Joint single-cell profiling reveals hidden gene regulatory insights of 5mC and 5hmC
Figure 1 – Schematic representation of Joint-snhmC-seq, relative analysis and results obtained on cortical brain cells of mice (Created in https://BioRender.com).
Abstract
The methylation of cytosine (5mC) is one of the pillars of epigenetic regulation of gene expression. Currently the most common technique to study the methylome exploits the use of sodium bisulfite (BS-seq). However, despite the great efficiency, this approach does not allow to discriminate between 5mC and its most abundant oxidized form, the 5-hydroxymethyl cytosine (5hmC), hiding the distinct role of these modifications. Here we present research regarding Joint-single-nucleus-hmC-seq, an innovative technique able to obtain simultaneously the methylome and the hydroxymethylome profile at a single cell-level. By exploiting the cross-analysis with transcriptome data, researchers have highlighted a class of highly expressed genes characterized by an unusual epigenetic profile: high 5mCG, moderate 5hmCG and low unmodified CG. This suggests a potential non-canonical role of 5mC and 5hmC in promoting and supporting the transcription. These findings could be the starting point to further explore the function of these epigenetic modifications, a crucial aspect to improve our knowledge regarding the aetiology of developmental disease and tumour, allowing more effective therapeutical approaches.
Review
DNA cytosine methylation on carbon 5 (5mC) is a key epigenetic modification, fundamental to finely regulate gene expression among different cell types. Considering this role, it is also essential for the proper development of organs, including the neuronal tissue. Nevertheless, 5mC is not an irreversible modification. In fact, it can be removed by Ten-Eleven Translocase (TET) enzymes, a family of dioxygenase made up of three members: TET1, TET2 and TET3. All these enzymes catalyse the progressive oxidation of 5mC to 5-hydroxymethyl cytosine (5hmC), 5-formyl cytosine (5fC) and 5-carboxy cytosine (5caC). Later, through the base excision repair, the cell will remove the oxidized cytosine and re-establish the unmodified cytosine. Despite sharing the same catalytic cycle, these three enzymes are expressed in different developmental stages1. In addition, TET enzymes have a quite slow kinetics, so the most abundant intermediate of 5mC oxidation cycle is the 5hmC.
Currently the most common technique to analyse the methylome is based on the use of bisulfite, a molecule capable of deaminate cytosine, 5fC and 5caC to uracil. Despite recent progress to ameliorate this technique2, it still cannot distinguish between 5mC and 5hmC. To overcome this barrier, today have been developed other techniques able to discriminate between these two modified forms of cytosine3. On the other hand, most of these approaches require several purification steps, so do not allow single-cell analysis.
Resolve between 5mC and 5hmC could be crucial because 5hmC could not only be a simple intermediate of TET’s cycle but could also have a distinct functional role. For instance, it is known that methyl-CpG binding protein 2 (MECP2) tightly binds 5mC but does not interact with 5hmC4.
Consequently, has become a priority for researchers to develop a technique able to resolve 5hmC from 5mC at single-cell level5.
An innovative technique: Joint-snhmC sequencing
To resolve between 5hmC and 5mC, in this study researchers have developed an innovative hybrid technique based on chemical and enzymatic deamination called bisulfite-assisted APOBEC-coupled epigenetic sequencing (bACE-seq). The first step requires the usage of bisulfite, which simultaneously fragments, denatures DNA, deaminates 5fC and 5caC and protects 5hmC via formation of cytosine 5-methylenesulphonate (CMS). Then, apolipoprotein B messenger RNA editing enzyme catalytic subunit 3A (A3A) is added, a deaminase capable of removing the amine group from 5mC but not from CMS. So, proceeding with a step of sequencing with NGS, all cytosine that will be read indicate the selective presence of a 5hmC. In this way we can quantitatively and qualitatively identify the 5hmC sites on the genome (5hmC-seq).
Next, to adapt 5hmC-seq to single-cell analysis, researchers have integrated a double-barcoding workflow of DNA fragments, creating the single-nucleus-hmC sequencing technique (snhmC-seq). This approach enables high-throughput analysis of maximum 2304 nuclei.
