We performed a systematic review and pooled analysis to compare the outcomes of HMA vs. CCR in patients with AML and MDS. The combined analyses revealed statistically significant OS and CR/PR benefit with HMA therapy in comparison to CCR (Fig. 2). These results confirm that HMA are reasonable therapeutic options with survival advantage, especially for elderly and transplant ineligible AML and MDS patients.

Aberrant DNA methylation has been suggested as a dominant mechanism of MDS progression to AML, and patients with MDS and AML have been shown to have unique patterns and abundance of aberrant DNA methylation compared to normal controls [40, 41], thus representing a suitable therapeutic target. Azacitidine (5-azacytidine) is metabolized into decitabine (5-aza-2′-deoxycytidine), which forms a covalent protein-DNA adduct, depleting intracellular methyl-transferase, leading to reversal of DNA hypermethylation on tumor suppressor genes and induction of apoptosis [8, 10–13]. As such, DNA demethylation has been widely accepted as the primary mechanism of cytotoxicity of HMA and a previous study showed the association of CDKN2B (that encodes p15 ^INK4B ) pre-treatment methylation level with treatment response to azacitidine [42]. However, interestingly, our subgroup analyses demonstrated an association of OS benefit with azacitidine treatment, but not with decitabine (Table 2), similar to a recent retrospective study with AML patients that showed superior outcomes with azacitidine therapy in comparison to decitabine [25]. Furthermore, Fandy et al. showed that reversal of methylation on four tumor suppressor genes (p15 ^INK4B , CDH-1, DAPK-1, and SOCS-1) had no prognostic impact on clinical response to the combination treatment of azacitidine and entinostat (histone deacetylase inhibitor) [43, 44]. Collectively, these results indicate potential cytotoxic mechanisms that are independent to DNA demethylation.

Both azacitidine and decitabine have been shown to induce DNA damage and cell cycle arrest, however, to different extents [45, 46], suggesting DNA damage as a possible underlying mechanism of HMA-induced cytotoxicity. However, a similar degree of γ-H2AX expression was observed in both responders and non-responders to azacitidine and entinostat treatment [43], questioning the role of DNA damage. Further studies are needed to define the role of DNA damage in the cytotoxic effect of azacitidine. In a recent study, azacitidine, but not decitabine, was shown to inhibit RNA methylation on cytosine 38 and 48 of tRNA^Asp, which are target sites of DNMT2, and reduce the metabolic activity in myeloid cell lines. This suggests that azacitidine induces cytotoxicity via tRNA demethylation rather than DNA although the detailed mechanisms still remain to be answered. Also, Roulois et al. showed that the anti-tumor effect of low-dose decitabine may depend on viral mimicry, activating MDA5/MAVS/IRF7 RNA recognition pathway in colorectal cancer-initiating cells [47], and its role in myeloid neoplasms need further investigation.

Cytogenetic risk and BM blast counts are known to be independent prognostic factors in AML and MDS [39, 48–50]. However, in our study, subgroup analysis according to cytogenetic risk or BM blast count failed to show any RR difference (Table 2). Ninety-eight percent of MDS patients enrolled in three trials [33, 35, 36] had intermediate or high IPSS risk and two studies [34, 37] included only AML patients whose prognosis is known to be dismal, partially explaining why there was no subgroup difference.

Previously, LDAC has been shown to be associated with higher response rates compared to supportive care in elderly AML or MDS patients who were not candidates for allogeneic SCT [51]. As such, three trials [34, 36, 37] in the current analysis incorporated LDAC as their control regimens. In a subgroup analysis comparing HMA and LDAC in AML patients, the OS rate in LDAC treatment group was significantly lower than that of azacitidine, but not decitabine, suggesting that azacitidine may be a better therapeutic option in this patient group (Additional file 1: Figure S4).

