A novel non-invasive monitoring assay of 5-azacitidine efficacy using global DNA methylation of peripheral blood in myelodysplastic syndrome

Introduction

Myelodysplastic syndrome (MDS) is a clonal hematopoietic disorder characterized by peripheral blood cytopenia and dysplastic features of hematopoietic cells. The prognosis of MDS patients has been defined by cytopenia, percentage of bone marrow (BM) blasts, and cytogenetic risk factors, which comprise the so-called International Prognostic Scoring System (IPSS) established in 1997.¹ Low and intermediate-1 scores have been defined as low-risk MDS according to the IPSS, while intermediate-2 and high scores have been defined as high-risk MDS. The goal of therapy for low-risk MDS patients is to deviate from transfusion dependency; for high-risk MDS patients, the goal is to prevent leukemic transformation.² The European LeukemiaNet has provided clinical recommendations for diagnosis and appropriate therapeutic interventions based on evidence- and consensus-based guidelines for adult patients with primary MDS in 2013.³ In addition, the US National Comprehensive Cancer Network (NCCN) guidelines for the management of MDS have been updated. Both of these guidelines commonly recommend hypomethylating agents (HMA) as a key medication for high-risk MDS patients.³ Moreover, NCCN guidelines have proposed HMA for low-risk MDS patients who do not respond to or are intolerant to other therapeutic options (https:www.nccn.org/professionals/physician_gls/pdf/mds/pdf).

5-azacytidine (azacytidine; AZA) is the most widely used HMA, and approximately 50% of AZA-treated MDS patients experience hematological improvement.^4,5 Because AZA incorporates both RNA and DNA, it exerts bipolar effects on hematological improvement. AZA induces apoptosis via RNA incorporation and demethylation via DNA integration; therefore, this double effect reflects the reduction in blast numbers and recovery from cytopenia. The remaining 50% of AZA-treated MDS patients do not show any response, even after treatment with at least six courses of 75 mg/m² AZA.⁶ For this primary AZA treatment failure, it remains unclear which MDS patients respond to AZA treatment, although several genes are considered linked to AZA sensitivity.^7–9 Moreover, some AZA responders do not maintain consistent hematologic improvement and subsequently develop acute leukemia, termed secondary AZA failure. Evidence suggests that hypomethylation of the promoter region in tumor suppressor genes is not always related to a clinical response attributable to AZA.^9–11 Recent reports have also demonstrated that somatic mutations, such as TP53 and TET2 mutation,^7,8,12–15 has been identified as a poor prognostic factor for MDS patients, and AZA response among TP53 mutated patients is better compared with non-mutated patients.¹² However, this response is temporary, and overall survival is not improved. TET2 is one of the most frequently identified mutation,¹³ although its effect on HMA response is controversial. Some studies have shown that TET2 mutation is a favorable indicator of AZA efficacy,^7,8 while others do not support such a relationship.^14,15 To date, no predictive biomarker for secondary AZA failure has been described. BM specimens are typically used in the initial diagnosis or relapsed phase, but are less appropriate during the follow-up period. We therefore applied a single-molecule methylation assay (SMMA), whereby methylated DNA in samples could be quantified using the methylation index (MI), to establish a non-invasive monitoring method for AZA efficacy using peripheral blood cells.¹⁶ We first analyzed MI in 34 untreated patients with MDS, including 6 post-MDS acute myeloid leukemia (AML) patients, and then measured the MI sequentially in 13 AZA-treated patients.

Our study indicates that whole-blood DNA from all 34 patients had low MI values compared with healthy volunteers, regardless of the percentage of blasts, indicating that the sequential analysis of MI in each patient may be more reliable for determining AZA efficacy. Indeed, appropriate MI recovery of peripheral blood after AZA treatment was found in AZA responders. The sequential analysis of MI using peripheral blood may therefore represent a non-invasive monitoring marker for use in the treatment of patients with MDS.

Materials and methods

Patients

Thirty-four patients with MDS and post-MDS AML (aged 23−83 years; mean, 66.2 years) were enrolled in this study. For the methylation assay, we subdivided patients according to percentage of BM blasts. Fifteen patients had <5% BM blasts (Table 1): four patients witefractory anemia (RA), one with RA with ringed sideroblasts, eight with refractory cytopenia with multilineage dysplasia, and two with unclassifiable MDS (MDS-U). Twenty-eight patients had ≥5% BM blasts but <20% BM blasts (Table 1): two patients with RA with excess blasts-1 and 11 with RA with excess blasts-2. Six patients with post-MDS AML were also enrolled in this study. Samples from all MDS or post-MDS AML patients were obtained before treatment. Among these, 13 MDS patient samples were analyzed sequentially before and after AZA treatment. Blood samples were collected weekly, and the response to AZA was evaluated using the International Working Group (IWG) response criteria for myelodysplasia.¹⁷ For NGS analysis, BM mononuclear cells were also used. Thirteen healthy volunteers (aged 35−58 years; mean, 48.9 years) served as the control group. This study was approved by the institutional review board of Tokyo Medical University (no. 1979). Written informed consent was obtained from all patients and volunteers prior to participation in accordance with the Declaration of Helsinki.

