PWWP-ADD and N-terminal domains of DNMT3B1 confer specificity for developmentally regulated CpG island methylation

  1. Komal Yasmin
  2. Tatyana B Nesterova
  3. Neil Brockdorff

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

CpG methylation in mammalian genomes is established by the closely related de novo DNA methyltransferases DNMT3A and DNMT3B. Whilst both enzymes contribute to pervasive genome-wide CpG methylation, DNMT3B has a unique role in developmentally regulated CpG island (CGI) methylation both on the inactive X chromosome and at other sites in the genome. The mechanistic basis for this specificity is poorly understood. Here we have developed an in vitro embryonic stem cell model system to dissect critical determinants of DNMT3B specificity. Our model faithfully recapitulates developmentally regulated CGI methylation and additionally provides novel insights into CpG methylation at cis-regulatory elements. Using genetic complementation, we show that DNMT3B specificity is attributable solely to the catalytic isoform DNMT3B1. Domain swap experiments demonstrate an important role both for the PWWP-ADD chromatin binding and unstructured N-terminal domains. Together, these findings advance our mechanistic understanding of the unique roles of DNMT enzymes in establishing CpG methylation in development.

Article activity feed

  1. Review Commons

    Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

    Learn more at Review Commons

    Reply to the reviewers

    The authors do not wish to provide a response at this time

  2. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #3

    Evidence, reproducibility and clarity

    The study by Yasmin & colleagues tackles an important question, what is the molecular nature of specificity that arises from otherwise highly similar proteins. In this case, they focus on two proteins with epigenetic activity, DNMT3A and DNMT3B, using a functional readout of their ability to methylate DNA in a model that specifically requires DNMT3B function at a subset of the genome, i.e. DNMT3B-dependent regions. This includes characterizing the role of DNMT3B in these regions in stem cell-to-embryoid body differentiation experiments, using genomic assays to probe DNA methylation dynamics. By removing DNMT3B and ectopically expressing a variety of sophisticated mutants, the authors attempt to show the protein domains required for specificity. However, several questions remain about the strength of the data to support the claims, particularly with respect to the ectopically expressed mutant DNMT3B proteins.

    Major comments:

    1. The strength of this study is in the very nice addback strategy for probing DNMT3B specificity, where the designed mutants seem highly useful to ask critical questions. However, the stability of the mutant proteins (i.e. cellular expression levels) and question of protien levels in the nucleus are insufficient evidence for the conclusions stated in the paper. With the exception of the Dnmt3b1-KI clones (top panel fig 3B), it seems like most mutants are not expressed at wildtype levels. How much of the results are driven by differences in expression, relative to the wildtype protein? While this a technically challenging problem, there are various methods to establish roughly matched expression such as integration into a stronger locus for expression or tuning the promoter sequence for expression of a construct. Given the mutants are key for the main conclusions of the study, this seems critical to address, though would substantially increase effort required for the paper.
    2. Characterisation of the datasets supporting effects seems lacking in several instances. For example, the text states that DMNT3B null cells behave similarly to wildtype cells but supporting data (FigS2A-C) or that Dnmt3b1-KI and Dnmt3b3-KI behave normally with respect to differentiation (FigS4C), seem insufficient evidence for this, with largely summary plots supporting the statements. Similarly, several of the MBDseq datasets seem discordant, such as FigS2G or FigS4D(right panel) where the x-y axis for scatterplots are clearly not equivalent suggesting global effects on the data. The authors should also clearly demonstrate the levels of DNMT3A throughout their EB timecourse for mutant lines, where this seems especially important given their readout is DNA methylation dynamics.
    3. An optional analysis that could support the claims of the paper would be to contrast the effect sizes in their cellular model with existing datasets that profile DNA methylation dynamics in vivo, where these have been captured at early developmental levels. This would nicely show that their functional readout in relation to normal processes.

