Myeloma Is Characterized by Stage-Specific Alterations in DNA Methylation That Occur Early during Myelomagenesis

Despite therapeutic advances, multiple myeloma (MM) remains incurable and needs newer insights into the pathogenic mechanisms that cause this disease. Gene expression profiling has been used extensively in myeloma and has led to the development of a risk model that remains robust even in the age of newer therapies (1). The use of gene expression profiling by different groups has also led to the identification of distinct molecular subgroups that show defined clinical features (2, 3). However, pathogenic mechanisms driving these differences in gene expression have not been well described, especially for early stages in myelomagenesis (4).

Recent studies of the epigenome and in particular the methylome have shown that cancer is characterized by widespread epigenetic changes. These changes lead to altered gene expression that can result in activation of oncogenic pathways. Furthermore, methylome profiling has shown greater prognostic ability than gene expression profiling in acute myeloid leukemia (AML) and has led to identification of newer molecular subgroups with differences in overall survival (5). We have shown that methylome profiling is able to identify changes that occur early during carcinogenesis in solid tumors such as esophageal cancer (6). These studies have been conducted with the use of the HpaII tiny fragment enrichment by ligation–mediated PCR (HELP) assay, which is a genome-wide assay that provides a reproducible analysis of the methylome that is not biased toward CpG islands (7). Analyzing the methylome of myeloma can help identify changes that can define disease subsets and lead to identification of newer therapeutic targets. Recent sequencing studies in myeloma have also revealed mutations in enzymes that are involved in epigenetic machinery, again reinforcing the need to study the epigenetic alterations in this disease (8). We have conducted a genome-wide analysis of changes in DNA methylation in monoclonal gammopathy of uncertain significance (MGUS), newly diagnosed MM, as well as relapsed MM, and compared them with normal plasma cell controls. We used a modification of the HELP assay (nano-HELP) that was developed to work with low amounts of DNA to interrogate the methylome of sorted CD138⁺ cells from patient samples. We report that widespread alterations in DNA methylation are seen in myeloma and have the power to discriminate between MGUS and new/relapsed cases. We report that hypomethylation is the predominant early change during myelomagenesis that is gradually transformed to hypermethylation in relapsed cases, thus providing the epigenetic basis for the differentiation between these different stages of myeloma.

Bone marrow aspirates were obtained with signed informed consent from 11 patients with MGUS, 4 patients with smoldering myeloma (SMM), 13 patients with newly diagnosed myeloma (NEWMM), 16 patients with relapsed myeloma (REL), including 2 patients with serial samples, and 2 patients in clinical complete remission (REM) who are followed at the myeloma clinic at the Robert H. Lurie Cancer Center at Northwestern University. All specimens underwent CD138⁺ microbead selection (Miltenyi Biotec). Purity was confirmed by flow cytometry staining for CD38 and CD45. Samples with a purity <90% by flow cytometry were not used for this study. Eight samples from normal donors without known malignancies were obtained commercially from AllCells (Emeryville, CA). All normal donor samples had a purity of ≥85%.

Genomic DNA was extracted using the Gentra PureGene kit using 1 × 10⁵ to 5 × 10⁵ cells. DNA methylation analysis was done using the HELP assay, as described previously (7, 9).

In brief, 100 ng genomic DNA (nano-HELP modification) was digested overnight using the two isoschizomers HpaII and MspI. The digestion products were then purified once with phenol-chloroform. After overnight adapter ligation, the HpaII and MspI products were amplified in a two-step protocol, as previously described (7). HpaII- and MspI-generated genomic fragments between 200 and 2000 bp in length were labeled with either Cy3- or Cy5-labeled random primers and then cohybridized onto a human (HG17) custom-designed oligonucleotide array (50 mers) covering 25,626 HpaII-amplifiable fragments (HAFs) annotated to 14,214 gene promoters. HAFs are defined as genomic sequences contained between two flanking HpaII sites found 200–2,000 bp apart. Each HAF on the array is represented by a probe set consisting of 14–15 individual probes, randomly distributed across the microarray slide. Methylation data presented in this article have been deposited in the National Institutes of Health Gene Expression Omnibus (National Center for Biotechnology Information, http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE 43860.