Table 1 Methylation index of patients with myelodysplastic syndrome and post-MDS AML

Isolation of DNA from whole blood, neutrophils, and mononuclear cells

For the methylation assay, DNA from either whole blood or the separated fraction was extracted using a robotic workstation (Magtration System 6GC; Precision System Science, Chiba, Japan) and an EZ1 DNA Blood 350 µl Kit (Qiagen, Valencia, CA, USA) according to the manufacturers’ instructions. To obtain a mononuclear lymphocyte-enriched fraction, heparinized peripheral blood cells were separated on a Ficoll-Hypaque density gradient. BM mononuclear cells containing blast fractions were also collected as described above. To collect a neutrophil-rich fraction, buffy coat was mixed with 0.2% methylcellulose to sediment the red blood cells. The leukocyte-enriched part was collected, washed with phosphate-buffered saline (PBS; pH 7.4), and resuspended in PBS and Hank’s balanced salt solution containing divalent cations (HBSS; Gibco, Grans Island, NY, USA).¹⁸

Methylation index measured by single-molecule methylation analysis

To determine global DNA methylation levels, we performed SMMA as previously described.¹⁶ The principle of this assay is based on two methodologies. One is the use of high-affinity methyl-CpG–binding domain protein 2 (MBD2) to bind methylated DNA, and the other is the use of fluorescence correlation spectroscopy, which determines the interaction of MBD2 with methylated DNA in the sample. We used 500 ng genomic DNA digested with Mse I (New England Biolabs, Beverly, MA, USA). The digested DNA was quantified by NanoDrop ND-1000 (Thermo Fisher Scientific, Waltham, MA, USA).¹⁶ To calculate the methylated DNA in the sample, we generated a standard curve using TAMRA-labeled MBD2, unlabeled MBD2, 0% methylated DNA (negative control), and 100% methylated DNA (positive control).¹⁶ Using this standard curve, we were able to assess the genome-wide DNA methylation status of whole blood, neutrophil-enriched fractions, and mononuclear fractions; the methylation level was expressed as the MI, which indicates the percentage of methylated DNA in the sample.

Next-generation sequencing analysis

Next-generation sequencing (NGS) analysis was performed as reported previously.¹⁹ Briefly, genomic DNA was extracted using a Gentra PureGene Cell Kit (Qiagen, Hilden, Germany). Whole exonic regions of 50 genes were amplified from 40 ng genome DNA from each patient sample using the GeneRead DNAseq Targeted Panel V2 (Human Myeloid Neoplasms Panel; Qiagen). We constructed a barcoded Illumina DNA library using the GeneRead DNA Library I Core Kit (Qiagen) and GeneRead Adapter I Set 12-Plex Kit (Qiagen); sequencing was performed on a MiSeq system (Illumina, San Diego, CA, USA) according to the manufacturer’s instructions. High-probability oncogenic mutations were extracted from annotated data in the available databases (ClinVar: https://www.ncbi.nlm.nih.gov/clinvar/and COSMIC: http://cancer.sanger.ac.uk/cosmic). All other non-polymorphous synonymous/non-synonymous mutations and intronic mutations were extracted individually by filtered population frequencies (<1%) using a sequencing data analysis tool. (VariantStudio, Illumina)

Statistical analysis

Data were expressed as mean ± standard deviation (SD). Mann–Whitney U and chi-square tests were used to determine statistical significance for comparisons between the control and test groups. Multiple groups were compared using one-way analysis of variance (ANOVA). GraphPad Prism software (version 5c for Macintosh; GraphPad Software Inc., La Jolla, CA, USA) was used for statistical analysis.