    Minor comments:

    1. Several figures require addressing, listed here:
      • Fig1B the points are not so legible when overplotted, consider reducing the size of the datapoint circles or turning into "*" representations.
      • Fig4I seems not to have a figure legend.
      • FigS2G should be represented as a square and not as a rectangle, as this visually condenses on axis relative to the other.
      • FigS3A is unclear, could more be added to the legend to describe what exactly is the schematic representing?
      • FigS3D the axis seems not aligned with the barplot positioning?
    2. The Dnmt3b-PAS-KI clone 1 does not seem to well-cluster with the 2nd and 3rd clone, could this be a clonal effect at the global level?
    3. The text states (page 7, third paragraph) that in the two differentiation models the identity of the CGIs that exhibit different dynamics largely match, though no direct comparison (i.e. delta-delta effect) is show, rather a summary plot of either is presented side-by-side. This seems insufficient evidence of the statement, and a direct comparison of the fold changes would help.
    4. The clonal effect sizes would benefit from more quantitative comparisons throughout the manuscript, broken down to raw data. For example, the statement in page 8 paragraph two that the effects on independent clones were fully consistent is show from largely a PCA plot, which seems incomplete evidence that replicates behave consistently. More transparent analysis of clonality from the raw data would be helpful for the reader.
    5. The statement in the discussion that the authors experimental system affords 'homogeneous and highly synchronised onset and progression of XCI", but it seems unclear from the data provided in the manuscript that cells exhibit differentiation in a synchronized manner. Softening this statement seems apt here.

    Significance

    The question of specificity is highly important, not just to the field of epigenetics and DNA methylation where this study is particularly relevant, but also to a broader audience. Many of our cells proteins are highly homologous but have nevertheless highly divergent activities. Molecular explanations of specificity are therefore critical to understand phenotype and how traits can be acquired through gene paralogue evolution. Here, by focusing on a particularly apt example, the similar DNMT3A/B proteins, this study offers a nice breakdown with the potential to tie back the results to locus specific activity in the genome. The strongest aspect is the comparison of sophisticated mutants in a matched experimental setting, however, the experiments do not seem sufficient to support the broad conclusions of the study. From a genomics standpoint, the experimental setup is impressive, but requires additional work to show that matched expression levels of wildtype/mutant proteins still maintains the phenotypes reported.

  3. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    DNA methylation controls gene expression and genome stability during development and in healthy adult cells. It is frequently abnormal in diseases including cancer. Therefore, there is a clear impetus for the community to better understand the mechanisms that underlie DNA methylation patterns in mammalian genomes, including how the mark is deposited on specific regions. Of particular interest are CpG islands, which correspond to the majority of promoters, and are typically devoid of methylation. However, in specific cases, including during differentiation and X chromosome inactivation, CpG islands acquire extensive DNA methylation in a de novo methylation process. There are 2 de novo DNA methyltransferases expressed in the embryo: DNMT3A and DNMT3B. They are globally similar, but only DNMT3B contributes to de novo DNA methylation during X inactivation (Gendrel 2012, from the authors' lab). The question of this paper is: why is that?

    The model system used by the authors is female mouse ES cells, which during differentiation in vitro inactivate one of their X chromosomes. They use a hybrid line to distinguish parental alleles, and a genetic trick to ensure that the same chromosome is inactivated in all cells. Figure 1 validates the system, showing that CGIs on the inactivated X acquire DNA methylation during the differentiation process into EBs, along with some autosomal CGIs (this is done by MBD-seq). Some are fast to gain methylation, some slower. A similar analysis is carried out during differentiation into NPCs.

    The authors then move on to functional experiments by knocking out DNMT3B in their system. The KO clones are extensively characterized (Fig 2 and S2). They lack DNMT3B but retain DNMT3A levels similar to WT. However, they fail to methylate the majority of CGIs during X inactivation, confirming that DNMT3B (and not DNMT3A) is the principal actor in this process.

    The next question is: which domain(s) of DNMT3B is/are involved in this function. For this, the authors rescue their KO clones with cDNAs encoding different isoforms of DNMT3B, namely 3B1 (active) and 3B3 (inactive). They found that 3B1 fully rescued proper DNA methylation and gene expression during differentiation, whereas 3B3 had no effect (Fig 3 and S4).

