Genome-bound enzymes as epigenetic drug targets in cancer

Epigenetic alterations have emerged as hallmarks of cancer [1]. Unlike genetic changes, epigenetic modifications are reversible, providing means for pharmacologic manipulation of gene expression. Here, we discuss the promise of targeting gene-bound epigenetic modifiers to advance precision oncology.

Large-scale sequencing studies revealed that about half of the somatic mutations in cancer are found in genes involved in epigenome organization across multiple types of cancer – clearly demonstrating the causal role that epigenetics plays in tumorigenesis. This has fueled great interest in this field, and epigenetic therapeutics is one of the fastest growing areas for the discovery of small molecules for the treatment of diseases, providing large numbers of compounds for testing. Some of these agents are already US FDA approved, and hundreds have gone through or are currently in clinical trials. Rational design of treatments is based on the knowledge of biological targets. And yet, with the exception of the methylated status of the MGMT gene promoter to guide the treatment of glioblastoma, epigenetics-based cancer therapies remain empiric rather than rational and biomarker based.

Epigenetic alterations in cancer are extensive, involving many chromatin modifications that span large regions of the genome. Some of these changes, but not all, may have gene transcription consequences at any given time. In many cases, epigenetic alterations are caused not only by somatic mutations of chromatin modifiers, but also by environmental factors which can have a significant contribution. Further, chromatin changes can increase the rate of somatic mutations. This interdependence may drive the coevolution of somatic mutations and epigenetic changes that converge to drive tumorigenic pathways. The list of epigenetic enzymes is extensive, and any given chromatin modification can be catalyzed by several enzymes within a class. When tested in vitro, these enzymes can exhibit promiscuous substrate activities. However, within a cell the specificity of actions of these enzymes is, in part, controlled by in situ compartmentalization mediated by anchoring or scaffold proteins (e.g., the nucleic acids-regulated docking protein, hnRNP K [2]). Thus, enzymes tethered to specific loci in the vicinity of their transcription and chromatin substrates could be viewed as druggable targets to alter expression of chosen cancer genes. As such, identification of enzymes bound to cancer driver genes would provide a paradigm to guide rational-design tumor epigenetic therapy.

Epigenetic plasticity is an important feature of cancer which, at least in part, may explain the success of epigenetic therapy in the treatments of some malignancies. Cancer is a disease where epigenetic therapy is both promising (given the plasticity) and needed, given the decades long lack of significant progress in the treatment of many solid tumors. Thus, targeted cancer therapy selectively modulating activities of one or more genome-bound enzymes at specific loci offers a new direction in precision oncology. Given the complexity of epigenetic processes, a long list of epigenetic modifiers and disease-related genes can the concept of gene-bound enzymes as disease biomarkers and druggable targets be rendered clinically practical? In this regard, there are several challenges to overcome to exploit the concept of gene-bound enzymes as druggable biomarkers in personalized oncology.

What genes to target epigenetically?

Tumors are initiated and maintained by somatic mutations which generate driver oncogenes and/or mutations in tumor suppressors. Thus, it is thought that cancer cell survival depends on relatively few driver genes. In fact, there is evidence that in many cancers inhibition of a single oncogene is sufficient to obliterate tumors, a concept termed oncogene addiction. In some cases, the oncogenic proteins can be inhibited directly with small molecules as in the case of BCR-ABL or mutated BRAF.

In other instances, there are no small molecule inhibitors available that target-mutated oncogenic proteins. Here, the ability to epigenetically repress such cancer genes would provide an alternative way to control expression levels of these oncogenic factors.

Epigenetic silencing of tumor suppressor genes is thought to be an early initiating event in the oncogenic processes of most cancers. These observations have been exploited in derepression of tumor suppressor loci, for example, hypermethylated promoter of the tumor suppressor – CDKN2A gene. However, there are examples where the tumor driver genes are unknown and epigenetic manipulations of elusive driver loci is not an option. How to approach such cases? Oncogenic pathways converge on genes that define cancer cell phenotype including uncontrolled cell proliferation, resistance to death, immortalization, angiogenesis, activated invasion and others [3]. Here, epigenetic targeting of cell proliferation genes as a group appears attractive not only in instances where driver mutations are unknown but also in cases when resistance emerges after targeting oncogenic proteins (e.g., tumor recurrence after vemurafenib inhibition of oncogenic BRAF). In this regard, as a class, genes encoding cyclin dependent kinases that drive cell proliferation seem attractive targets for epigenetic repression. Similar approach can be taken to epigenetically target other classes of cancer genes that are driven by oncogenic proteins-mediated signaling.

Data-driven approaches can also be used to provide potential candidate genes to target epigenetically. The chromatin accessibility of 410 tumor samples spanning 23 cancer types from The Cancer Genome Atlas (TCGA) were recently profiled via ATAC-seq [4]. Combined with existing TCGA DNA methylation data of tumor and adjacent normal tissues, as well as normal tissues from the NIH Roadmap Epigenome Project [5], it is possible to computationally identify genomic loci and the associated genes that are aberrantly silenced or activated epigenetically across multiple types of cancer. This candidate gene list can be further prioritized by considering their functional roles in the hallmarks of cancer (proliferation, metabolic reprogramming, etc.).

