Patients entered onto the study were required to have a diagnosis of Philadelphia chromosome (Ph) –positive CML and to have evidence of resistance or intolerance to imatinib. Resistance to imatinib was defined by elevated WBC or platelet counts while on therapy, lack of any cytogenetic response after 12 months of therapy (Ph, 100%), appearance of accelerated or blastic phase features while on therapy, reappearance of the Ph-positive clones after an initial complete cytogenetic response to imatinib, or an increase in Ph-positive cells by 30% or more while on therapy. Intolerance to imatinib was defined by the development of life-threatening adverse effects requiring discontinuation of therapy.

Other eligibility criteria were adequate performance status (Eastern Cooperative Oncology Group performance status ≥ 2) and adequate cardiac (New York Heart Association class III-IV excluded) and hepatorenal functions (creatinine < 2 mg/dL; bilirubin < 2 mg/dL, and hepatic enzymes < 2 × upper limit of normal). All patients gave written informed consent indicating that they were aware of the investigational nature of the study, in keeping with the policies of the M. D. Anderson Cancer Center (Houston, TX). Consent was also obtained for the (optional) correlative studies, which included the collection of additional blood samples.

The initial plan was to treat patients at 15 mg/m² daily IV over 1 hour for 10 days (5 days on, 2 days off, 5 days on), approximately every 6 weeks as indicated by follow-up counts and marrow studies. After the first 10 patients with CML-CP were treated, the initial dose was reduced to 10 mg/m² daily IV over 1 hour for 10 days (5 days on, 2 days off, 5 days on) because of myelosuppression. The starting dose was not changed for patients with AP or BP. After the first cycle, patients were dose-adjusted for toxicity. If ≥ grade 3 nonhematologic toxicity was felt attributable to the drug, the patients were taken off-study. Patients experiencing grade 3 or worse hematologic toxicity were evaluated for persistent disease. If myelosuppression was attributed to disease, the treatment was continued at the same dose level. If myelosuppression was attributed to the drug, subsequent cycles were given at a dose that was lowered by 5 mg/m²/d of treatment. The use of hydroxyurea was allowed for the first two cycles of therapy, at the discretion of the investigators. Prophylactic antibiotics were recommended for all patients who became neutropenic. Growth factors were not routinely used, but were allowed in patients who developed fever and neutropenia. In general, patients were treated every 6 weeks for two cycles, and treatment was continued beyond two cycles unless clear evidence of lack of response or disease progression was observed.

Response criteria were as previously described.^15,16 A complete hematologic response (CHR) required disappearance of all signs and symptoms related to disease, normalization of peripheral counts (absolute neutrophil count ≥ 10⁹/L, platelet count ≥ 100 × 10⁹/L), and a normal bone marrow morphology with ≤ 5% marrow blasts. A partial hematologic response was defined as a CHR, but with persistence of peripheral immature cells (< 5% myelocytes + metamyelocytes), or persistent splenomegaly or thrombocytosis that were reduced by 50% or more from pretreatment levels. A hematologic improvement was defined as a return to a second CP, referring to disappearance of accelerated-BP criteria. Cytogenetic responses were measured separately by G-banding in at least 20 metaphases and were categorized by the degree of suppression of Ph-positive cells as a complete cytogenetic response (Ph, 0%), partial cytogenetic response (Ph, 1% to 34%), or minor cytogenetic response (Ph, 35% to 95%). Remission was calculated from date of first response until relapse. Survival was calculated from start of therapy until death from any cause.

Peripheral blood was collected from consenting patients treated as part of this protocol. Whenever possible, blood was obtained before or on the first day of treatment (day 0 or 1), at the end of the first week or at the beginning of the second week (days 5 to 7), at the end of the second week (day 12), and at recovery of counts. DNA was isolated from peripheral-blood samples after ficoll separation of mononuclear cells using standard phenol-chloroform extraction methods. After extraction, DNA was modified with sodium bisulfite. This induces deamination of unmethylated cytosines converting unmethylated CpG sites to UpG without modifying methylated sites. Bisulfite treatment of genomic DNA was performed as described.¹⁷ Two μg of DNA were used. DNA was denatured in 0.2 N NaOH at 37°C for 10 minutes and incubated with 3 M Na-bisulfite at 50°C for 16 hours. DNA was then purified using the Wizard cleanup system (Promega, Madison, WI) and desulfonated with 0.3 N NaOH at 25°C for 5 minutes. DNA was then precipitated with ammonium acetate and ethanol, washed with 70% ethanol, dried, and resuspended in H₂O.