    Having found that 3B1 expression fully rescues the DNMT3B KO, they move on to a more precise delineation of the important domains (Fig 4). Domain-swapping experiments show that the catalytic domain itself does not contribute to the specificity of DNMT3B (see my note on this experiment below). A similar strategy is then employed to test the contribution of the PWWP and ADD domains to 3B function. I did not find this part very clear. My understanding is that the swapped rescue construct has some activity on gene bodies even before differentiation. I gather from the lower part of Panel 4F that the PAS construct mostly fails to rescue DNA methylation during differentiation, but I am a little confused by the phrasing. It would be very easy to solve this with a summary graphic.

    Lastly, they move on to an examination of the Nter part of DNMT3B, using their domain swapping/rescue approach (Figures 5 and S6). A first experiment suggests that the Nter of 3A cannot substitute for the Nter of 3B (see note below). A second set of experiments shows that the presence of residues 1-218 is necessary for DNMT3B to function, and that smaller deletions within this region also inactivate the protein (see notes below).

    The paper is very well written. The figures are clear. The experiments are well controlled and correctly interpreted, except for the points below.

    Fig 4A: it is looking like the DNMT3B1-3A-Cat rescue construct is vastly overexpressed relative to the endogenous protein. This is surprising as the DNMT3B1 rescue construct is not (Figure 3B). What is the reason for this? Is a different promoter or rescue method being used? I feel that the overexpression level weakens the conclusion that the catalytic domain of 3A can perfectly replace that of 3B.

    Fig 4G: similarly, the two different DNMT3B-PAS-KI clones show widely different levels of DNMT3B expression. Are they both used to generate the data of Fig 4H? A third rescue clone is shown in Fig S5B. What experiment(s) was it used for?

    The clarity of the section concerning DNMT3B-PAS-KI clones can be improved easily.

    Fig 5B: the DNMT3a2(N)3b-KI clones show 3B expression levels that are ~10% of endogenous. I find it hard to conclude from this that the Nter of 3A cannot replace the Nter of 3B. Unless the authors can show that a 10% level of 3B expression is enough to fully rescue the KO of 3B.

    Fig S6D: same comment for the DeltaA and DeltaB construct. I am not seeing the data for DeltaC. In the absence of expression data, the methylation data for this mutant are not interpretable.

    Fig 5B. Is it clear that the Delta 1-218 protein is nuclear? What about the Delta A-E mutants?

    I suggest that the authors add the following paper to their reference list: Wapenaar EMBO Rep 2024 PMID: 39528729

    Significance

    I feel that the target audience is somewhat specialized. In addition, the novelty of the paper is diminished by the existence of published papers, in particular: DNMT3B PWWP mutations cause hypermethylation of heterochromatin. Taglini F, ..., Sproul D. EMBO Rep. 2024 Mar;25(3):1130-1155. doi: 10.1038/s44319-024-00061-5. Epub 2024 Jan 30. PMID: 38291337

    This paper uses a different system (human cancer cells), but arrives at the same conclusion, ie the Nter of DNMT3B is necessary for de novo DNA methylation. In addition, it shows that the Nter interacts with HP1.

  4. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    Summary:

    Provide a short summary of the findings and key conclusions (including methodology and model system(s) where appropriate).

    In this study the authors used a previously established mESCs iXist-ChrX cell line to investigate the mechanisms underpinning developmentally regulated CGI methylation through differentiation into embryoid bodies and MBD-seq profiling. They show that their system recapitulates developmentally regulated DNA methylation at CGIs on the X chromosome and autosomes before using a knockout and rescue system to determine that this is dependent on the DNMT3B1 isoform. Through domain swap experiments, they then go on to suggest that this requires the PWWP-ADD domain and that the N-terminal region of DNMT3B1.

    Major comments:

    • Are the key conclusions convincing?
    • Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether?
    • Would additional experiments be essential to support the claims of the paper? Request additional experiments only where necessary for the paper as it is, and do not ask authors to open new lines of experimentation.
    • Are the suggested experiments realistic in terms of time and resources? It would help if you could add an estimated cost and time investment for substantial experiments.
    • Are the data and the methods presented in such a way that they can be reproduced?
    • Are the experiments adequately replicated and statistical analysis adequate?