Results

Genome-wide DNA methylation status for untreated MDS

We first measured the genome-wide DNA methylation status of healthy volunteers. In healthy subjects, there was no significant difference in the MI between whole blood and neutrophils (82.8±5.9 vs 79.7±9.3; P=0.315); however, the MI of mononuclear cells was significantly lower than that of whole blood (P=0.0001) or neutrophils (P=0.0087). We therefore measured the methylation status of whole blood in 34 patients and of neutrophils and mononuclear cells in 28 patients when separated samples were available. The MI values of each fraction for all patients analyzed are shown in Table 1. To estimate the influence of blasts in the sample, MDS patients were subdivided into MDS-I (BM blasts <5%) and MDS-II (BM blasts >5%) groups (Table 1 and Figure 1A, B, C). A significant reduction in MI for the whole blood DNA sample was evident for patients with MDS-I (P=0.0001), MDS-II (P<0.0001), and post-MDS AML (P=0.0001) (one-way ANOVA; P<0.0001) compared with normal controls (Figure 1A). There was no significant difference in MI among the three groups, indicating that it did not reflect the percentage of blasts in the samples. This was also evident in DNA from neutrophil fractions (one-way ANOVA; P=0.0008) (Figure 1B), but not from mononuclear cell fractions (one-way ANOVA; P=0.0713) (Figure 1C). For instance, the MI of neutrophils was significantly lower in patients with MDS-I (P=0.001), MDS-II (P<0.0003), and post-MDS AML (P=0.0027) (Figure 1B), whereas that of mononuclear cells was not significantly decreased in patients with MDS-II (P=0.2622) (Figure 1C). There was no consistent pattern for MI levels found among MDS subtypes (P=0.3064), IPSS categories (P=0.2827), or IPSS-R categories (P=0.1940); however, we found that the MI was significantly higher for MDS patients in the very poor cytogenetic category according to the IPSS-R compared with patients in the good cytogenetics category (Figure 1D). We did not identify any correlation between the MI and hematologic features. No correlation was evident between the MI and neutrophils (P=0.6775) or mononuclear cells (P=0.9965) (data not shown).

Figure 1 Methylation index (MI) measured using a single-molecule methylation assay (SMMA). The percent of methylation DNA in the sample is shown. (A). MI of whole blood. (B). MI of the neutrophil fraction. (C). MI of mononuclear cells. (D). MI is significantly increased in MDS patients with a very poor cytogenetic score according to the Revised International Prognostic Scoring System.Abbreviations: MDS, myelodysplastic syndrome; post MDS-AML, post myelodysplastic acute myeloid leukemia.

Sequential analysis of methylation index for AZA-treated patients

Based on the results obtained from various sample fractions in the current study, we concluded that MI measurements in whole blood were the most robust and reliable for estimating genome-wide DNA methylation. Since MI was highly variable among the patients, regardless of the proportion of blasts, we analyzed the MI sequentially for 13 patients treated with at least four courses of AZA (Table 1). The clinical response to AZA and a summary of the mutation analysis related to myeloid neoplasia are shown in Table 2. Detailed information from the NGS analysis is shown in Table S1. Although the number of patients in the current study is small, six patients showed hematological improvement (unique patient number [UPN] 5, 7, 13, 15, 18, and 30). Among these six, five patients (all except for UPN 30) developed resistant to AZA and exhibited secondary failure after several courses of treatment. The remaining seven patients did not respond to AZA. TET2 mutation was detected in four patients, including two AZA responders, whereas TP53 mutation was detected in another four patients including one AZA responder.

Table 2 Clinical response to AZA in 13 patients whose methylation index was subsequently measured

To address the question of whether the MI before AZA administration could be a predictive marker for AZA response, we first compared the MI between responders and non-responders; however, there was no significant difference in the MI (66.6±6.3 vs 67.4±9.5; P=0.8689). A transient reduction in the MI at day 7 after AZA administration was noted in the majority of patients, but there was no significant difference in the MI between responders and non-responders (55.0±14.1 vs 52.0±16.2; P=0.7289).

We then focused on the sequential change in the MI for each case. Because the MI for each patient was highly variable, we attempted to calculate the normalized rate rather than the MI at a single point. The restoration rate of MI at day 28 (the day of next AZA treatment) is expressed as follows: ∆MI^day28 (MI^day28 − MI^day0). The ∆MI^day28 for AZA responders was significantly higher than that of non-responders (P=0.0352) (Figure 2A). In addition, ∆MI^day120 was significantly higher than that of non-responders (P=0.0251) (Figure 2B). It is notable that ∆MI^day120 was significantly higher than ∆MI^day28 for AZA responders, although there was no significant difference in ∆MI at day 28 and day 120 for non-responders (Figure 2C and D). This indicates that, in responders, the genome-wide DNA methylation level is gradually normalized after four courses of AZA. In contrast, there were no significant differences in ∆MI^day7 (MI^day7 − MI^day0) (P=0.6915), ∆MI^day14 (MI^day14 − MI^day0) (P=0.937), or ∆MI^day21 (MI^day21 − MI^day0) (P=0.6623) between AZA responders and non-responders. This indicates that the genome-wide DNA methylation level for responders tended to normalize before the next AZA administration (on day 28), whereas that for non-responders did not.

Figure 2 The restoration rate of methylation for AZA-treated patients. (A). ∆MI at day 28 (MIday28 – MIday0) for AZA responders was significantly higher than that of non-responders (P=0.0352). (B). ∆MI at day 120 was significantly higher than that of non-responders (P=0.0251). (C). For AZA responders, ∆MI at day 120 was significantly higher than at day 28 (P=0.0216). (D). For AZA non-responders, there was no significant difference between ∆MI at day 28 and at day 120 (P=0.2961).Note: *Represents a significant difference.Abbreviation: MI, methylation index.