To simultaneously investigate the methylome and hydroxymethylome profiles at single-cell level, the authors next developed Joint-single-nucleus-hmC-seq. This approach splits the genomic fragments derived from a single nucleus: one half undergoes snhmC-seq, the other one will be treated only with bisulfite (single-nucleus-5mC seq). At the end of the process, the hydroxymethylation profile is obtained from the first half, while they mapped both methylation and hydroxymethylation from the other. Thus, by subtracting the hydroxymethylation signal obtained by snhmC-seq from the signal obtained by snmC-seq, considering genomic regions of fixed dimension (100 kb), it is also possible to obtain the “true” methylation profile. This approach allows to simultaneously obtain two distinct epigenetic profiles at a single-cell level, maintaining a high quality of the analysis on both CpG and CpH (H=A, C and T).
Transcriptome cross-analysis reveals complex role of 5hmCG
It is known from former studies that cytosine methylation can correctly cluster main neuronal and non-neuronal cell-types2; the authors further investigated this aspect on 5hmC. They have applied the technique on brain cortical cells of mice (n=3). Then they have performed a clustering analysis based on hydroxymethylation data obtained from Joint-snhmC seq and they have compared the relative proportions of cellular population with those derived from Fluorescence-Activated Nucleus Sorting (FANS). Through this technique, they were able to divide neuronal and non-neuronal population, exploiting an antibody against a neuronal marker (NeuN), and excitatory neurons from inhibitory ones, using an antibody against an excitatory marker (Neurod6). From this comparison, it was observed an accuracy of almost 98.6%.
One of the main advantages of Joint-snhmC seq is the possibility to combine data obtained from this approach with those derived from other omics technique. In this study, researchers have made a cross-analysis between transcriptomic data, obtained from single-nucleus drop sequencing (sNucDrop-seq), and those regarding cytosine epigenetic modifications. They compared the UMAP obtained from the clustering analysis on transcriptomic data with three UMAPs derived respectively from 5mCG, 5mCG+5hmCG and 5mCH+5hmCH ones. From the assessment of discrepancy, they identify the 5mCG-based UMAP as the best clustering whilst the other ones displayed some limitations in clustering non-neuronal cell types.
To evaluate the relationship between epigenetic modifications and RNA levels, researchers started by performing a 5mCG+5hmCG-based clustering. The UMAP obtained has been annotated exploiting the fact that specific marker genes are hypomethylated in their respective cell subtypes. Subsequently, they generated heatmaps to visualize the RNA levels or epigenetic modification of specific gene among the cell types previously identified. As expected, they found a negative correlation between RNA levels and 5mCG in both neuronal and non-neuronal cell types, and lower levels of 5hmCG in non-neuronal population. Surprisingly, the correlation between 5hmCG and RNA levels is more complex across different cell types.
To better focus only on neuronal cells, they performed a clustering analysis using CH total methylation, based on the fact that this epigenetic modification is significantly more abundant in neuronal populations than non-neuronal ones5. They annotated the resulting UMAP as before and they have identified main excitatory and inhibitory neurons. They visualized again using heatmaps the RNA, 5mCG and 5hmCG levels, and, in addition, they also gathered data regarding unmodified CG level. In this case they also identified a negative correlation between RNA levels and 5mCG. Furthermore, they have highlighted higher level of 5hmCG in inhibitory neurons than in excitatory ones and the expected positive correlation between RNA levels and unmodified CG. Interestingly, they have observed a direct correlation between 5hmCG and RNA levels inside the same subtype, but a more complex relationship considering the expression of a single marker gene among different subtypes. This aspect could be confirmed by looking at relationship between 5hmCG and RNA levels of Arpp21, Erbb4 and Tle4 among different expressing subtypes. In fact, Arpp21 displays low level of 5hmCG and elevated level of unmodified CG where expressed, whereas Erbb4 displays the opposite pattern. On the other hand, the 5hmCG in Tle4 seems to act simply in a neutral way. All these findings could suggest a context-dependent role of 5hmCG.
