Epigenetic drugs including DNA methyltransferase inhibitors, which are meant to correct for DNA methylation imbalances in cells, constitute promising therapeutic approaches in the battle against cancer. The FDA-approved azanucleotides 5-azacytidine and decitabine are already administered to patients with hematologic neoplasias. Zebularine has emerged as a new member of this type of agents that has shown potentials for long-term oral applications, as a result of systematic comparative analyses
[23–27, 70, 71]. However, most of the assessments have been performed utilizing molecular methods that reveal precise information regarding CpG methylation profiles of non-repetitive sequences, but are currently costly and time-consuming, if not challenged, when applied in a cell-by-cell mode. Nevertheless, we believe that analysis of cultured cell models at single-cell resolution is necessary to obtain a more global and cell systemic picture of drug action and efficacy in the search for new drugs as well as the epigenetic evaluation of existing drugs. Thus, high-content and high-throughput analyses, which have been supported by recent advancements in imaging technology and computational capacities, offer valuable means for rapid and cost-effective cellular phenotyping in drug screening
. Furthermore, the vast majority of studies so far have been focusing on assessing the hypomethylating potential of drugs on selected gene promoters in combination with cell viability testing for drug cytotoxicity and genotoxicity. However, hypomethylating agents can also perturb the epigenetic regulation of chromatin conformation, thus having an impact on the higher-order genome organization and nuclear architecture that regulate genome integrity and gene expression
. We were interested in tracking the progression and extent of such global structural changes, also in correlation with drug cytotoxicity to additionally elaborate on the verification of the 3D-qDMI system’s utility for the therapeutic field. Towards this end, we have conducted a comparative cell-by-cell evaluation of zebularine and its extensively characterized isoform 5-azacytidine based on their effects on global nuclear DNA and its higher-order organization in the cell nucleus. For the purpose of generating comparable topological data, we chose human cell culture models that have rendered themselves as sensitive to both agents, as well as cell culture conditions and drug doses that have been used previously in comprehensive studies to explore differential changes on the level of DNA methylation for targeted single-copy CpG sites. Our study includes standard viability testing for measuring cytotoxicity and upgraded 3D-qDMI for evaluating the demethylation effects on two levels: (i) changes in the load of nuclear MeC (I
MeC), and (ii) alterations in the spatial codistribution of MeC and global DNA, including condensed heterochromatin regions that are represented by bright DAPI areas in the nuclei of cells. Our cytotoxicity data as well as the results of our topologic approach are strongly concordant with data presented by other investigators
[23, 25, 26, 73–75]. Drug response efficacy, as judged by the degree of spatial nuclear MeC/DAPI patterns, was comparably high for the two drugs across all concentrations.
In terms of cytotoxicity, we found that the Huh-7 hepatocarcinoma cells reacted more sensitively to zebularine than the prostate cancer cells. Nevertheless, for both cell types, zebularine elicited similar cytotoxicity levels at doses that were one to two orders of magnitude higher than for 5-azacytidine, thus can be considered as much less cytotoxic at near-equimolar concentrations. The results are in accordance with data from other investigations that have probed the two agents in various other cancer cell models such as bladder (T24), colon (HCT-116), ovarian (A2780 and HEY) and breast (MBA-MD-231 and MCF-7) cancer cell lines, as well as in acute myeloid leukemia cells (AML 193)
[23, 25, 73–75]. Investigations addressing the chemistry behind this phenomenon have led to cumulative evidence indicating the formation of a permanent covalent bond between human as well as selected bacterial DNMTs and 5-azacytidine that can trap the enzyme in a suicide complex (triggering apoptosis). In comparison, only a stable but no permanent covalent bond has been proven between zebularine and the same DNMTs, which would allow the enzymes’ release after binding in vitro as well as in vivo. This may explain why higher concentrations of zebularine are necessary for similar levels of global DNA hypomethylation in cell nuclei and its lower cytotoxicity (at equimolar concentrations), compared with AZA
Furthermore, we observed that the increase in cytotoxicity correlates with global 5-methylcytosine levels, especially the extent of DNA hypomethylation at DAPI-positive heterochromatic sites as revealed by 3D-qDMI through scatter plotting of MeC/DAPI codistribution. This was also true for AZA-treated cells (data not shown). Along the same lines, when localizing low-intensity MeC and DAPI sites in the same nuclei, we could map the gradual increase in LIMs from the nuclear periphery into the more interior parts of the nuclei. However, we experienced that a strong level of LID increase within the nuclei interior was already seen at the lower zebularine concentration (8–40 μM), compared to naïve cells, which did not significantly change up to the highest concentration applied (1000 μM). These LID-patterns were very similar to the one in AZA-treated cells (Figure
8), in which the majority of LIDs were found to be located in the nuclear periphery. These conclusions are drawn from images of cells with seemingly intact nuclear envelope. In fact, for drug concentrations ≥5 μM for 5-azacytidine and ≥500 μM for zebularine, a large number of cells were found to present DAPI and MeC signals outside of their nuclei, leading to the assumption that the drugs had also affected the nuclear envelope and caused DNA leakage. In these cells the respective LID curves were located below the diagonal of the graphs (not shown). Due to the cytotoxic effect induced by high drug concentrations, such cells were not included in our further analyses.
Therefore we cannot exclude any contribution of topological changes of gDNA/heterochromatin to cytotoxicity. On the contrary, we assume that global DNA demethylation may lead to both DNA hypomethylation as well as gDNA reorganization, which are bilateral and together could lead to cellular decline. Although, our data here suggest that cytotoxicity is more fine-correlated with DNA hypomethylation than with bulk DNA reorganization. However, it may also be possible that only local gDNA rearrangements occurred under the conditions applied in our study. The latter effect is conceivable from the increase of LIMs in nuclear areas that harbor heterochromatin: as a significant LIM increase was detected for cells already at low zebularine doses, a compounding of both DNA demethylation effects may have triggered cellular disruption. Figure
6 underlines the fact that 5-azacytidine has equivalent effects at concentrations that are much lower than of zebularine.