    Overall, this is an interesting study that provides insights into the role of different domains of DNMT3B in establishing DNA methylation patterns during development. However, it suffers from poor annotation and description throughout. More specific comments are:

    1. The manuscript provides some details as to the replication of experiments but fails to show the replicate data in the vast majority of cases. Instead, representative data is presented alongside global correlations for some experiments. However, global correlations can mask differences between replicates. The data from all replicates should be shown in the manuscript and clear details provided regarding the replication strategy in the methods and figure legends. For example, were the different knock-in clones generated from independent DNMT3B knockout clones? For individual experiments this would be a minor point however, collectively this is a major point. It is particularly important given the variation highlighted in point 2 below.
    2. Different DNMT3B knock-in clones show high variability in expression levels. Have the authors investigated whether the discrepancy in protein levels contributes to the differences in methylation patterns seen? A non-comprehensive list of examples is given in the minor comments section.
    3. There is variation in the level of expression of different DNMT3B constructs detected by Western blot. Could this be caused by differences in protein stability? It would be helpful to see an assessment of protein stability to determine whether this contributes to the variable expression. For example, the DNMT3B3-KI has lower levels than the DNMT3B1-KI (Figure 3B) and this could potentially contribute to the differences in DNA methylation observed.
    4. The study lacks statistical tests to support the conclusions drawn from the analysis of the sequencing data. For example, are the differences in CGI methylation between DNMT3B1-KI and DNMT3B3-KI statistically significant? For individual analyses this would be a minor comment but given the lack throughout the study, this classes as a major comment.
    5. Chromatin marks play major roles in DNMT3A and DNMT3B recruitment (Tibben & Rothbart 2024) and the N-terminal region, PWWP and ADD domains have direct or indirect chromatin reading activity. However, the manuscript does not detail the chromatin environment of the CGIs studied. This could potentially be addressed through experiments, analysis of existing data or discussion.
    6. As the authors state in their discussion, MBD-seq may only detect very dense methylation. This could potentially obscure lower levels of DNA in some conditions. Analysis of a few loci by alternative methods, such as targeted bsPCR or EM-PCR would help support key results and rule out the possibility that some of the rescue constructs are able to partially rescue DNA methylation patterns.
    7. While some expression data is shown, there is currently no investigation as to whether the different DNMT3B domain swap constructs have impact on transcriptional silencing on Xi/autosomal sites.
    8. The text relating to the section on the PWWP-ADD domain is very brief and currently unclear. Expanding this section and specifying which data are derived from ES cells vs differentiated cells would help to clarify. We also suggest that it would be clearer to move this data to the main figure and to move the results of the catalytic domain experiments, which are negative, to the supplementary.
    9. The authors suggest that PWW-ADD domain region of DNMT3B is required for developmental methylation of Xi and autosomal CGIs. However, there is no further dissection as to whether this requirement is due to the PWWP, ADD or the intervening region.
    10. Throughout the text autosomal and Xi CGIs are both analysed. The introduction highlights SMCHD1 as important in methylating CGIs on Xi but that PCGF6 complex for the autosomal targets. This suggests separate mechanisms target DNMT3B to these loci. However, based on the results presented here, these two different types of targets have similar DNMT3B1 domain requirements. It would be interesting to see discussion with regard to this point in the manuscript.

