 days every 4 weeks; the most commonly used standard regimen of decitabine administers 20 mg/m²/d for 5 days every 4 weeks. This more frequent administration is beneficial because DNMT1 targeting by azacitidine or decitabine requires overlap between treatment exposure windows and malignant cell S-phase entries. That is, DNMT1 targeting is S-phase or exposure time dependent, and lowering the decitabine dose from 45 to 20 mg/m²/d, and increasing the frequency of administration from 3 days every 6 weeks to 5 days every 4 weeks produced 2–3-fold improvements in remission and hematologic improvement rates (18, 19). Building on this logical progression, we scheduled non-cytotoxic, DNMT1-depleting doses of decitabine 0.1–0.2 mg/kg/d (∼5 mg/m²/d) for frequent and distributed administration 1–2X/week. The overall response rate (ORR) per International Working Group criteria was 44%, including complete cytogenetic remissions of TP53-mutated disease containing complex cytogenetic abnormalities (2, 12, 22). This non-cytotoxic regimen has been effectively and safely administered for >5 years in patients with MDS >80 years old (12).

Thus, pulse-cycled schedules of HMA administration reflect conventions inherited from cytotoxic treatments. Lower doses that target DNMT1 without cytotoxicity enable alternative schedules of administration that increase drug exposure durations and distributions to increase S-phase–dependent DNMT1 targeting.

HMAs can produce non-toxic clinical responses but relapse/resistance is routine. Resistance, in vitro, in mice and in patients, is characterized by failure to deplete DNMT1 (23, 24). To deplete DNMT1, azacitidine, a cytidine analog, and decitabine, a deoxycytidine analog, must be processed by pyrimidine metabolism into a nucleotide analog that incorporates into the newly synthesized DNA strand during S-phase (Fig. 1). Pyrimidine metabolism is a network evolved for nucleotide homeostasis. The interactions between decitabine and azacitidine and this network perturb nucleotide balances, and thus trigger automatic, balancing responses that dampen decitabine or azacitidine pro-drug processing into DNMT1-depleting nucleotide (Fig. 2; ref. 23). One key pyrimidine metabolism adaptation underlying this mode of resistance is upregulated de novo pyrimidine synthesis, which competes directly with the DNMT1-depleting nucleotide Aza-dCTP by building natural cytidines and deoxycytidines from amino acid building blocks (Fig. 2). Inhibiting de novo pyrimidine synthesis can thus restore AML cell sensitivity to HMAs (14, 15). Venetoclax is a BCL2 inhibitor that dislodges from BCL2 sequestration BH3-only proteins; BH3-only proteins then bind to and enable the effector proteins BAX and BAK1 to permeabilize mitochondrial membranes (Fig. 3). This depowers mitochondrial processes, including de novo pyrimidine synthesis (Fig. 3; ref. 25). Also decreased are mitochondrial outputs of nicotinamide adenine dinucleotide (NAD⁺/NADH) and flavin adenine dinucleotide (FAD⁺/FADH), mandatory cofactors for the transcription repressing enzymes C-terminal binding protein 1 (CTBP1), and lysine demethylase 1A (KDM1A), that like DNMT1, are implicated in oncogenic, aberrant epigenetic repression of hundreds of lineage-differentiation genes in oncogenesis (Fig. 3; refs. 15, 25–27). Thus, successful mitochondrial membrane depolarization by venetoclax, short of triggering full-blown apoptosis, can augment DNMT1 targeting/terminal differentiation in several ways (Fig. 3).

Other cofactors produced by mitochondria, for example, alpha-ketoglutarate (AKG), serve epigenetic enzymes that promote terminal differentiation, for example, ten-eleven translocation 2 (TET2) that demethylates DNA (Fig. 3). AKG is antagonized by an oncometabolite 2-hydroxyglutarate (2HG) that is produced by mutant IDH1 or IDH2 (IDH1 and IDH2 mutations are recurrent in myeloid malignancies). Accordingly, inhibitors of mutant IDH1/2 align with therapeutic index/pathway goals of HMA therapy, and early clinical trial data for such combinations are encouraging (Fig. 3).