We recognize several limitations of the current analysis. First, there was significant heterogeneity in the OS analysis (I ² = 89 %, p < 0.00001). The primary source of heterogeneity was experimental agents as shown in intra-subgroup homogeneity of three azacitidine trials [35–37] (I ² = 0 %, p = 0.38) as well as two decitabine trials [33, 34] (I ² = 0 %, p = 0.85). Although the proportion of females in decitabine studies was relatively higher, other demographic profiles or clinical parameters were not substantially different from azacitidine trials (Table 1 and Additional file 1: Table S2). BM morphologic abnormalities were shown to be associated with worse prognosis in de novo AML [52], and 32.4 % of patients (n = 158) in one azacitidine study [37] had AML-MRC, which may result in heterogeneity of treatment outcomes. However, a recent ad hoc study with AML-MRC patients [53] showed significantly higher OS and CR/CRi rates with azacitidine, similar to the original data, rendering this possibility less likely. Second, in one study [35], 53 % of patients who were initially randomized to the CCR group crossed over to azacitidine treatment. However, the outcomes of these patients were analyzed along with the CCR group, which may underestimate the efficacy of azacitidine. Third, inclusion of LDAC and IC as BSC in three studies [34, 36, 37] may have contributed to the heterogeneity, although the result from direct comparison of HMA and LDAC was not significantly different from the original analysis (Additional file 1: Figure S4). Fourth, schedule and administration method and number of cycles of HMA used in the individual studies were different, potentially generating additional heterogeneity based on previous studies that showed better treatment response with prolonged azacitidine treatment [54] and lower bioavailability of subcutaneous azacitidine compared to intravenous administration (AUC values 89 %) [55]. Lastly, the maximum tolerated dose (MTD) of decitabine is known to be 1500–2000 mg/m²/course [56]. However, decitabine trials in our analysis used a 10–25 mg/m²/day dose, which is significantly lower than the MTD. This is based on previous phase I trial that focused on pharmacodynamics rather than MTD [57]. The optimal dose-schedule of decitabine, an S-phase specific agent, needs be further investigated in the future study.

Our systematic review and pooled analyses have identified several areas that require further study. First, the mechanisms of action of HMA and their therapeutic targets remain to be poorly defined. Inhibition of tRNA has been suggested as a potential mechanism of azacitidine; however, details supporting this need further elucidation. Recently, activation of the MDA5/MAVS RNA recognition pathway was suggested as an underlying mechanism of decitabine-induced cytotoxicity in colorectal cancer-initiating cells, and its role in myeloid neoplasm requires further elucidation. Second, the benefit of initial treatment response with decitabine failed to translate into OS improvement, supporting the possibility that the superior OS with azacitidine may result from better disease control, the mechanisms of which remain unknown. In the same context, the optimal dose of decitabine still remains to be defined. Third, potential biomarkers that might have prognostic relevance with regards to response to HMA therapy, including nucleoside transporters like hCNT1 [58] and cytosine deaminase activity [59], also need further study. Finally, a second-generation HMA has been developed to reduce elimination of decitabine by cytidine deaminase, thereby increasing the in vivo exposure of decitabine. A recent phase I clinical trial with SGI-110 (dinucleotide of decitabine and deoxyguanosine) demonstrated a comparable safety profile to decitabine with a significantly longer half-life [60]. An ongoing phase II clinical trial (NCT01261312) will hopefully provide more data in regards to the clinical activity of second generation HMA.

In an analysis of prospective randomized controlled trials in elderly patients with AML or MDS, HMA therapy was associated with improved response rates and OS in comparison to CCR, which included BSC, LDAC, and IC. Further analysis demonstrated that the observed survival benefit was restricted to azacitidine therapy, suggesting that azacitidine may be a better therapeutic option in AML and MDS patients. Finally, we also conclude on the need for additional mechanistic and epigenetic work to better understand mechanisms of action of HMA, optimal dosing strategies, and the identification of prognostic markers and biomarkers to help better predict and monitor response to therapy with these agents.