We used a LINE1 repetitive element bisulfite/pyrosequencing assay to estimate global methylation.¹⁸ After bisulfite treatment, a 50 μL polymerase chain reaction (PCR) was carried out in 60 mmol/L Tris-HCl pH = 8.8, 15mM ammonium sulfate, 0.5 mmol/L MgCl2, 1mM dNTP mix, and 1 unit of Taq polymerase. The PCR primers used were: 10 pmol of 5′-TTTTTTGAGTTAGGTGTGGG-3′, 1pmol of 5′-GGGACACCGCTGATCGTTTATCTCACTAAAAAATACCAAACAA-3′, and 10 pmol of a universal biotinylated primer 5′-GGGACACCGCTGATCGTTTA-3.′ PCR cycling conditions were 95°C for 30 seconds, 50°C for 30 seconds, and 72°C for 30 seconds for 35 cycles. The PCR product was purified and methylation quantitated using the PSQ HS 96 Pyrosequencing System (Pyrosequencing Inc, Westborough, MA). The sequencing primer for pyrosequencing was 5′-GGGTGGGAGTGAT-3′.

We also used bisulfite-pyrosequencing¹⁹ to analyze methylation of the p15 gene. A 50-μL PCR was carried out in 60 mmol/L Tris-HCl pH = 8.8, 15mM ammonium sulfate, 2 mmol/L MgCl2, 1mM dNTP mix, and 1 unit of Taq polymerase. PCR primers used were 10 pmol of 5′-GTTTTTTTTTAGAAGTAATTTA-3′, 1 pmol of 5′-GGACACCGCTGATCGTTTATCCTTCTACGACTTAAAACC-3′, and 10 pmol of a universal biotinylated primer 5′-GGGACACCGCTGATCGTTTA-3′. PCR cycling conditions were melting temperature of 95°C for 30 seconds, annealing for 45 seconds at a temperature of 50°C for three cycles, 48°C for four cycles, 46°C for four cycles, 44°C for four cycles, 42°C for 43 cycles, and extension temperature of 72°C for 45 seconds. The sequencing primer for pyrosequencing was 5′-TTTTTAGAAGTAATTTAGG -3′.

Simple descriptive statistics were used in the analysis of the clinical efficacy of decitabine. Survival was computed using Kaplan-Meier curves and the provided P values are two-sided. A P value of .05 was considered significant. The methylation data generated was expressed as a percentage and showed a normal distribution. The data was summarized (mean, median, standard error of the mean) using the Excel software (Microsoft, Redmond, WA). T tests or paired t tests were used to compare methylation at different time points or in different groups, as appropriate. Provided P values are two-sided. A P value of .05 was considered significant.

A total of 35 patients were registered on this study. Their clinical characteristics and diagnoses are summarized in Table 1

Table 1. Patient Characteristics

Previous studies have repeatedly shown that decitabine works slowly, and that multiple cycles are required for optimal response.^5,15 For this reason, the protocol recommended that patients complete at least two cycles of therapy, regardless of response. Thirty of the 35 patients received at least two cycles of therapy. In five patients, only one cycle was administered because of disease progression (two patients), early death (one patient) and noncompliance (two patients).

Responses are summarized in Table 2

Table 2. Summary of Responses

While some of the responses were of short duration, there were responses that lasted a year or more. To explore the clinical significance of achieving a response in this setting, we used Kaplan-Meier curves to plot the survival of patients with or without response to decitabine. Overall, patients in BP had a median survival of 4 months, patients in AP had a median survival of 12 months, and the median survival of patients in CP has not been reached yet. Because the majority of patients with CP responded and had a relatively favorable prognosis, we excluded them from analysis. There were 23 patients with CML AP or BP enrolled in the study, with 13 responders and 10 nonresponders. As shown in Figure 1

Fig 1. Kaplan-Meier curve of survival in chronic myelogenous leukemia accelerated phase or blastic phase by response to decitabine. Straight line: responders; dashed line: nonresponders.

A summary of toxicities is shown in Table 3

Table 3. Grade 2 to 4 Toxicities Observed

We used a bisulfite/pyrosequencing assay to measure DNA methylation across LINE1 repetitive elements in peripheral-blood mononuclear cells as a surrogate marker for global methylation. We first evaluated the reproducibility of the assay; 105 samples with values ranging from 32.1% to 81.3% were assayed in duplicate. The correlation between duplicates was R = 0.82 (P < .001). We next evaluated the stability of LINE1 methylation in CML over time. Fourteen patients had two separate samples independent of decitabine therapy assayed for LINE1 methylation at intervals ranging from 1 day to 2 years. The mean of sample 1 was 71.1%, and the sample 2 mean was 71.0% (P = .9 by paired t test for a difference). The correlation between the two independent samples from the same patient at different times (unrelated to decitabine) was R = 0.84 (P < .001).