The pyrimidine metabolism enzyme, cytidine deaminase (CDA), rapidly catabolizes azacitidine and decitabine into uridine counterparts that do not target DNMT1, and also contributes to HMA resistance (Fig. 2; ref. 23). CDA in the liver and gastrointestinal tract moreover severely limits HMA plasma half-life and oral bioavailability (28, 29). A fixed-dose co-formulation of the CDA inhibitor cedazuridine with oral decitabine was recently approved for MDS treatment in the United States based on pharmacokinetic equivalence to parenteral decitabine (30). Another CDA inhibitor, tetrahydrouridine, has also been co-formulated with oral decitabine and 5-azacytidine, with weight-based dosing and schedules of administration designed for non-cytotoxic DNMT1 targeting (28, 31, 32).

Clinical trial experience thus suggests utility in combining HMAs with agents aiming to manipulate pyrimidine and mitochondrial metabolism, to thereby enhance pro-drug processing and non-cytotoxic corepressor targeting, with open opportunities to refine and optimize such combinations.

We analyzed nine completed, unsuccessful, combination RCTs (Table 2) and identified departures from, and workable design modifications to better align with the mechanism and pathway principles learned thus far, to in this way inform future efforts.

Histone deacetylase inhibitors (HDACi), by promoting acetylation of histones and other proteins, can also target the epigenome to reverse oncogenic gene silencing. HMA combination with HDACis was found synergistic in preclinical studies, and produced high response rates in small, single-arm studies (33). Entinostat is a potent oral inhibitor of HDAC class I and was combined with azacitidine in the E1905 RCT in patients with MDS and AML with myelodysplasia-related changes (34). Azacitidine alone was favored with hematologic normalization (HN) in 32% versus 27% for the combination (Table 2). Pracinostat is a class I/II/IV HDACi and was combined with azacitidine in an RCT in patients with International Prognostic Scoring System (IPSS) intermediate (INT)-2 or high-risk MDS (35). The combination did not provide a significant ORR or OS benefit versus azacitidine plus placebo (Table 2). Recently, an RCT of pracinostat plus azacitidine versus placebo plus azacitidine in patients with AML was terminated prematurely, due to failure to meet its primary endpoint of complete remission (CR) rate superiority (36). Vorinostat is an HDAC class I/II inhibitor; multicenter open-label RCT compared vorinostat plus azacitidine versus azacitidine monotherapy in patients with newly diagnosed or relapsed/refractory AML or IPSS INT-2 or high-risk MDS (37). Unfortunately, this trial also failed to meet its ORR and OS primary endpoints. Valproic acid is a pan-HDAC inhibitor (38); an RCT evaluation of valproic acid plus decitabine versus decitabine alone in INT or high-risk patients with MDS and AML older than 60 failed to demonstrate significant ORR or OS benefit for the combination (Table 2; ref. 39). Finally, SWOG-S1117 randomized patients with higher-risk MDS (81%) or chronic myelomonocytic leukemia (CMML; 19%) into three arms: azacitidine plus lenalidomide versus azacitidine plus vorinostat versus azacitidine monotherapy (40). No significant improvement in the ORR primary endpoint was observed between either combination arm versus azacitidine monotherapy (Table 2). An ORR difference was observed in subset analyses of patients with CMML, favoring azacitidine plus lenalidomide (ORR 68%, n = 19) over azacitidine alone (ORR 28%, n = 18; P = 0.02), but without a difference in OS (P = 0.87). On the basis of the overall non-superior ORR and OS outcomes, the phase III part of the study was not conducted.