All microarray hybridizations were subjected to extensive quality control using the following strategies. First, uniformity of hybridization was evaluated using a modified version of a previously published algorithm (10, 11) adapted for the NimbleGen platform, and any hybridization with strong regional artifacts was discarded and repeated. Second, normalized signal intensities from each array were compared against a 20% trimmed mean of signal intensities across all arrays in that experiment, and any arrays displaying a significant intensity bias that could not be explained by the biology of the sample were excluded. Signal intensities at each HpaII-amplifiable fragment were calculated as a robust (25% trimmed) mean of their component probe-level signal intensities. Any fragments found within the level of background MspI signal intensity, measured as 2.5 mean-absolute-differences above the median of random probe signals, were categorized as failed. These failed loci therefore represent the population of fragments that did not amplify by PCR, whatever the biological (e.g., genomic deletions and other sequence errors) or experimental cause. In contrast, methylated loci were so designated when the level of HpaII signal intensity was similarly indistinguishable from background. PCR-amplifying fragments (those not flagged as either methylated or failed) were normalized using an intra-array quantile approach wherein HpaII/MspI ratios are aligned across density-dependent sliding windows of fragment size-sorted data. The log₂(HpaII/MspI) was used as a representative for methylation and analyzed as a continuous variable. For most loci, each fragment was categorized as either methylated, if the centered log HpaII/MspI ratio was generally <0, or hypomethylated, if, in contrast, the log ratio was >0.

Validation of the findings of the HELP assay was carried out by MALDI-TOF mass spectrometry using EpiTyper by MassArray (Sequenom, CA) on bisulfite-converted DNA, as previously described (6). Validation was done on all samples with sufficient available DNA. MassARRAY primers were designed to cover the flanking HpaII sites for a given HAF, as well as for any other HpaII sites found up to 2000 bp upstream of the downstream site and up to 2000 bp downstream of the upstream site, to cover all possible alternative sites of digestion within our range of PCR amplification.

Unsupervised clustering of HELP data by hierarchical clustering was performed using the statistical software R version 2.6.2. A two-sample t test was used for each gene to summarize methylation differences between groups. Genes were ranked on the basis of this test statistic, and a set of top differentially methylated genes with an observed log fold change of >1 between group means was identified. Genes were further grouped according to the direction of the methylation change (hypomethylated versus hypermethylated), and the relative frequencies of these changes were computed among the top candidates to explore global methylation patterns. Validation with MassArray showed good correlation with the data generated by the HELP assay. MassArray analysis validated significant quantitative differences in methylation for differentially methylated genes selected by our approach.

Screening for mutations in the DNMT3A catalytic domain was carried out using direct genomic sequencing. PCR primers were designed to amplify and sequence coding exons 18–23 (available upon request). For each PCR, 20 ng genomic DNA was used for PCR amplification, followed by purification using Montage Cleanup kit (Millipore, Billerica, MA). Sequencing was performed using ABI 3730xl DNA analyzer (Applied Biosystems, Foster City, CA). PCR was carried out using the following: 94°C, 4 min, followed by 35 cycles of 94°C, 30 s, specific annealing temperature for each primer, 30 s, and extension 72°C, 30 s, final 72°C, 5 min. Results were compared with reference (NM_175629) and SNP databases (dbSNP; http://www.ncbi.nlm.nih.gov/projects/SNP). If chromatograms suggested a possible mutation, bidirectional resequencing was performed.

The Ingenuity Pathway Analysis software (IPA, Redwood City, CA) was used to determine biological pathways associated with differentially methylated genes, as performed previously (6, 10). Enrichment of genes associated with specific canonical pathways was determined relative to the Ingenuity knowledge database for each of the individual platforms and the integrated analysis at a significance level of p < 0.01. Biological networks captured by the microarray platforms were generated using IPA and scored based on the relationship between the total number of genes in the specific network and the total number of genes identified by the microarray analysis. The list of differentially methylated genes was examined for enrichment of conserved gene-associated transcription factor (TF) binding sites using IPA as well as other published gene sets available through the Molecular Signatures Database (MSigDB) (12).

We obtained gene expression data from the Arkansas datasets GSE5900 and GSE2658 (4) from National Center for Biotechnology Information’s Gene Expression Omnibus database. The datasets were quantile normalized to ensure cross-study comparability, based on our previous approach (13, 14). Analyses were performed using SAS (SAS Institute, Cary, NC) and the R language (http://www.r-project.org). The final database had 22 normal controls, 44 MGUS, and 559 new MM samples.