We used this LINE1 assay to study global methylation in the patients entered onto this trial. Pretreatment samples were available on 30 patients. LINE1 methylation had a mean ± standard error of the mean of 71.4% ± 1.1% (range, 56.5% to 85.8%). There was no difference in methylation between patients in CP (72.1% ± 1.7%), AP (70.3% ± 1.3%), or BP (74.4% ± 4.0%). On-treatment samples were available for 22 patients for the first cycle. Typically, samples were collected before or on the first day of treatment, at the end of week 1 of therapy (day 5), at the end of week 2 of therapy (day 12), and at recovery of counts (days 30 to 50). In these 22 patients, methylation decreased from 71.3% ± 1.4% to 60.7% ± 1.4% at day 5 (P < .001) and 50.9% ± 2.4% at day 12 (P < .001 compared with day 5), increasing back to 66.5% ± 2.7% (P = not significant compared to pretreatment) at the time of recovery (Fig 2

Fig 2. Methylation dynamics following cycle 1 (straight line) or cycle > 1 (dashed line) of decitabine therapy.

Fig 3. Methylation dynamics in four individual patients (A, B, C, D). Patients were started on treatment on day 1, and subsequent cycles are indicated by arrows.

Most patients on study received more than one cycle, and samples were available for analysis from a total of 47 cycles administered to 22 patients (19 cycle 2, 12 cycle 3, 16 cycle > 3). This allowed us to determine the dynamics of hypomethylation after subsequent cycles. Overall, the decreases in methylation in cycles 2 and greater were similar to cycle 1. Methylation decreased from 67.4% ± 1.1% pretreatment to 55.2% ± 1.1% at day 5, 49.5% ± 1.7% at day 12 and recovered to 67.4% ± 1.0% at count recovery (Fig 2). The study allowed dose reduction for toxicity in subsequent cycles. Of the 47 cycles analyzed, 24 were given at full dose and 23 at a dose reduced by one third (10 mg/m² IV daily for 5 days per week on 2 consecutive weeks). Interestingly, the lower dose resulted in lower degrees of hypomethylation at days 5 (57.0% ± 1.7% v 53.4% ± 1.4%; P = .1) and at day 12 (54.3% ± 2.4% v 45.2% ± 2.1%; P < .008; Fig 4

Fig 4. Methylation dynamics for cycles > 1 given at standard dose or reduced dose (reduced by one third).

We next analyzed the relationship between induction of hypomethylation during cycle 1 and response in CML. Of the 22 patients with samples available, 10 had partial or complete hematologic responses, while 12 had no responses. Figure 5

Fig 5. LINE1 methylation dynamics after cycle 1 of decitabine in patients who subsequently responded (straight line) or did not respond (dashed line) to therapy.

One hypothesis to explain the paradoxical correlation between more profound hypomethylation and a lack of response to decitabine is a cell-death mechanism of response, whereby the most hypomethylated cells die rapidly in patients sensitive to therapy, while cells resistant to therapy can withstand higher degrees of hypomethylation without cell death. If this hypothesis were correct, by extension, resistance to decitabine would be distal to hypomethylation induction (ie, nonpharmacologic). This hypothesis can be tested by examining the relationship between hypomethylation induction and maintenance of response or relapse during therapy. Of the 47 cycles administered subsequent to cycle 1, for which samples are available, 19 were associated with maintained or improved response, 17 with consistent lack of response, and 11 with a relapse on therapy (ie, response in the previous cycle, loss of response and progression during the current cycle). We therefore examined hypomethylation induction in these three situations. In the cycles where remissions were maintained, the relative decrease in methylation was 10.9% ± 3.3% at day 5 and 18.6% ± 4.8% at day 12, compared with nonresponders at 22.8% ± 2.6% at day 5 (P = .008) and 30.7% ± 3.9% at day 12 (P = .06), respectively. Remarkably, in patients who responded and subsequently lost their response while on therapy, the relative decrease in methylation was intermediate, with a decrease of 17.5% ± 1.9% at day 5 and 25.8% ± 5.6% at day 12. In this latter group, the decreases at days 5 and 12 in the relapsing cycle were all the more remarkable when one compares them to the decreases seen during cycle 1 (for the same patients): 11.9% ± 3.4% at day 5 and 25.1% ± 5.0% at day 12. Thus, hypomethylation induction was similar or higher in relapse cycles compared to initial cycles.

Finally, we studied interactions between global methylation or methylation of the p15^INK4b gene and response in these patients. As measured by LINE1, there was no correlation between global methylation and response; thus, LINE1 methylation had a mean of 69.3% in responders compared to 73.2% in nonresponders (P = .1). p15^INK4b was densely methylated (methylation density > 10%) in only three patients, precluding meaningful correlations with response. Nevertheless, of these three patients, one in BP was nonassessable, one in AP did not respond, and another patient in AP had a transient, minimal response (hematologic improvement).