    Minor comments:

    • Specific experimental issues that are easily addressable.
    • Are prior studies referenced appropriately?
    • Are the text and figures clear and accurate?
    • Do you have suggestions that would help the authors improve the presentation of their data and conclusions?
    1. Introduction: Methylation of tumour suppressor gene CGI promoters is mentioned alongside examples of developmental CGI methylation. It would be useful to place rare event in context. Methylation of CGIs is common in cancer but very few of these correspond to tumour suppressor genes. It would also be useful to discuss how DNMT3B might be involved in these events.
    2. Introduction: The authors mention the DNMT3A1's N-terminal region recruits it to H2AK119ub1. It would also be useful to discuss recent work on the N-terminal of DNMT3B which can bind HP1-alpha to mediate H3K9me3 recruitment (Taglini et al 2024). This paper is currently only cited in the discussion.
    3. Throughout the manuscript DNMT3B1 is referred to as the catalytic isoform, however it is not the only catalytic isoform of DNMT3B.
    4. Page 5: 'the presence of non-catalytic isoforms, notably DNMT3L and DNMT3B3'. This statement incorrectly suggests DNMT3L is a non-catalytic isoform of DNMT3B.
    5. The authors refer to the protein complex as heterodimers when referring to a DNMT3B -DNMT3L or DNMT3B3-DNMT3A complex. However, the consensus of structural studies is that they form tetramers.
    6. Marker sizes are not included on blots with the exception of Figure 3B.
    7. Western blots are cropped closely. It would be useful if full blots were shown in the supplementary particularly given the presence of extra bands in some blots and different DNMT3B isoforms.
    8. Details about the Western blot methods (eg visualisation and antibodies) are missing.
    9. It would be useful if the size of different groups was annotated in plots. This is given in Figure 1D for example but not in Figure 2E.
    10. The authors show data confirming DNMT3B knockout by western blot. However, they do not provide details of the strategy for generating the knockout (ie vector used, sgRNAs, screening process). Could the authors also provide additional details as to whether there is any sequencing to confirm the results on the knockout?
    11. Page 9: The authors state "Since Dnmt3b-/- cells have normal levels of DNMT3A", but show no data to support this statement. This is particularly relevant as they have generated new DNMT3B knockout clones for this study so they cannot be assumed to behave similarly to previous studies.
    12. Figure 1B: There is poor PCA clustering between replicates at some timepoints, particularly Day 8.
    13. Figure 1G: Different colours are used for the different timepoints in this figure. We are unclear if this is deliberate.
    14. Page 7: The authors state "... the two categories largely matched between the two differentiation systems (Figure S1C-G)". It is difficult to draw this interpretation from the data presented as no explicit overlap is shown.
    15. Figure 2E: MBD-seq peaks for DNMT3B-independent loci in WT sample have a dip in the middle of the peak (also seen in Figure 5F). Could the authors explain why this might be and why it only appears in some experiments?
    16. Clarification as to whether the DNMT3B -dependent and -independent loci are located on chromosome X or autosomes (e.g.: Figure 2D, E).
    17. Figure S2C: the chromatin RNA-seq is not explained in the text or figure.
    18. Figure S2E: Suggests WT is one of 5 replicates. The authors should show all replicates.
    19. Figure S2H: What genes are included in the metaplot?
    20. DNMT3B rescue knock-in clones. As shown in figure 2A, there are two different Dnmt3b-/- cell line clones. Could the authors clarify whether all the CRISPR KI clones are produced from the same parental Dnmt3b-/- cell line clone?
    21. Figure 3B: The two clones shown for Dnmt3b3-KI have variable expression. Do the individual replicates for the Dnmt3b3-KI clones show similar methylation patterns?
    22. Figure 3E: in the MBD-seq metaplots, there is a peak present at -/+ 4kb. What are these peaks and why do they appear at 4kb distance? Similar peaks are seen in other metaplots.
    23. In figure 3F, the signal for both Dnmt3b1-KI and Dnmt3b3-KI at DNMT3B-independent CGIs is higher than in the KO. This suggests that these may not be DNMT3B independent but this point is unaddressed in the text.
    24. Figure S3A: it is currently unclear what has been modelled in this figure, adding labels of what has been plotted along the x- and y- axis may aid in understanding.