Cell lines U266 and RPMI8226 were purchased from American Type Culture Collection (Manassas, VA). Cell lines MM1.S and MM1.R were provided by S. Rosen (Northwestern University, Chicago, IL). All cell lines were cultured in RPMI 1640 medium (Invitrogen, Carlsbad, CA) supplemented with 10% heat-inactivated FBS and penicillin (100 U/ml), streptomycin (100 mg/ml), and 4 mM glutamine. Cells were maintained at 37°C and humidified with 95% air and 5% CO₂ for cell culture. For the viability assays, the cells were cultured in 0.5, 1, and 5 μM decitabine (Sigma-Aldrich) for 5 d. Decitabine was added to the culture daily; DMSO was served as control. Viability was measured on day 6 using MTS (Promega), according to the manufacturer’s instruction.

Extensive study of gene expression profiling of myeloma cells has led to several molecular models that allow the classification of myeloma in different risk categories, which has led to significant improvements in treatment strategies and outcomes (1, 15). Epigenetic alterations including aberrant DNA methylation can regulate gene expression and can also be used as better prognostic markers in tumor models (5). To determine whether there was aberrant differential methylation in different stages of multiple myeloma, we used the nano-HELP assay (7) on CD138⁺ selected bone marrow cells from 11 patients with monoclonal gammopathy of uncertain significance (MGUS), 4 patients with SMM, 13 patients with NEWMM, and 16 patients with REL, which also included 2 patients with serial samples. We also included 2 myeloma patients in REM. CD138⁺ selected bone marrow plasma cells from 8 healthy donors (normal [NL]) served as controls. Clinical characteristics of the 48 patients with plasma cell dyscrasias are listed in Supplemental Table 1.

At first we performed an unsupervised analysis of the generated methylation profiles with hierarchical clustering and the nearest shrunken centroid algorithm. Both methods showed that the healthy controls formed a tight cluster that was distinct from samples with abnormal plasma cells, demonstrating epigenetic dissimilarity between these groups (Fig. 1). Interestingly, MGUS samples also clustered in a distinct group, suggesting definite alterations in DNA methylation that occur early during myelomagenesis. The new (NEW) and relapsed (REL) myeloma cases demonstrated great epigenetic dissimilarity to the NL and MGUS samples and separated in two subgroups with the majority of NEWMM in one and the majority of REL samples in the other group (Fig. 1A). These findings show that methylation profiling can differentiate between NL, MGUS, NEWMM, and REL, in an unsupervised manner. Included in our analysis were two sets of consecutive samples taken from the same patient at different time points. In the first case, the initial sample (REL15) clusters with the group of relapsed myeloma and the posttreatment sample clusters with the NEWMM samples (REL9). The second case included samples taken at distinct time points without any treatments in the interval. Both samples cluster together in the REL group (REL18, REL6) and thus show the biological validity of our analysis.

After demonstrating global epigenetic dissimilarity between normal and myelomatous plasma cells, we wanted to determine the specific differences in DNA methylation between the different stages of myeloma. We performed a supervised analysis comparing the MGUS, NEWMM, or REL cases versus control samples. Volcano plots comparing the difference of mean methylation of all individual loci were plotted against the significance (log [p value] based on t test) of the difference (Fig. 2). Stringent cut-offs comprising an absolute fold change of ≥2 (= log[HpaII/MspI] > 1) and a p value <0.005 were used to identify differentially methylated loci. All probes thus identified were significant after multiple testing with Benjamini-Hochberg correction with a false discovery rate of <5%. We observed that majority of differentially methylated loci in MGUS were found to be hypomethylated when compared with normal plasma cells (Fig. 2A, Supplemental Table 2A, 2B). Hypomethylation was the predominant change also seen in NEWMM samples, yet this was less pronounced than in MGUS (Fig. 2B, Supplemental Table 2C, 2D). Interestingly, in cases of relapsed myeloma, we observed increased hypermethylation, demonstrating that there was a trend toward predominant hypomethylation in MGUS to predominant hypermethylation in REL (Fig. 2C, 2D, Supplemental Table 2E, 2F). Overall, differential methylated genes at the transition between stages are depicted in Supplemental Fig. 1A. When analyzing this differential methylation at the transition from one disease state to the next (i.e., NL to MGUS, MGUS to NEWMM, and NEWMM to REL), we found that of the 2963 unique genes that were hypomethylated at the transition from NL to MGUS, 2472 became hypermethylated at the transition of MGUS to NEWMM. Of those, 2 were also significantly hypermethylated at the transition from NEWMM to REL, albeit with a q value >0.05 (p < 0.05, Supplemental Fig. 1B). A list of the differentially methylated genes between the different disease states can be found in Supplemental Table 2A–K.

