- © 2012 by American Society of Clinical Oncology
DNA Methylation As a Clinical Marker in Oncology
- Corresponding author: Jean-Pierre Issa, Fels Institute for Cancer Research and Molecular Biology, 3307 North Broad St, Rm 154, PAHB, Philadelphia, PA 19140; e-mail: jpissa{at}temple.edu.
Personalized oncology is critically dependent on accurate measurement of biomarkers in cancer tissues or surrogate tissues if appropriate. Historically, the field started with protein biomarkers such as hormonal status in breast cancer1 and genetic biomarkers such as chromosomal changes in the management of patients with acute myeloid leukemia.2 The genomics era brought RNA into the mix, and gene expression profiles are now commonly sought in breast and other cancers.1 The supremacy of protein and RNA as diagnostic, prognostic, and predictive biomarkers now faces a serious challenge with the resurgence of DNA-based information. The recent successes of mutational profiles in predicting response to targeted therapies have been stunning, and this is likely to expand with genome-wide studies.3 Moreover, the clinical impact of biomarkers is dependent on the stability of the information provided. In this respect, protein and RNA studies have serious drawbacks. In Journal of Clinical Oncology, Claus et al4 report that DNA methylation analysis of the ZAP-70 gene provides more accurate prognostic information in chronic lymphocytic leukemia (CLL) than protein-based analysis using flow cytometry.
DNA methylation in mammals refers to a covalent modification of the cytosine base that is prevalent throughout the genome.5 Methylation affects protein-DNA interactions and therefore has a regulatory role in gene expression. Generally, methylation in or near transcription start areas is associated with reduced or absent gene expression, as observed here for the ZAP-70 gene. One of the key features of DNA methylation is stability over time. DNA methylases faithfully replicate information from the parent strand to the newly synthesized strand.5 DNA methylation in gene promoters has been shown to remain stable over years in culture and to be relatively similar in primary cancer cells, xenografts, and early passage cell lines.6 Thus, the epigenetic information gleaned from DNA methylation analysis is not just expression but a prediction of (relatively) stable expression over time7 (Fig 1). In other words, a gene that is repressed by DNA methylation in a given cancer or leukemia will remain repressed regardless of the sampling time.
Gene expression patterns from methylated versus unmethylated promoters. Top panel: putative gene exon 1 (light blue box) embedded in a methylated promoter (M), which leads to a stable repression complex (gold ellipse). As a result, gene expression (right) is uniformly low, regardless of sampling time (black arrows). Bottom panel: the same gene if unmethylated; in this case, several proteins compete for binding (eg, transcription factor [TF], transcription repressor [TR], histone acetylase [HAT], histone deacetylase [HDAC]) to trigger transcription through polymerase II (POLII). The levels of mRNA produced (dark blue line) are further regulated by microRNAs (red lines) and other factors. As a result, gene expression (right) is stochastic and variable over time. Sampling time (black arrows) may result in large differences in measured gene expression. Because cancer biomarkers often require stability, in some cases, DNA methylation may provide more useful information than gene expression.
In contrast to methylated promoters, gene expression from unmethylated promoters can be quite variable. The measured expression results from a balance of distinct factors including RNA polymerases, transcriptional activators and repressors, histone acetylases and deacetylases, various microRNAs, and other processes. This balance results in a significant stochasticity in gene expression8 (Fig 1). Consequently, gene expression measurement at a single time point can be misleading, because it is not always predictive of gene expression a few hours or days later, particularly in the face of shifting conditions and resources available to the cell. Indeed, Claus et al4 show that expression of ZAP-70 is uniformly low when the gene is methylated but quite variable when it is unmethylated. Thus, DNA methylation, when present, could provide a more reliable approximation of the contribution of a particular gene to the physiology of the cancer. In addition, DNA is considerably more stable than RNA or protein, and analysis of degraded DNA is still more reliable than that of degraded RNA. This should have a significant impact in improving accuracy of predictors outside tertiary medical centers, where collection and handling of specimens for molecular testing can be difficult to control.
There are precedents for DNA methylation as a more accurate predictor of outcome than expression. For example, inactivation of the methyl-guanine methyltransferase (MGMT) gene is associated with better response to alkylating-agent therapy, including temozolamide in glioblastoma.9 Methylation of MGMT has been found to be a more reliable predictor of outcome after therapy than protein expression.10 In addition, methylation could provide clinical information beyond the mere function of the gene measured. It is now clear that many cancers are characterized by intense abnormal methylation of multiple genes, a phenomenon termed CpG island methylator phenotype (CIMP).11 CIMP confers a unique biology and distinct clinical outcome in different tumors and can only accurately be ascertained by DNA methylation analysis.12 For example, CIMP is associated with a good prognosis in brain tumors13 and in breast cancer.14 Interestingly, it has recently been reported that the commonly used gene expression predictors of outcome in breast cancer may simply be surrogates for CIMP status.14 If confirmed, these data suggest that DNA methylation analysis could supplant these tests in the clinical management of patients with breast cancer, in the same way that the current report4 might lead to ZAP-70 methylation replacing ZAP-70 flow analysis in predictive models of CLL outcome.
An added wrinkle to the information provided by DNA methylation analysis is the availability of drugs that inhibit DNA methyltransferases and reduce methylation in vitro and in vivo.15 These drugs are useful in myeloid malignancies and have shown intriguing occasional responses in solid tumors.16 Therapy-induced hypomethylation of specific genes correlates with response to these drugs,17,18 but it is not yet known whether there are DNA methylation patterns at diagnosis that predict for response. The effects of these drugs are pleiotropic and difficult to predict given the large number of genes affected. At first glance, there are instances in which one might interpret DNA methylation patterns in light of potential response. Thus, ZAP-70 methylation is associated with favorable outcome in the study by Claus et al,4 which might be interpreted as precluding a beneficial effect of this therapy in CLL. However, ZAP-70 is one of potentially hundreds or thousands of genes variably methylated in cancer.19 One assumes that a potential therapeutic ratio of the drugs is related to the sum total of the effects rather than the outcome of a single gene. This issue requires considerably more data, particularly data relating genome-wide patterns with response to therapy.
Biomarkers are playing an increasing role in the management of patients with malignancies. Arguably, it may one day be less important to see cancer (pathology and scans) and more important to know what we cannot see—the cancer genome. It is also increasingly likely that DNA sequencing will not be sufficient for this purpose and that comprehensive DNA methylation analysis will be equally important in determining outcomes and selecting patients for various therapies.
AUTHOR'S DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST
The author(s) indicated no potential conflicts of interest.
