Accidentals of the DNA Symphony

Prostate cancer is the most common cancer type in men in Western countries. Localized prostate cancers are often indolent, but prostate cancers that have spread eventually progress and develop resistance to intensified androgen receptor (AR)–targeted therapies or chemotherapy. Following large investments in next-generation sequencing, studies in the past decade have characterized the genome and transcriptome landscapes of prostate cancer. These have identified therapeutic targets, such as BRCA gene alterations, that have led to the development of new treatments for advanced prostate cancer and have started to elucidate drivers of resistance, which often center around AR (1, 2).

More recently, Zhao and collaborators integrated whole-genome bisulfite sequencing with whole genome and transcriptome data to characterize 5-methylcytosine (5mc) levels in 100 metastatic castration-resistant prostate cancer (mCRPC) biopsies (3). In this issue of Cancer Research, Sjöström and colleagues have added another layer of DNA epigenomic information: 5-hydroxymethylcytosine (5hmC) on solid tumor tissue biopsy and cell-free DNA samples (4). 5hmC, the first oxidative product of 5mC, is a transition state of the reverse methylation, or DNA demethylation, process that involves TET enzyme to oxidize 5mC with the help of α-ketoglutarate substrate. It has been shown that 5hmC is preferentially located on gene bodies and enhancers. The presence of 5hmC may indicate active demethylation processes and therefore provide complementary and potentially clinically useful information on altered gene expression, cancer proliferation, progression, and tissue dedifferentiation.

There are several approaches to profile 5hmC including capture-based or single-base resolution methods such as TET-assisted pyridine borane sequencing (5), chemical-assisted pyridine borane sequencing, or oxidative bisulfite sequencing. In general, these approaches have demonstrated high concordance for 5hmC profiling (6). In this current study, the authors have used a capture-based approach and complemented their mCRPC cohort with samples from 52 localized prostate cancers and 12 normal tissues (5 benign prostates and 7 samples of normal tissue adjacent to metastatic biopsies). Comparison of localized prostate cancers and mCRPC identified 5hmC levels at sites that could explain differences in expression, phenotype, and tumor behavior between these two states and of distinct mCRPC phenotypes. The most extreme mCRPC phenotype, which remains rare despite potent AR targeting, is often referred to as treatment-emergent small cell neuroendocrine prostate cancer. This is characterized by low levels of AR signaling and neuroendocrine features on a genomic background not markedly different from the rest of mCRPC. In contrast to genome similarity and in keeping with prior studies of 5mC marks (7), 5hmC patterns are strikingly different with emergence of this de-differentiated state (4). This in part corresponds to the very significant transcriptional differences, with lower 5hmC levels in AR pathway genes and higher levels in upregulated neuroendocrine genes. A potentially important but challenging line of inquiry could be to identify the “trigger” that regulates or induces these changes in 5hmC levels and that could be targeted or reversed to treat this aggressive phenotype. 5hmC levels further split the more common AR-active mCRPC phenotype into two groups, one group aligning with the authors’ previously described hypermethylated tumor subtype associated with TET, DNMT3B, IDH, and BRAF somatic mutations (3).

Given the limitations of tumor biopsies, both in obtaining them from patients and in their limited sampling range, the authors then expanded their study to plasma cell-free DNA from 15 patients obtained at the same time as biopsy in their mCRPC cohort and an additional 64 mCRPC cases obtained prior to initiation of second-generation AR-targeted therapies. Liquid biopsies have been widely implemented clinically for patient monitoring and treatment selection using genome-based tests. However, a major limitation remains the inability to identify disease subgroups not characterized by genomic alterations. Extracting epigenetic information from plasma DNA and using this to influence patient management is probably one of the next frontiers for cancer biomarker discovery.

Both 5mC and 5hmC levels can be extracted from plasma DNA. However, the proportion of cancer-specific DNA in a plasma sample is usually lower than in a tumor tissue biopsy, and the consequently high levels of admixed noncancer plasma DNA arise mostly from leukocytes and hepatocytes. The large amount of prostate-specific 5mC and 5hmC markers can be used for sensitive detection of circulating cancer DNA, for example, in monitoring of residual disease or early diagnosis; in fact, 5mC-based tests are in advanced clinical trials of population screening (8). For patients with advanced cancer, these methylome data can be used to estimate the circulating tumor fraction and prognosticate patients, evaluate treatment response, and even monitor neuroendocrine differentiation (9). In addition, these data could be integrated with the plasma DNA fragmentome that, as reported in a recent study using plasma samples overlapping with this 5hmC series, more accurately infers tumor transcriptomes (10). Plasma DNA arises from multiple sites of cancer in a patient, and it is potentially the more clinically relevant ones that are consequently actively represented in circulation. Many now agree that plasma DNA could therefore be more informative than single tumor biopsies for determining resistance and stratifying patients for treatment. There is a biomarker opportunity in using plasma DNA at genes that harbored distinct 5hmC patterns and associated with clinical outcome but not commonly altered at the gene level. Of particular interest to the authors were EZH2 and TOP2A, the former amongst the top differential 5hmC regions in mCRPC compared with localized prostate cancer (4). 5hmC gene body levels across both these genes were prognostic in the discovery cohort, after adjusting for tumor fraction levels and clinical variables. Further clinical qualification is required to determine the clinical utility of such information.

The major challenge now for clinical implementation of epigenetic plasma data in advanced prostate cancer is the expansion of this technology to samples with lower tumor fractions (<20%). This is necessary for inclusion of the majority of patients with mCRPC and especially patients with a better prognosis for whom treatment change may be more effective. This is, however, very challenging and will probably require deeper sequencing and computational approaches suited for low tumor fraction samples. In conclusion, the work of Sjöström and colleagues represents an important step in improving on the granularity of current genome-centric biomarkers used in mCRPC. Further studies are now required to define the clinical utility and optimal disease settings to implement 5hmC markers for prostate cancer management.

A. Wu and G. Attard have a patent on blood methylation markers (PCT/GB2020/052706) and are co-founders of a company (Cansor) to develop methylation-based blood tests. G. Attard is also listed as a co-inventor on a patent related to plasma genomic analysis (P032355GB). G. Attard has received personal fees, grants, and travel support from Janssen and Astellas Pharma; personal fees or travel support from Pfizer, Novartis/AAA, Bayer Healthcare Pharmaceuticals, AstraZeneca, and Sanofi-Aventis; in addition, G. Attard's former employer, The Institute of Cancer Research, receives royalty income from abiraterone and G. Attard receives a share of this income through the Institute's Rewards to Discoverers Scheme. No other disclosures were reported.