Baseline Patient Characteristics and Follow-Up
Of the 170 patients enrolled onto the phase III trial, we excluded 81 patients from the training cohort because no specimen was available. We compared patient clinicopathologic characteristics between the patients with tissue available and those without and found similar distributions for patient age; sex; hemoglobin, WBC, platelet, and marrow blast counts; cytogenetic categories; prior malignancy; prior treatment for MDS; and IPSS status. However, compared with patients who were excluded, there were slightly more patients with chronic myelomonocytic leukemia among the patients with tissue available (13 of 89 studied patients v one of the other 81 patients; P = .001). In the first validation cohort, we excluded 20 patients as a result of sample unavailability. The data for patients included and for those excluded had similar distributions with regard to demographic and clinicopathologic characteristics. lists the clinical and demographic characteristics of the patients in the three cohorts. Patients in the training cohort and those in the validation cohorts had similar characteristics, except that compared with the other two cohorts, the second validation cohort included a broad spectrum of patients with more low-risk MDS ().
Baseline Patient Demographics Clinical and Hematologic Characteristics
The median follow-up time was 13.7 months in the training cohort, 14.4 months in the first validation cohort, and 7.1 months in the second validation cohort. In the training cohort, we found no significant association between specimen sources and either overall survival (P = .80) or progression-free survival (P = .56).
Selection of a Panel of Genes for Methylation Analysis
We screened promoter CpG island methylation of 24 genes in a group of 24 patients with MDS; these genes were selected based on previous reports18–20
and our ongoing effort to identify hypermethylated genes in cancer by a genome-wide methylated CpG island amplification/representational difference analysis technique21,22
(details are summarized in the Data Supplement). After the initial screening, low levels of methylation were detected in 14 genes, and we excluded them from further study. Aberrant promoter CpG island methylation of 10 genes, including E-cadherin (CDH1
), N-cadherin (CDH13
), estrogen receptor-α (ER
α), oxidored-nitro domain-containing protein isoform 1 (NOR1
), nucleoplasmin 2 (NPM2
), oligodendrocyte lineage transcription factor 2 (OLIG2
), cyclin-dependent kinase 2B inhibitor (p15INK4B
), progesterone receptor A (PGRA
), progesterone receptor B (PGRB
), and PDZ and LIM domain 4 (RIL
), was found in more than 10% of the patients, and these genes were selected for further analysis (Data Supplement).
To confirm that the observed methylation differences were not merely measurement variation or cellular heterogeneity (such as analyses performed on unfractionated cells), we repeated bisulfite treatment, polymerase chain reaction, and pyrosequencing for six genes in 20 MDS patient samples, compared methylation for all genes between bone marrow and blood samples obtained at the same time from 25 patients, and compared methylation level between sorted CD34+ and CD3–/19– cells from the same patients (Data Supplement). The results from two replicate experiments were almost identical, with the Spearman correlation coefficient (r value) of 0.92 (n = 93). A significant correlation was found between methylation levels measured in bone marrow and blood (n = 117, r = 0.93), as well as between methylation levels measured in CD34+ and CD3–/19– cells (n = 13, r = 0.91).
Model of DNA Methylation Profiling for Predicting Survival in the MDS Training Cohort
Among the 89 patients in the training cohort, promoter methylation (> 15%) was found in 7% at ERα, 15% at CDH1, 15% at NOR1, 20% at NPM2, 21% at CDH13, 23% at p15INK4B, 41% at OLIG2, 45% at PGRB, 45% at PGRA, and 70% at RIL. Using methylation level as a continuous variable and analyzing the correlation between each gene by Spearman correlation analysis, we found significant positive associations among methylation of different genes within the same patients (Data Supplement). These results indicate concordant methylation in a subset of patients (Data Supplement) and hence that combining multiple gene methylation profiles could provide greater accuracy than individual markers in predicting clinical outcomes. Therefore, we used average z scores based on methylation of all genes to compare clinical characteristics and build predictive models of survival for individual patients. There were no significant associations between average methylation z score and age, sex, hemoglobin level, absolute neutrophil count, platelet count, bone marrow blast percentage, IPSS status, French-American-British type, or cytogenetics ().
Correlation Between Average Methylation of 10 Genes (by z score) and Clinical Features
To evaluate the prognostic significance of the methylation profile, we performed both univariate and multivariate analyses. In the univariate analysis, only methylation was significantly associated with shorter overall survival (hazard ratio [HR] = 1.68; 95% CI, 1.01 to 2.80; P = .05). Methylation was most significantly associated with progression-free survival (HR = 1.03; 95% CI, 1.00 to 1.06; P = .06 for age; HR = 1.78; 95% CI, 1.01 to 3.13; P = .05 for IPSS status; and HR = 1.88; 95% CI, 1.15 to 3.07; P = .01 for methylation). shows Kaplan-Meier curves of overall survival and progression-free survival according to the baseline levels of methylation. The median overall survival time was 12.3 months in patients with high methylation compared with 17.5 months in patients with low methylation (P = .04; A), and the median progression-free survival time was 6.4 months in patients with high methylation compared with 14.9 months in patients with low methylation (P = .01; E). In multivariate analysis, methylation remained the only independent predictor of overall survival (HR = 1.68; 95% CI, 1.0 to 2.81; P = .05) and progression-free survival (HR = 1.95; 95% CI, 1.18 to 3.21; P = .009; ). The effects of methylation on overall and progression-free survival were similar in patients who received decitabine and those who were on supportive care (Data Supplement).
