2000 Publication

Traditionally, cancer treatments have been selected on the basis of tumor type, pathological features, clinical stage, the patient's age and performance status, and other nonmolecular considerations. We have generally accepted with a certain fatalism that some patients pigeonholed into a given category will have a response to a particular therapy, whereas others will not. The difference is often viewed as a matter of luck, like the result of a coin toss, but in fact, treatment response can be predicted in some cases, whereas it is close to impossible to predict the results of a coin toss. The field of pharmacogenomics, through the study of large numbers of genes that influence drug activity, toxicity, and metabolism, provides the opportunity to tailor drug treatments and to eliminate many of the uncertainties of current therapy for cancer. 

Strong support for this concept is provided by the study of genetic polymorphisms that influence drug metabolism. (1) CYP2D6, for example, affects the metabolism of a wide range of agents, including beta-blockers, antidepressants, antipsychotics, and opioids. Dihydropyrimidine dehydrogenase influences the metabolism, and therefore the neurotoxicity, of fluorouracil. DNA-sequence variants may also directly influence the toxic side effects of a drug or its ability to interact with its target.  

In this issue of the Journal, Esteller and colleagues (2) provide clinical evidence of an intriguingly different sort of mechanism -- an epigenetic one that does not involve any change in DNA sequence -- to explain the resistance of some gliomas to nitrosourea alkylating agents. Carmustine (BCNU) and other nitrosoureas kill by alkylating the O6 position of guanine and thereby cross-linking adjacent strands of DNA. Formation of these cross-links can be prevented by the DNA-repair enzyme O6-methylguanine-DNA methyltransferase (MGMT), which rapidly reverses the alkylation. Wide variations in the expression of MGMT are found within and among tumor types. In particular, about 30 percent of gliomas lack MGMT. Although the literature on this subject is complex, a lack of MGMT appears to correlate with sensitivity to carmustine. 

Mutations in the DNA sequence of MGMT are unusual and cannot be invoked to explain the variation in levels of expression. So what is the mechanism? It has been proposed that methylation of the MGMT promoter region, with consequent transcriptional silencing of the gene, may account for this variation. (3) DNA methylation of normally unmethylated CpG (cytidine-phosphate-guanidine) islands in the promoter regions of genes for tumor suppressors, DNA-repair enzymes, receptors, and cell-cycle proteins can silence those genes in cancer cells and thus influence tumor evolution. Using a methylation-specific form of the polymerase chain reaction, (4) Esteller et al. (2) studied methylation of the MGMT promoter in 47 consecutive newly diagnosed grade III and IV gliomas and found a striking relation to the response to treatment with carmustine. Twelve of 19 patients with methylated promoters in their tumors had a partial or complete response to treatment, whereas only 1 of 28 patients with an unmethylated promoter had a response (P<0.001). Overall survival and time to progression were also longer in patients whose tumors had methylated promoters. (2) These findings suggested that methylation of the MGMT promoter could be used to predict responses to treatment with carmustine. 

Further clinical studies will be necessary, of course, to validate these impressive first results, and it would be interesting to verify directly that methylation of the MGMT promoter in these tumors correlates strongly with MGMT expression and activity. But the implications of the study, if its results can be replicated, are clear: carmustine therapy might be reserved for patients whose gliomas have methylated MGMT promoters, and the response to carmustine might be increased by agents such as O6-benzylguanine that inhibit MGMT activity. A very simple calculation indicates the potential power of such therapeutic markers. For the 47 patients in the study by Esteller et al., the overall tumor response rate was 27.7 percent. Given that response rate, if this had been a phase 1 trial of a new drug, 10 patients taking an effective dose would have been required to produce a 95 percent chance of at least one response. In contrast, if only those with methylated promoters (with the observed 63.2 percent response rate) had been admitted to the trial, only three patients would have been needed to achieve similar certainty of seeing at least one response. More important for clinical practice, patients with unmethylated MGMT promoter regions in their tumors could be spared the considerable toxicity of carmustine and could instead be given an agent more likely to be effective against the tumor. 

Pharmacogenomic studies will inevitably produce benefits such as these for both clinical research and standard practice. From the perspective of the pharmaceutical industry, they have the potential disadvantage of dividing the market for a successful drug, but their larger potential advantages include the discovery of better drugs, elimination of poor candidate drugs early in the development process, and dramatic decreases in the size and expense of clinical trials. 

The study by Esteller and colleagues provides a case in point. It presents clinical correlations with respect to a particular promoter and a particular class of drugs. But it immediately raises broader questions. The nitrosoureas have activity in tumors other than gliomas, including some lymphomas, cancers of the gastrointestinal tract, and melanomas. Will MGMT-promoter methylation serve to identify patients with those cancers who might benefit from therapy with nitrosoureas? More generally, how often will epigenetic methylation of CpG islands in promoters of other genes prove useful for the selection of treatments beyond the nitrosoureas? To address the latter question, various methods are being developed to scan large numbers of promoters for differences in methylation. (5) Thus it seems likely that progress in this field will require a survey of many genes. 

Such comprehensive approaches to biology can be characterized as "omic" research (6) -- that is, research in which one generates large resources of information on biologic molecules in aggregate without necessarily knowing in advance which pieces of information and which correlations will prove most important. (7) "Omic" research is hypothesis-driven, but the hypothesis relates to information and its usefulness, rather than to particular molecules or processes. "Omics" began with genomics and the Human Genome Project. Then, as coined by various researchers, there came proteomics, kinomics (for the kinases in aggregate), CHOmics (for the carbohydrates), metabolomics, immunomics, toxicomics, and clinomics -- as well as compound forms, such as functional genomics, structural genomics, and pharmacogenomics. In view of the study by Esteller et al., (2) and as we search for other clinically relevant instances in which promoter methylation affects therapy, can "pharmacomethylomics" be far behind?