The HELP assay has been validated quantitatively in many studies (5, 6), and we also validated our findings by analyzing the methylation status of differentially methylated genes, ARID4B and DNMT3A, by bisulfite Massarray analysis. Examination of the promoter regions of these genes demonstrated a strong correlation of quantitative methylation values obtained from MassArray with the findings of HELP assay, demonstrating the validity of our findings (Supplemental Fig. 2).

We next analyzed the biological pathways that were associated with differentially methylated genes in myeloma and observed that pathways involved in cell proliferation, gene expression, cell cycle, and cancer were significantly involved by aberrantly methylated genes in MGUS, NEWMM, as well as relapsed cases of MM (Table I). The genes affected by aberrant methylation of these pathways included many genes that were hypomethylated (CEBPalpha, ILs, and their receptors, GFI1, etc.) and hypermethylated (EZH2, various HOX members, SOX, and WNT family members) that have not been previously implicated in myelomagenesis.

Aberrant methylation was not distributed randomly across chromosomes. Differentially methylated HpaII fragments showed significant regional differences with positional association with chromosomes 1, 9, 11, 15, 17, 19, 20, and 21. Furthermore, to determine whether these aberrantly methylated regions shared any common DNA elements, we performed a search for TF binding sites enriched in these loci. Significant overrepresentation of binding sites for TFs that have been implicated (PAX, MEF, and MAZ) in myelomagenesis was observed (Table II). Various TFs such as HNF6, PAX4, STAT3, EVI1, and others were significantly enriched in relapsed cases of MM, including some that have not been implicated in the pathophysiology of MM previously.

Finally, we also sought to determine whether CpG islands were predominantly affected by differential methylation in myeloma. We observed that aberrant methylation occurring in MGUS was significantly found to occur within CpG islands, whereas both new and relapsed cases of MM had aberrantly methylated loci located outside of CpG islands (Fig. 3). This shows that there are qualitative differences in epigenetic alterations between MGUS and myeloma and is also consistent with recent findings that have highlighted that aberrant methylation in cancer can occur outside of CpG islands (16).

Because we saw an overall decrease in methylation in MGUS, we analyzed our dataset for the methylation status of genes involved in this process and also evaluated for the expression of these genes in large gene expression datasets (Arkansas datasets GSE5900 and GSE2658) that have been published previously (4). Specifically, we analyzed the methyltransferases DNMT1, DNMT3a, and DNMT3b and observed that DNMT3a was significantly underexpressed in a large independent cohort of myeloma gene expression profiles (Fig. 4A) and had a significant impact on survival. After separating patients into four quartiles according to their DNMT3A expression, we found that high DNMT3A expression (Q4) at baseline conveyed a significantly overall survival compared with patients with very low DNMT3A expression (n = 140 in each group, log rank test, p = 0.01, Fig. 4C). In our cohort of patients, we found the promoter of DNMT3A to be aberrantly hypermethylated (Fig. 4B). Mutations of the gene encoding DNMT3a have recently been reported for myelodysplastic syndrome (MDS) and AML (17) and have not been examined in myeloma. Sequencing of the catalytic domain of DNMT3A in 23 of our CD138 purified plasma cell samples revealed no exonic mutation or SNPs. These results suggest that DNMT3A expression in myeloma is mainly reduced by aberrant hypermethylation.

Lastly, because we observed that relapsed cases of myeloma exhibit increased aberrant methylation, we wanted to determine the sensitivity of these cells to hypomethylating agents. We tested the efficacy of DNMT inhibitor, decitabine, in longstanding myeloma cell lines that exhibit various cytogenetic alterations and are reflective of relapsed cases. Decitabine was able to significantly inhibit the proliferation of these myeloma cell lines at both low and high doses (Fig. 5). In fact, the dose of 0.5 μM has shown to be hypomethylating without causing DNA damage and was significantly able to inhibit myeloma cell proliferation, thus suggesting that reversal of hypermethylation can be a potential therapeutic strategy in these cases.

We show that myeloma is characterized by widespread alterations in DNA methylation that are specific for different stages of the disease. Early stage of MGUS is characterized by predominant hypomethylation that is prevalent enough to distinguish these cells from normal control plasma cells. The later stages acquire progressive hypermethylation with maximum methylation seen in relapsed cases. These data provide a comprehensive epigenomic map of myelomagenesis and also identify important oncogenic gene pathways that are targeted by these aberrant changes.