HMAs have also been combined with drugs targeting mutant FMS-like tyrosine kinase 3 (FLT3) or mutant p53, and with thrombopoietin (TPO)-receptor agonists. Again, single-arm trial results were encouraging, but not RCT readouts. FLT3 is among the most recurrently mutated genes in AML and encodes for a transmembrane ligand–activated receptor. Gilteritinib is a potent second-generation FLT3 inhibitor active against both FLT3 internal tandem duplication and FLT3 tyrosine kinase domain mutations. Several RCTs have failed to demonstrate superiority of FLT3 inhibitors combined with HMAs versus HMAs alone; most recently, the phase III, multicenter, open-label LACEWING RCT compared gilteritinib plus azacitidine to azacitidine alone in patients with newly diagnosed FLT3-mutated AML ineligible for induction chemotherapy. Although the phase I–II results looked promising, the phase III failed to meet its primary endpoint—there was no OS benefit (Table 2; ref. 41).

TP53 mutations in MDS or AML predict poor prognoses, and a small molecule, APR-246 (eprenetapopt), has been postulated to convert the mutant-p53 protein to its original wild-type form. In RCT evaluation, patients with TP53-mutated MDS received APR-246 plus azacitidine versus azacitidine alone. Again, phase IB/II results were encouraging but phase III failed to show significantly higher CR rate for the combination versus azacitidine alone (Table 2; ref. 42). Insofar as azacitidine/decitabine success to treat MDS/AML is because of a p53-independent mechanism-of-action, it may be more rational to combine this class of agent (p53 stabilizer) with agents intending p53-dependent cytotoxicity, for example, cytarabine (Fig. 2).

Thrombocytopenia in patients with MDS/AML is a major source of morbidity, and single-agent TPO-receptor agonists have shown efficacy in raising platelet counts. A double-blind RCT-assigned IPSS INT-1, INT-2, and high-risk patients with MDS with baseline platelet count less than 75 × 10⁹/L to eltrombopag plus azacitidine versus placebo plus azacitidine (43). The study ended prematurely because it did not meet its primary endpoint of platelet-transfusion independency, and there was no significant improvement in ORR or OS with the combination (Table 2).

These combinations did not add benefit and sometimes demonstrated worse responses than with HMAs alone (34, 40). Here are reasons:

• (i) Direct antagonism of S-phase–dependent HMA by concurrent administration of cytostatic/cytotoxic investigational agent. A common attribute of these RCTs is that the investigational agents can have cytostatic effects—even eltrombopag has cytostatic effects on leukemia cells (44, 45)—and were given concurrently/overlapping with the HMA (Table 2). Such scheduling has the potential to antagonize the mainstay HMAs, for example, correlative analysis of DNA methylation showed that combination entinostat plus azacitidine produced less demethylation than the azacitidine-alone arm (34). It would be reasonable, therefore, to avoid concurrent (same day) scheduling of less active agents that have cytostatic effects, including HDACi, immunomodulatory imide drugs (IMiD), or kinase inhibitors, with the HMA backbone, because this would antagonize S-phase–dependent DNMT1 depletion (Fig. 1).

• (ii) Investigational agent suppression of functional myelopoiesis. Known side-effects of HDACi as single agents, even in patients without myeloid malignancies, are thrombocytopenia, neutropenia, and anemia (40). Such myelosuppression may delay or decrease feasible exposure times to the more active HMA. How often an HMA is administered dictates the fraction of the malignant cell population that has the possibility of being treated because DNMT1 targeting is S-phase dependent. In successful RCT evaluations of azacitidine plus venetoclax, low blood counts attributed to therapy were managed by reducing venetoclax frequency and/or dose (46). By contrast, in other RCTs, myelosuppression was managed by reducing azacitidine frequency and dose, even discontinuing azacitidine altogether, favoring continuation of investigational agents shown to be less active as single agents (40). Thus, myelosuppression attributed to treatment should be managed as much as possible with dose and/or frequency reductions of the less-active investigational agent, not the HMA.