Fig 1. Kaplan-Meier survival estimates of overall and progression-free survival in patients with myelodysplastic syndromes. Overall survival in (A) training cohort, (B) first validation cohort, (C) second validation cohort, and (D) all patients. Progression-free (more ...)
Multivariate Cox Proportional Hazards Analysis of Overall and Progression-Free Survival
Validation of the DNA Methylation Prognostic Model in Two Independent Cohorts
As shown in the Data Supplement, concordant methylation of the 10 genes was also observed in the two validation cohorts. No significant association was found between average methylation z score at baseline and clinical variables in the first validation cohort (). In the second validation cohort (consecutive series that included patients with low-risk MDS), significantly lower levels of methylation were found in patients with lower bone marrow blast percentage, lower risk IPSS status, and refractory anemia with ringed sideroblasts.
shows overall survival and progression-free survival according to baseline levels of methylation in the validation cohorts. Methylation was significantly associated with progression-free survival in both cohorts (P = .03 and P = .03; F and G). We found significant association between methylation and overall survival in the first validation cohort (P = .05; B); however, the difference in overall survival was not statistically significant in the second validation cohort (P = .097; C). Because the second validation cohort comprised more patients with low-risk MDS, one possible explanation is that the relatively short follow-up in these patients may not be sufficient to detect significant differences. We then performed separate analyses in patients with or without low-risk MDS and found that methylation was not significantly associated with overall or progression-free survival in patients with low-risk MDS during the follow-up period (Data Supplement). For patients with other than low-risk MDS, methylation was confirmed to predict shorter overall and progression-free survival (P = .035 and P = .012, respectively; Data Supplement). In the original and validation cohorts combined, patients with high methylation had a worse survival than patients with less methylation (14.3 v 22.8 months, respectively; P = .014 for overall survival and 10.2 v 18.3 months, respectively; P = .0017 for progression-free survival; D and H).
Multivariate analyses were performed using data from the two validation cohorts or from all patients (). As in the training cohort, after adjusting for age, sex, and IPSS status, methylation was independently associated with overall survival (HR = 1.83; 95% CI, 1.23 to 2.71; P = .0027) and progression-free survival (HR = 1.87; 95% CI, 1.31 to 2.67; P = .0006).
Because hypomethylating agents may alter the natural history of disease, we then analyzed the correlation between baseline methylation score, clinical characteristics, and standard prognosis separately according to whether or not the patients received treatment with either decitabine or azacitidine. Including only patients who did not receive hypomethylating agents yielded the same results as including all patients (Data Supplement lists clinical characteristics and shows Kaplan-Meier analysis; lists results of multivariate survival analysis).
Methylation-Based Prediction Within the Same Cytogenetic Risk Groups
Taking advantage of the large data set, we were able to determine the impact of the DNA methylation prognostic model in patients within similar cytogenetic subgroups. Cytogenetic risk groups were defined by the IPSS as good, intermediate, and poor. Among patients who were in cytogenetic intermediate- and poor-risk groups, the methylation model was a significant predictor of survival (B and C show overall survival, and E and F show progression-free survival). Interestingly, survival curves showed no significant differences between high methylation and low methylation groups in patients with good cytogenetic risk (A and D).
Fig 2. The DNA methylation prognostic model and cytogenetic risk groups. The Kaplan-Meier estimates show survivals for groups of patients with cytogenetic good risk (A: overall survival; D: progression-free survivals), intermediate risk (B: overall survival; (more ...)
Correlation Between DNA Methylation and Treatment Responses
To determine whether DNA methylation could predict responses to decitabine treatment, we compared DNA methylation at baseline with clinical responses in 163 patients enrolled onto the phase III and phase II trials. Nine percent of the 89 patients from the phase III trial achieved complete remission (CR) or partial remission (PR), and 47% of the 75 patients from the phase II trial achieved CR or PR. Although the patient groups are comparable between these two trials (), the difference in clinical responses may be related to the different dose-intensity of decitabine given and to the median number of decitabine courses, which was six in the phase II trial and three in the phase III trial.5,6
Treatment response was not correlated with either methylation of single genes or a combination of all genes (Data Supplement).
We then analyzed methylation changes at multiple time points for correlation with response in 34 patients from the phase III trial. These patients were selected solely based on tissue availability, and all available tissues were evaluated. Of 14 patients who received decitabine, two patients achieved CR, three achieved PR, four achieved hematologic improvement (HI), four had stable disease (SD), and one had progressive disease (PD). Of 20 patients on supportive care, two patients achieved HI, six had SD, and 12 had PD. Methylation levels at each time point were averaged across the 10 genes. At the latest available time point (> 4 months on therapy), we found that methylation decreased by 11.2% in patients on decitabine but increased by 20.1% in patients on supportive care (P = .04; ). A greater decrease was observed in patients with CR or PR (40.6% ± 15.7%) compared with HI (9.8% ± 13.2%). In contrast, methylation increased by 15.4% in patients with SD and by 27.2% in patients with PD (P = .02; ).
Fig 3. Methylation changes at multiple time points after treatment. Average methylation changes (before and after 4 months on therapy) were compared between patients treated with decitabine (DAC) and supportive care (SUP). Methylation decreased by 11.2% in patients (more ...)