    25. Figure 4A: Dnmt3b1-3a-Cat-KI appears very highly expressed. Is the WT shown the endogenous protein? Could this higher expression be because the chimeric protein is more stable than DNMT3B1? There are also multiple bands on this blot.
    26. Plots panels are inconsistently ordered, e.g.: Figure 3F is dependent then independent. 4F is independent then dependent.
    27. Figure 4G: the expression level of the Dnmt3b-PAS-KI varies significantly between the clones shown. There are also two bands on the blot, both for the wt and KI. Clarify if WT is endogenous.
    28. Figure 4H: The figure lacks a legend to indicate the scale of the colour density used.
    29. Figure 4F,H: Could the authors clarifying what data (clones and number of replicates) are presented in the representative plots. Does the different protein levels between the clones result in any differences in DNA methylation?
    30. Page 12: The authors cite Boyko et al when discussing potential differences between the ADD domains of DNMT3A and B. However, they do not cite the study of Lu et al., 2023 (https://doi.org/10.1093/nar/gkad972) which reaches a different conclusion.
    31. Figure S4A: The position of the 750 residue is inconsistent across the isoforms in this schematic.
    32. Figure 5A: Schematic suggests the chimeric protein is DNMT3A2(N)-3B. However, DNMT3A2 lacks the N terminal region so presumably this should be DNMT3A1(N)-3B. This applies other figure panels using this construct.
    33. Figure 5B: Many lanes on this blot are unlabelled and it would be useful to clarify what these extra lanes show.
    34. Figure 5C: For Dnmt3a2(N)3b-KI the levels of methylation appear to be lower than Dnmt3b-/- and it would be useful to understand why this might be the case.
    35. Figure 5D: would be helpful to indicate which CGIs are DNMT3B dependent and independent.
    36. Figure 5F: Dnmt3a2(N)3b-KI data not included for the autosomal peaks
    37. Figure 5G, H: It is difficult to see if there are any differences between the deletions in this heatmap. For example, it appears that levels of methylation on autosomal DNMT3B-dependent loci are very similar between the KO and rescue constructs. ∆D also appears to have a lesser effect than the other deletions on the Xi CGIs. A more quantitative representation of the data would help with interpretation.
    38. S5E has different colour scale to other heatmaps. Red is low and in other heatmaps red is high.
    39. Figure S6C: sequence conservation is shown for primates. However as mouse Dnmt3b is used throughout the paper, including the mouse NT would be a useful comparison. This is particularly relevant given that the NT is the region that varies the most between mouse and human proteins (Molaro et al 2020 https://doi.org/10.1093/molbev/msaa044 ).
    40. Figure S6D: There is variable expression levels between the clones of the different deletions. The deletion ∆C is also not shown in this figure meaning that no data is shown to support the statement that it is unstable.
    41. It would be useful to clarify in the text that "deletion of residues 98-146" corresponds to ∆C. It is also unclear why MBD-seq data for this deletion was included if it is unstable.
    42. Discussion, page 15: The authors propose that DNMT3B could directly bind to H3K9me3. However, a study they cite, Taglini et al., 2024 (Figure S8C, D), suggests this is not the case.
    43. Discussion: When discussing regulatory element methylation on Xi. An uncited statement is included: 'This observation may help to explain a prior observation that loss of DNMT3B1 alone does not result in significant de-repression of Xi during embryogenesis'. However this model appears contradictory to the observation that DNMT1 is not required for Xi silencing, given that DNMT1 KO embryos would be expected to have very low DNA methylation (eg Sado et al 2000, https://doi.org/10.1006/dbio.2000.9823 and Sado et al 2004, https://doi.org/10.1242/dev.00995).

    Referee cross-commenting

    Having reviewed the comments of the other reviewers, we agree that they are very similar and we have no issues with them. We note that Reviewer 3 notes that considering nuclear protein levels is important in the context of this study and we agree that this is an important additional consideration that we did not consider in our review.

    Significance

    The manuscript is an interesting study on the role of DNMT3B in X inactivation and development. It will be of interest to scientists who work on these fundamental processes. In addition, given the roles of DNA methylation in gene regulation, cancer, aging and disease more generally the findings are likely to be of interest to many others.

    Our expertise is in epigenetics and its regulation in disease, with a specific focus on DNA methylation and DNMTs.

  5. Version published to 10.1101/2024.10.22.619628v1 on bioRxiv