The observation of global hypomethylation in MGUS builds upon two recent studies that noticed loss of methylation in plasma cell neoplasms (18, 19). The first study by Salhia et al. (19) reports hypomethylation as early as in MGUS and noted no difference between methylation of new and treated myeloma samples. The second study by Walker et al. (18) noticed hypomethylation in myeloma samples, but did not see any significant differences between normal plasma cells and MGUS samples. Remethylation was seen in plasma cell leukemia samples in comparison with MM. In contrast to the latter study, which interrogated 27,578 CpG sites (corresponding to 14,495 genes), the former study used an approach looking at a limited set of only 1,505 CpG sites (corresponding to 807 genes), thus limiting generalizability of those findings. The study presented in this work was able to distinguish between normal, MGUS, and myeloma (new and relapsed) samples on unsupervised clustering and revealed clear-cut, widespread differences between these groups. We used the HELP assay, a high-resolution assay, that is not biased toward CpG islands and has been able to show stage-specific differences in various tumor models (5, 20). This assay analyzes 25,626 CpG sites and thus is comparable to the system used by Walker et al. (18). Just like Walker et al. (18), we observe relative hypermethylation in late stages of MM; however, we observe hypomethylation much earlier, that is, already at the level of MGUS, and not only MM, similar to the reports of Salhia et al. (19). This discrepancy might be due to the limited number of NL and MGUS samples (3 and 4, respectively) used by Walker et al. (18). Our findings of early hypomethylation in MGUS, which is maintained throughout the early MM stage, but then converts to predominant hypermethylation, thus represent a novel assessment of myelomagenesis.

The finding of hypomethylation in MGUS and myeloma is also different from the previous single locus studies that have focused on hypermethylation of tumor suppressor genes in myeloma (21). The classical cancer-associated epigenetic alteration is promoter CpG island hypermethylation. Even though global hypomethylation was reported in the pioneering epigenetic studies in cancer (22), most investigators have subsequently focused on hypermethylation in CpG islands within selected gene promoters. Array-based DNA methylation assays that have mainly focused on CpG islands have supported the biased perception that CpG islands are uniquely responsible for methyl-DNA–induced genomic changes. Newer iterations of assays for the analysis of DNA methylation, which are based on next-generation sequencing, have revealed a much more complex picture with DNA methylation occurring further upstream in CpG island shores, in introns, as well as extending much further downstream. With this in mind, it is difficult to speculate about the true meaning of differential methylation in CpG islands, and this clearly points to the need of further in-depth investigations with these newer methods. Hypomethylation has been hypothesized to lead to carcinogenesis by promoting genomic instability (23, 24) as well as by aberrant activation of oncogenes (23). Because myeloma is characterized by various chromosomal translocations and deletions, the finding of early hypomethylation may be an important pathogenic mechanism that promotes secondary genetic events that lead to the development of full-blown disease.

Efforts to distinguish MGUS from MM in an unsupervised manner using gene expression profiling data from large clinical trials have not been successful to date (25, 26). Although MGUS samples can readily be separated from NL samples, they appear identical to MM at a gene expression profiling level. Using methylation data, we were able to clearly distinguish between the majority of NL, MGUS, MM, and REL samples. Gene expression profiling can be affected by large-scale changes in just a few transcripts. Because methylation analysis is dependent on evaluation of DNA, it is not globally biased by differences in a few loci, and thus is not affected by changes that may occur only in a minority of the analyzed cells. It is therefore more reflective of biology of premalignant conditions, as illustrated by our previous study in esophageal carcinogenesis (6). In the present study, we used CD138⁺ selection to isolate the myelomatous and normal plasma cells; thus, it is possible that the observed difference between MGUS and MM samples is in part due to dilution with normal samples. Strategies such as multiparameter flow cytometry sorting are being used to ensure clonality in future studies. At this point it should be noted that, although two samples included in our study were in clinical remission with <3% of plasma cells observed on bone marrow biopsy, they were quite different at the epigenetic level. Whereas REM2 as expected clustered with MGUS samples, REM1 clustered with the relapsed samples, suggesting the presence of epigenetic higher risk features that were not appreciated with conventional methods.