There is increasing evidence that the bulk of tumor cells are derived from a subset of cells designated as cancer stem cells (CSC). These are a small and unique population of cells with stemness-like properties for self-renewal and capacity to generate different cell types that constitute the bulk of the tumor mass. Critically, CSC are thought to account for resistance to chemotherapy and tumor recurrence. Single-cell chromatin immunoprecipitations (ChIP)-seq technologies enabled the characterization of rare cell types within the heterogeneous tumor, such as the CSC, and resistant or persistent cells and their unique epigenomic profiles [6]. The pluripotent character of CSC is thought to reflect plasticity of their epigenome which can be reprogrammed to induce these cells to differentiate and as such render them sensitive to chemotherapy [7]. Stemness genes that induce and maintain the CSC phenotype are known and could be the pharmacological loci for epigenetic manipulation through targeting of enzymes tethered to such sites.

What epigenetic modifiers to target?

One hallmark of the epigenome is its plasticity, a property that is more prominent in cancer compared with differentiated cells [1] and thus can be more readily manipulated with drugs. Genome-bound epigenetic modifiers are druggable, providing critical avenues to exploit the epigenome’s plasticity in order to mitigate epigenetic changes responsible for cancer cell phenotypes. Thus, the ability to accurately assess epigenetic profiles and chromatin-bound modifiers is critically important for designing new cancer treatments. Although the list of chromatin modifying enzymes is extensive, advances in ChIP assays have overcome the challenges to study low abundance gene-bound proteins allowing high throughput parallel detection of dozens enzymes at loci of interest [8,9]. Many drugs, including selective agents, have been introduced that target DNA and histone modifying enzymes including EZH2 (H3K27me methyltransferase), DOT1L (H3K79me methyltransferase), HDACs (histone deacetylases), DNMTs (DNA methyltransferases) and ERKs (protein kinases). These modifying enzymes are involved in oncogenic processes including initiation and maintenance of CSCs [7]. Agents that inhibit activities of these enzymes can induce CSC differentiation and render them sensitive to chemotherapy [7].

Chromatin components and transcription machinery are regulated by a wide range of signaling processes in which protein kinase cascades play a prominent role. Many factors involved in chromatin and transcription are kinase targets. There is a long list of chromatin-bound protein kinases that are implicated in cancer representing virtually every family of these enzymes (e.g., mitogen-activated protein kinases [MAPKs], PKCs, CDKs, Aurora and AMPK). The MAPK cascades are perhaps most studied because of their prominent role driving abnormal cell proliferation in cancer. The extracellular signal-regulated kinases (ERKs) MAPK pathways have emerged as being particularly important in cancer (e.g., activating mutations in EGFR, KRAS and BRAF). Nearly the entire ERK cascade, from the canonical receptor tyrosine kinases (RTKs such as EGFR) to ERKs, are recruited to target loci illustrating an example for spatiotemporal control of enzyme activity and as such identify genes directly controlled by these pathways [2,9,10]. Thus, detection (e.g., by ChIP assays) of druggable epigenetic enzymes including components of MAPK pathways at cancer-related genes could be used to guide rational tumor treatments.

There are several approaches available to prioritize confirmation of epigenetic modifiers as druggable gene-tethered proteins by high throughput ChIP analysis. First, their aberrant expression, copy number variation or mutation are documented in patient samples listed in TCGA. Second, their known physical interactions with cancer-associated epigenetic enzymes. Third, their known interactions with transcription factors whose target genes are aberrantly expressed and/or exhibit altered chromatin accessibility in cancer. Fourth, their effective enzymatic activity on sets of cancer genes is predictable using machine learning approaches. For example, based on multiomic features from Cancer Cell Line Encyclopedia [11] and Genomics of Drug Sensitivity in Cancer projects [12] that profiled the sensitivities of hundreds of cancer cell lines to a wide range of compounds (including MAPK/ERK inhibitors).

How to overcome epigenetic intratumor heterogeneity?

Advances in sequencing are rapidly uncovering the molecular basis for tumor phenotypic heterogeneity. These studies have shown that cells within spatially separated tumor regions have distinct somatic mutation profiles (suggesting multiple tumor clonal expansions), accounting for intratumor heterogeneity. This could be more profound in tumors that are drug resistant or in cancers that recur [13]. And yet, despite heterogeneity, cells from different tumor regions share some mutations, indicating a common cellular evolutionary ancestry (presumably early tumor evolution events such as CSCs [14]). Thus, it has been suggested that the identification of the earliest branch points in a tumor’s evolutionary history could be a way to identify the most effective druggable targets (e.g., gene-bound enzymes) [14]. As such, the ability to map epigenetic enzymes and factors bound to shared cancer genes (e.g., oncogenic drivers, CSC and others) provides a practical way to identify the best common drug targets within an otherwise heterogeneous tumor epigenetic environment. The latest technologies, such as single cell RNA-seq, ATAC-seq, ChIP-seq and spatially resolved RNA-seq [15], can be used to identify and target rare cells that underlie drug resistance, recurrence or metastasis, and infer the epigenetic enzymes and transcription factors regulating the unique epigenomic profiles of these cells that are amenable to targeted therapies. Thus, the increased understanding of the molecular mechanisms that underlie tumor heterogeneity shifts diagnostic and therapeutic paradigms in that multiple, rather than single, tissue sampling (biopsies) are needed to identify mutual, across tumor regions, druggable gene-bound enzymes, paving the way toward advancing personalized therapies.