The mode of action of azanucleosides is quite complex
. Cytosine hypomethylation by azanucleosides, including zebularine, has been extensively reported to reactivate tumor suppressor genes and apoptosis-related genes
[79–81] but also the relaxation of highly compacted chromatin that can be seen as a loss of gDNA (DAPI) signal per voxel
[43, 44], as chromatin conformation is linked to DNA methylation and its bilateral relationship to histone tail modifications
. Therefore, we believe that cell-by-cell topological analysis as used in our approach, i.e. the topology of LIMs and LIDs in combination with the display of differential MeC/DAPI colocalization patterns shows a potential to serve as a valuable indicator for the observed phenomena: cytotoxicity-correlated global DNA hypomethylation and DNA reorganization, as consequences of drug effects. For the selected combinations of cell types and agents, the measurement of mean MeC signal (I
MeC) — a derivative of DNA methylation load, across all imaged cells — corresponded well with the level of cytotoxicity (Figure
4). However, for the majority of cases I
MeC presented a relatively high standard deviation, whereas for the same cell populations we observed a low fraction of dissimilar cells in terms of MeC/gDNA distribution (Figure
5). The discrepancy between the two signatures becomes more plausible with the analysis of synchronized DU145 cells: high similarity was measured between G0/G1-cells and G2-cells in MeC/DAPI codistribution (Figure
10). On the contrary, individual intensity values for global 5-methylcytosine (MeC) and overall DNA (DAPI) nearly doubled between G0/G1-phase and G2-phase as expected, although with a large spread in both signal distributions indicating high signal variability even in synchronized cells (Figure
11). Based on these findings, we believe that signatures based on spatial MeC/DAPI codistribution are more robust in MeC-phenotyping of cells than simply measuring DNA methylation loads, as they can better distinguish between drug-induced demethylation effects and the variation of methylation among individual cells. In combination with K-L divergence measurement, such a cell-by-cell cross-examination as performed with 3D-qDMI can provide structure-based quantities for studying epigenetic drug response.
Finally, in order to test the quantitative accuracy of 3D-qDMI a comparative analysis was performed utilizing MethyLight assays that have been specifically designed for and proven to measure differential levels of DNA methylation in repeat sequences such as Alu, Sat2, and Satα with high confidence
. These sequences are highly methylated in human cells and also represent a significant portion of their genomes. Therefore, they have been proven to serve as surrogates for measuring the global content of 5-methylcytosine in cells. Our comparative analyses revealed a significantly high degree of correlation between the outcomes of the two methods. We chose MethyLight as a validated technique over high-pressure liquid chromatography (HPLC), used as a standard method for measuring global DNA methylation: as the latter method requires significantly more input DNA (5–10 μg).
We conclude that the results of our work strongly support the idea of utilizing the spatial higher-order genome organization as a sentinel for drug-induced toxicity effects in liaison with global DNA hypomethylation. In particular, nuclear DNA methylation distribution patterns have proven to serve as an indicator of topological changes of the genome that could perturb spatial interactions of genomic loci and subsequent expression programs leading to cytotoxicity in treated cells. This is quite conceivable as it has been observed that DNA hypomethylation after treatment with DNMTi can be accompanied by additional changes in histone-tail modifications and nucleosome depletion that decrease DNA-repressive mechanisms and support a more open chromatin conformation
[83–86], an effect that we could reconcile with 3D image analysis for H3K9me3. A decrease in this repressive and compacting chromatin landmark with increasing doses of 5-azacytidine correlates well with a decrease in gDNA signal, and could be interpreted as chromatin decondensation (Figure
7). These downstream effects remain to be evaluated by determining the underlying molecular effects of possible cellular reprogramming, including the degree of heterochromatin demethylation
. Especially, the loss of global DNA methylation at heterochromatic areas of the genome that harbor highly repetitive DNA sequences such as highly abundant Alu repeats, transposable long interspersed nuclear elements (LINEs) and satellite DNAs can be associated with multiple risks towards genome instability
[8, 88]; through an adverse reorganization of the genome with side effects, such as transcriptional activation of oncogenes, activation of latent retrotransposons, chromosomal instability, and telomere elongation of chromosomes
[11, 89–92]. More specifically, Satα and Sat2 DNA hypomethylation may favor centromeric and pericentromeric instability, respectively. Alu retroelements, if left unchecked, would insert throughout the genome into non-coding and coding regions. The result would be mutations, and activation of oncogenes: spontaneous insertion of an Alu element causes nearby promoters to be hypomethylated, increasing gene expression
[93, 94]. Diseases directly associated with Alu insertion into coding regions include neurofibromatosis, haemophilia, agammaglobulinaemia, leukemia, breast cancer and ovarian cancer
. Any malignancy caused by Alu insertion is both heritable along somatic cell lines as well as in the germline. This concern has been recently strengthened by observations, in which specific genomic areas were found to become re-methylated during a following DNA replication step after initial drug-induced demethylation; as a possible mechanism to protect these sequences from permanent hypomethylation
. The study showed that exposure of cancer cells to agents such as 5-azacytidine and decitabine preferentially led to demethylation of CpGs not located in CpG-islands, whereas island-associated CpGs became preferentially re-methylated, suggesting that CG-dinucleotides in repetitive elements could become more persistently hypomethylated than gene-associated CGs.