Discussion

Genome-wide DNA hypomethylation and hypermethylation in the promoter region of tumor suppressor genes are hallmarks of cancer.^20,21 Several attempts have been made to determine the association between clinical response and the effect of HMA. Although many gene-specific methylations have been found to be associated with leukemia progression²² or effectiveness of HMA treatment, including AZA, it is likely that the effect of AZA is not restricted to aberrantly hypermethylated regions of single genes.^9,10 Shen et al proposed a methylation prognostic model for predicting response to therapy by analyzing the promoter region in 10 selected genes rather than in single genes.²³ In contrast, Grövdal et al recently demonstrated that AZA induced genome-wide DNA hypomethylation in progenitor cells obtained from MDS patients. A clear decease in active chromatin was observed, suggesting that molecular mechanisms underlying AZA resistance are more complex than reactivation of the silenced genes.¹¹ We previously showed marked genome-wide DNA demethylation together with constitutive activation of the ATM/BRCA1 pathway in AZA-resistant human leukemia cell lines.²⁴ We also found aberrant histone modification in two AZA-resistant human leukemia cell lines.²⁵ Therefore, we considered that the architecture of genomes, such as in genome-wide DNA demethylation, may play some role in the clinical efficacy of AZA in MDS patients.

To date, the analysis of DNA methylation has focused on the promoter region of genes, even when a comprehensive analysis was performed using the methylation array. Genome-wide DNA methylation can be measured by several methods, such as methylation-sensitive restriction enzymes, methyl-binding proteins, and anti-methylcytosine antibodies/^26,27 However, these techniques are complex and, in some cases, have been modified using bisulfate conversion. In the current study, we used an SMMA with 500 ng genomic DNA, which is less time-consuming than the methods described above: the fluorescence correlation spectroscopy measurement can be performed in 1–2 mins per assay, without bisulfate modification.¹⁶ Our previous studies demonstrated that the MI (originally designated the SMMA index) measured by SMMA correlates with the global methylation status investigated by the DNA methylation array.¹⁶

The binding of MBD2 to methylated DNA occurs not only in the CpG islands and its “island shores” in the promoter area, but also in the gene body; therefore, it is likely that the MI mainly reflects genome-wide DNA methylation in the gene body. Theoretically, it would be logical to perform assays using isolated CD34-positive MDS cells. In practical terms, however, this is time-consuming and methodologically complex because the percentage of blasts varies widely among patients. We found that the MI for MDS patients was significantly lower than that of normal individuals, regardless of the percentage of blasts, but significantly higher for MDS patients with very poor cytogenetics according to IPSS-R. These results suggest that the MI involving peripheral blood may reflect epigenetic changes in MDS cells in some cases, because it depends on the cells from which the DNA originated.

Regarding promoter-associated CpG site analysis, it has been reported that the methylation dynamics of tumor suppressor genes do not predict clinical response in MDS patients treated with AZA and entinostat.²⁸ Similar to previous reports,^23,28 neither MI at diagnosis nor MI after AZA therapy could predict clinical efficacy in the present study. A reduction in MI was observed in the majority of patients at day 7 after AZA administration; therefore, we could not distinguish responders from non-responders at day 7. We next focused on the difference between the initial MI and that after cycle 1 of therapy (ie, ∆MI^day28: MI^day28 − MI^day0). Eventually, we found that ∆MI^day28 could represent a marker for normalization of genome-wide DNA methylation. It is notable that the normalization of MI was more evident after cycle 4 of therapy (∆MI^day120).

Histone modification in accordance with genome-wide DNA methylation may have a key role in sustaining genome architecture, which regulates transcription.^10,11,29 Therefore, a genome-wide DNA methylation assay could function as a robust technique for managing MDS patients receiving AZA treatment. In clinical practice, it would be more reliable to use the MI value in combination with the methylation prognostic model based on the promoter analysis of 10 selected genes and mutation analysis of methylation-associated genes,²³ such as TET2.⁷ Because the number of MDS patients studied was too small to draw final conclusions for both primary and secondary AZA resistance, further studies using larger numbers of patients should be performed to clarify the efficacy of MI.

In conclusion, we established a genome-wide DNA methylation assay using whole blood to monitor the efficacy of AZA in MDS patients. Compared with healthy individuals, MDS patients had hypomethylated DNA, and AZA responders showed recovery of hypomethylated DNA during the next AZA administration. Therefore, it was possible to discriminate responders and primary AZA non-responders after the first cycle of therapy. Patients with secondary AZA failure showed a progressive decrease in MI, indicating that the MI recovery of peripheral blood after AZA could represent a non-invasive monitoring marker for AZA efficacy in MDS patients.