Despite the continued progress and the improvement of treatment with newer drugs, drug resistance remains a big concern. Strategies to use highly active drugs in combination upfront have resulted in a significant improvement of progression-free survival; however, they have only had limited effect on overall survival. Multiple studies in other cancers have shown association of hypermethylation phenotypes with resistance to treatment via the inactivation of various cell cycles and other genes involved in chemosensitivity. The presented study identifies a relative hypermethylation in cases of relapsed myelomas compared with normal plasma cells. This difference is further enhanced in the comparison between newly diagnosed MM and relapsed MM (data not shown) and consistent with the observation made by Walker et al. (18) comparing plasma cell leukemia, a very advanced high-risk stage of MM, with newly diagnosed MM. At this point it is not possible to say whether this hypermethylation phenotype is the result of treatment or due to the biology of the disease. However, it suggests inclusion of demethylating agents such as DNMT inhibitors as a viable treatment option. In vitro findings by us and others (27, 28) with representative myeloma cell lines using the DNMT inhibitor decitabine support this assumption, and clinical trials testing this hypothesis are currently underway.

In addition to global quantitative differences in methylation, we also found differences in sites of aberrant methylation between MGUS and myeloma. We observed that even though changes in MGUS involved CpG islands, the later changes seen in myeloma preferentially occurred outside of CpG islands. Recent work has similarly shown that cytosines present outside of CpG islands can be aberrantly methylated/hypomethylated in cancer, and assays that cover these loci are critical to discovering the full landscape of altered methylome of malignancies (16). Our novel findings thus demonstrate both qualitative and quantitative differences between these subsets of disease.

Finally, we also provide insights into the genesis of these epigenetic changes. We observed that DNMT3A is significantly reduced in myeloma and is aberrantly methylated early in both MGUS and myeloma. DNMT3A is an important methyltransferase that has been shown to be mutated in a large proportion of AML cases (17). We did not find any DNMT3A mutations in our cases and showed that this enzyme can be aberrantly methylated in cancer as well. It is generally accepted that DNMT3A and DNMT3B are responsible for the establishment of methylation, whereas DNMT1 ensures maintenance of methylation, thus guaranteeing the faithful transmission of DNA methylation marks from one cell generation to the next. However, this longstanding view is being challenged, and newer models suggest a role for the de novo enzymes DNMT3A and DNMT3B in methylation maintenance (29). In fact, recent data show that DNMT3A mutations impart an adverse prognosis in AML (30). Our data show that dysregulation of this methyltransferase can also be seen in myeloma and adds to the importance of this gene in hematologic malignancies. Analysis of the publicly available data from Walker et al. (18) confirms a significantly higher level of methylation for newly diagnosed MM compared with normal samples and a very high methylation status of MGUS samples with a strong trend toward significance (data not shown). However, generalization of this data is limited due to the low number of normal and MGUS samples included in that study. Unfortunately, due to sample limitation, we were not able to obtain RNA for gene expression profiles from the samples used to perform the methylation analysis. Although we are not able to directly correlate methylation status and gene expression, the finding of hypermethylation and underexpression (compared with normal plasma cells) seen in two different datasets, respectively, as well as the DNMT3A expression effect on survival suggest a role for DNMT3A in myelomagenesis that warrants further investigation. If DNMT3A expression and methylation play a role in the progression of MGUS to MM, requiring treatment is the subject of an ongoing investigation.

In contrast to early MM, relapsed MM shows predominant hypermethylation. Although we cannot show a clear reason for this, others have shown that exposure to chemotherapy can lead to large-scale genetic and epigenetic changes in tumor samples [reviewed in (31–33)]. The relapsed cases examined in our cohort were heavily pretreated patients who have been exposed to multiple MM agents, including bortezomib. It is plausible that our relapsed population is enriched with patients who have a higher overall degree of methylation, as hypermethylated phenotype has been associated with chemotherapy resistance in multiple cancers (34–36) and has been linked with resistance to bortezomib (37). In the current study, we have shown that low-dose decitabine significantly reduces viability of several MM cell lines as models of late-stage disease. Low doses of decitabine have been shown to significantly reduce DNMT levels without the additional DNA damage activity (38, 39). Ongoing longitudinal studies following patients from diagnosis to relapse will hopefully answer the question as to whether hypermethylation can be detected early on and whether it is associated with survival. If hypermethylation were to be detected early and to predict for short progression-free survival, it would be rational to add a demethylating drug such as decitabine to the initial therapy.

J.M. and S.B.S. are on the Speakers' Bureaux of Celgene and Millenium. The other authors have no financial conflicts of interest.