How to maximize epigenetic drug tumor-specificity?

Multicomponent drug treatment is increasingly becoming the standard of care for diseases such as AIDS and hypertension. It refers to a therapeutic regimen that aims at multiple targets using several agents. The goal of drug combination treatments is to achieve synergistic, or at least additive, therapeutic effects using lower doses of drugs to minimize toxicity. Inhibition of two or more targets counteracts pathway redundancies and/or induces synergistic responses that are greater than using a single agent alone [16]. One of the most successful examples of drug combinations is the highly active antiretroviral therapy regimen for management of HIV/AIDS, where three or more drugs target different viral proteins. Until now epigenetic agents have been used and tested largely as monotherapies. Given the increasing availability of small molecules that target different classes of epigenetic modifiers, combinatorial use of these agents can be highly synergistic lowering inhibitory concentrations (IC₅₀) by an order of magnitude [7,17]. Further, it is increasingly recognized that epigenetic drugs can exhibit gene selectivity opening exciting avenues to effectively target sets of specific loci at lower doses and as such reducing off target effects [18]. By systematically generating transcriptomic and epigenomic profiles before and after epigenetic drugs and their combinations [18], one can assess the selectivity of epigenetic drugs.

Enzymes (epigenetic modifiers, MAPK pathway components, etc.) bind directly or indirectly to many loci across the genome, and their binding sites can be systematically revealed using high throughput technologies such as PIXUL-ChIP (stands for PIXulated ULtrasound) [8]. However, the genomic and chromatin features that determine their binding patterns remain elusive. Deep learning has been recently used to predict DNA accessibility patterns from genomic sequence alone [19]. Deep learning models have also been trained to predict the binding affinities of individual transcription factors and predict how noncoding genetic variants affect binding using ENCODE ChIP-seq data [5]. This deep learning approach can also be used to learn the regulatory logic of gene-bound enzyme binding, and discover sequence motifs that are associated with the selectivity of gene-bound enzymes. This deep learning model can then be used to predict the effects of cancer mutations on gene-bound enzyme drug responses specificity.

Pharmacologically targeting enzymes that colocalize at given loci, each enzyme with a different agent, could result in desired and potent synergistic and selective effects on target gene expression. Thus, colocalization analyses of genomic elements could identify two or more druggable epigenetic modifiers that significantly colocalize to the same or neighboring genomic sites [20]. In some cases, such colocalization could be mediated by nucleic acids docking platforms [2]. For example, hnRNP K is implicated in cancer and interacts with both protein kinases and components of the PRC2 complex which includes EZH2. Thus, large-scale protein–protein interaction datasets such as the Bioplex (http://bioplex.hms.harvard.edu/index.php) can be used to further prioritize enzymes that not only colocalize to similar genomic loci, but also interact with each other. Colocalization of a given sets of druggable epigenetic modifiers at functionally-related cancer genes (e.g., proliferation, angiogenesis and CSC) can be confirmed by ChIP assays. If so, such finding would serve as the bases for testing synergistic action and dose reduction of epigenetic drugs (minimizing off target effects given much lower drug concentrations) first in vitro and then in clinical trials.

Another avenue to search for drug selectivity is to correlate analysis of cancer genes-bound enzymes in tumor biopsy samples (e.g., ChIP assays of tissue specimen including FFPE samples) from epigenetic drug clinical trials with clinical outcomes (response vs no response). For example, in clinical studies using a MEK inhibitor presence of MEK (or ERK) at transcribed cell proliferation genes may correlate with certain tumor histology and patients response to such a treatment. Further, epigenetic analysis of cancer tissues from clinical trials that use drug combinations could provide information to exploit drug synergism while minimizing unwanted off target gene expression effects. For example, in studies using azacytidine (DNA methyltransferase inhibitor) and valproic acid (HDAC inhibitor) presence of cognate enzymes at tumor suppressor genes may be correlated with patients response to this combination and as such allow lowering the doses of each one of these drugs maintaining the desired action while minimizing unwanted effects.

In summary, advances in high-throughout sensitive epigenetic technologies and computational tools provide clinically practical means to profile binding of dozens of druggable epigenetic modifiers to cancer-related genes in tumor biopsy samples. We suggest that this approach combined, with histology, presents an avenue toward rational-design epigenetics-based precision oncology.