Gene length dictates TET processivity
To further examine if the processivity of TET is related to cell type-specific gene regulation, researchers have visualized the relationship between unmodified CG, 5hmCG and 5mCG and expression patterns of marker genes inside a specific subtype through ternary plots. From previous findings1 they would expect that expressed genes should be characterized by high levels of unmodified CG and low levels of 5mCG. Nevertheless, this is partially true. In fact, there are some genes that are strongly expressed despite high level of CG methylation. From these findings authors have posited the presence of two different genes classes based on different TET processivity, deduced by the percentage of unmodified CG. The first class is charachterized by low TET processivity (unmodified CG<30%), high levels of 5mCG, moderate 5hmCG and higher gene lenght. On the other hand, the second class is characterized by high TET processivity (unmodified CG>30%), display opposite level of epigenetic modification and lower gene lenght. This hypothesis has been confirmed on both excitatory and inhibitory neuronal subtypes by plotting the correlation curve between 5hmCG and unmodified CG. They have noticed a positive correlation for class one genes, due to accumulation of 5hmCG caused by low TET processivity. On the other hand, the second class display the expected negative correlation, due to high TET processivity.
Suprisingly, they have found that class one genes show comparable or even higher levels of expression than class two ones despite the aspects previously described. These findings suggest a potential non-canonical role of 5hmCG and 5mCG in maintaining or promoting the expression of specific genes in mature cells.
Conclusions
In this study, the authors have developed a novel approach capable of overcoming the limitations of previous techniques. Joint-snhmC-seq enables the resolution of epigenetic ambiguity between 5mC and 5hmC and allow the directly mapping of methylome and hydroxymethylome at single-cell level. In addition, this technique offers practical and computational advantages, enabling the high-throughput analysis of low-input DNA and making easier the cross-analysis with other omics techniques.
This study unveils through an innovative approach a potential non-canonical role of 5hmC that could have an active function in promoting or maintaining transcription of specific gene in mature cells. This hypothesis could be the starting point to better understand the aetiology of neuronal-development disease and tumour, leading to possible therapeutical strategies6,7.
Personally, although the promising results obtained with this innovative technique, we think that the experimental procedure display some limitations. In fact, the bins dimension (100 kb) does not guarantee a resolution able to discriminate between crucial element of the gene, like promoter and enhancer, crucial for gene expression regulation. In addition, although the novel classification of marker genes based on TET processivity, we think that this correlation between gene-length and TET processivity could be biased from the fact that longer genes require much time to be demethylated by TET. From these considerations, future studies could build upon these aspects to further improve the precision and the resolution of this innovative technique, enabling more accurate, reliable and high-resolution single-cell mapping and a deeper understanding of gene regulatory mechanisms.
References
- Wu, Hao, and Yi Zhang. “Reversing DNA methylation: mechanisms, genomics, and biological functions.” Cell 156,1-2 (2014): 45-68. doi:10.1016/j.cell.2013.12.019
- Luo, Chongyuan et al. “Single-cell methylomes identify neuronal subtypes and regulatory elements in mammalian cortex.” Science (New York, N.Y.) 357,6351 (2017): 600-604. doi:10.1126/science.aan3351
- Mooijman, Dylan et al. “Single-cell 5hmC sequencing reveals chromosome-wide cell-to-cell variability and enables lineage reconstruction.” Nature biotechnology 34,8 (2016): 852-6. doi:10.1038/nbt.3598
- Kinde, Benyam et al. “Reading the unique DNA methylation landscape of the brain: Non-CpG methylation, hydroxymethylation, and MeCP2.” Proceedings of the National Academy of Sciences of the United States of America 112,22 (2015): 6800-6. doi:10.1073/pnas.1411269112
- Fabyanic, Emily B et al. “Joint single-cell profiling resolves 5mC and 5hmC and reveals their distinct gene regulatory effects.” Nature biotechnology 42,6 (2024): 960-974. doi:10.1038/s41587-023-01909-2
- Gabel, Harrison W et al. “Disruption of DNA-methylation-dependent long gene repression in Rett syndrome.” Nature 522,7554 (2015): 89-93. doi:10.1038/nature14319
- Zhao, Ying-Tao et al. “Long genes linked to autism spectrum disorders harbor broad enhancer-like chromatin domains.” Genome research 28,7 (2018): 933-942. doi:10.1101/gr.233775.117