• (iii) Destruction of normal HSPC (Fig. 1; Table 1). Separation of epigenetic pharmacodynamic effects that terminate malignant self-replications, from cytotoxic effects that destroy normal HSPC, has been shown for HMA but not for HDACi (47–50). This could be because HDACs have pleotropic cell physiology functions, not restricted to epigenetic repression of genes (HDAC deacetylate substrates other than histones; refs. 47–50). HDACis thus have significant side-effects but modest activity for treating myeloid malignancies (reviewed in ref. 33). If biochemically possible, it would be useful to design, identify, and use doses of the HDACi that inhibits HDAC, or the IMiD that alters cereblon substrate specificity, or the kinase inhibitor to inhibit FLT3, without cytotoxicity to normal HSPC (dosage selection based on pharmacodynamics instead of maximum-tolerated levels). Investigational agent dose, as per the HMA, should be selected per molecular-targeted pharmacodynamic (“optimal-biological-dose”) not MTD principles, to limit or avoid indiscriminate antimetabolite effects and cytotoxicity to normal HSPC needed for response and durable benefit.

• (iv) Fidelity to schedules of administration. Non-superiority with investigational agent addition to HMAs in RCTs implies the HMA backbone also largely accounts for preceding phase II results that were found exciting. One reason for apparently better results than historic for the same HMA regimen could be fidelity to schedules-of-administration, important for S-phase–dependent DNMT1 targeting. In one phase II evaluation of decitabine 20 mg/m²/d for 5 days every 28 days, growth factor support was used to alleviate neutropenia that might otherwise threaten on-time administration of cycles every 28 days (18, 19), whereas in another phase II evaluation of the same regimen, neutropenia was managed by cycle delays (19); the trial pursuing on-time HMA administration reported ORRs of 63% versus 35% for the trial using cycle delays (Fig. 1; Table 1; refs. 18, 19). Hence, it could be useful to consider growth factor support, or other measures, to enable on-time administration of the HMA; avoid cycle delays (17).

Another observation is lower response rates for the same combination treatment in phase III versus prior phase II, for example, eprepranocept + azacitidine produced a CR rate of 33.3% in the phase III RCT (n = 77) but 50% in the prior single-arm phase II (n = 40; ref. 42). A straightforward explanation is less selected and larger (more representative) sample sizes in phase III.

In sum, >40 years of clinical trial evidence, complemented by mechanism studies, suggests selecting HMA dose and schedule to increase S-phase–dependent DNMT1 targeting, manipulating pyrimidine metabolism to increase HMA pro-drug processing into DNMT1-depleting nucleotide, and targeting of other epigenetic enzymes implicated in oncogenic repression of lineage-maturation programs, all the while adhering to basic therapeutic index principles that are especially critical when treating myeloid malignancies in the elderly.

A. Shastri reports other support from Kymera Therapeutics and OncLive, as well as personal fees from Janssen Pharmaceuticals outside the submitted work. A.K. Verma reports grants from BMS, Janssen, Curis, and Prelude, as well as other support from Stelexis during the conduct of the study. Y. Saunthararajah reports other support from EpiDestiny and Novo Nordisk during the conduct of the study. In addition, Y. Saunthararajah has a patent for tetrahydrouridine and decitabine issued, licensed, and with royalties paid from EpiDestiny; has a patent for tetrahydrouridine and 5-azacytidine pending; has a patent for ISWI inhibitor (tumor differentiation therapy) issued; is in research and development of drug strategies for non-cytotoxic targeting of epigenetic proteins, defining how to combine tetrahydrouridine, decitabine, and 5-azacytidine; is leading a team that is developing a non-cytotoxic inhibitor of the ISWI family of ATP-dependent chromatin remodelers; and reports patents and industry partnerships in this field of research and development. No disclosures were reported by the other authors.

This work was supported by the National Heart, Lung and Blood Institute PO1 HL146372, National Cancer Institute P30 CA043703, National Cancer Institute RO1 CA204373, Robert and Jennifer McNeil, Lescek and Jolanta Czarnecki, and Dane and Louise Miller (to Y Saunthararajah). This work was also supported by National Heart, Lung and Blood Institute R01HL139487 and R01HL150832, Albert Einstein Cancer Center Support grant (P30CA013330), Evans MDS grant, and V Foundation for Cancer Research (to A.K. Verma).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.