Write a comment

In this manuscript, Taglini and colleagues explore the role of de novo DNA methyltransferase DNMT3B in human heterochromatin methylation. Using a human colorectal cancer cell line lacking DNMT3B, the authors demonstrate a specific loss of DNA methylation over pericentromeric DNA and constitutive heterochromatin, which are normally characterized by intermediate levels of DNA methylation and high H3K9 methylation. This model is consistent with previous studies that link DNMT3B loss of function mutations with pericentromeric DNA demethylation, genome instability and ICF-1 syndrome.

The authors then reconstituted DNMT3B-less cells with DNMT3B variants that carry a series of clinically relevant mutations, including the previously described S270P mutation in the PWWP domain, which was hypothesized to be the domain required for DNMT3B targeting to pericentromeric DNA. The authors also generated DNMT3B variants carrying PWWP domain mutations at paralogous amino acids to those that cause Heyn-Sproul-Jackson Syndrome (HESJAS) syndrome when mutated in DNMT3A (W263A and D266A) which result in methylation of DNA methylation valleys normally marked by H3K27me3. Unexpectedly, using two types of protein stability assays and Western blotting, the S270P mutation was shown to result in protein destabilization, conferring a loss of function mutation. Further characterization of pericentromeric DNA in revealed that S270P leads to loss of DNMT3B at heterochromatin hypomethylated pericentromeric DNA, likely due to protein activity loss. On the other hand, WGBS, southern blotting and bisulphite-PCR of W263A and D266A mutants show that both lead to an increase in pericentromeric and heterochromatin DNA methylation. Thus, while DNMT3B normally targets transcribed gene bodies and heterochromatin, mutations that alter PWWP function shift the balance of DNMT3B localization from genic regions to heterochromatin regions. Surprisingly, complete abrogation the PWWP domain does not result in DNMT3B redistribution, suggesting the W263A and D266A are gain of function mutations. Finally, the authors demonstrate that the uncharacterized N terminal of DNMT3B is required for heterochromatin targeting, but is dispensable in a W263A mutant context.

Overall, this is an excellently conducted and well written study. The authors employ orthogonal technologies to conclude DNMT3B’s role in the methylation of pericentromeric and constitutive heterochromatin DNA. While evidence already suggested DNMT3B associates with pericentromeric DNA in mouse via HP1a and Suv39h proteins (Lehnertz et al., 2003 Current Biology), Taglini and colleagues widened the scope of this interaction to generalize the association between DNMT3B and constitutive heterochromatin genome-wide. Furthermore, the integration of disease-relevant mutations deconvoluted a standing mystery in clinically observed DNMT3B mutation in ICF-1 syndrome.

Major comments

I wonder whether DNMT3A can compensate for the lack of DNMT3B in the ‘parental’ DNMT3B KO cell line. Both at gene bodies (which seems to be the case) and over heterochromatin. The authors could do a Southern on Satellite II DNA to check methylation levels in the DNMT3A/B double KO to confirm.

During my reading on the Results section it was unclear the extent of overlap between pericentromeric regions and H3K9me3 domains. If possible, the authors could include pericentromeric DNA annotations (if these exist) or the location of Satellite II DNA (again, assuming reliable annotations exist).

Does the ΔPWWP mutation change the stability of DNMT3B?

Do the H3K9me3 domains correspond to LADs? Could mine data from (van Schaik et al., 2020 EMBO Reports).

Minor comments

Rescue with DNMT3A instead of DNMT3B showed little gain of DNAme at H3K9me3 domains, revealing a difference in targeting of the DNMTs. However, if the cells normally express DNMT3A, this is an overexpression experiment, which could be discussed.

Could the authors speculate about the many isoforms of DNMT3B and how they may encode proteins with variable N terminal domains or missing PWWP? Such isoforms were discussed in a 2016 paper using the HCT116 cell line (Duymich et al., 2016 Nature Communications).

P5. “They also significantly overlapped heterochromatic regions resistant to nuclease digestion identified using Protect-seq in the same cells (Jaccard=0.623, p=2.11x10-30, Fisher’s test) (Spracklin and Pradhan, 2020)” has no associated figure, though I’m not sure one is required. Maybe a supplemental table?

What is the overlap of K9me3 domains in DNMT3B WT & KO cells?

The title of Fig 2c (right) could be improved as it’s not JUST W263A cells, but also 3BWT cells.

I think Figure 7 model is a tad ambiguous about K9me3/K27me3 marked regions and omits interesting biology, namely, that H3K27me3 is redistributed over K9me3-marked regions in DNMT3B KO cells. Further, different “PWWP mutations” have different effects on DNMT3B targeting.

Boxplot Fig 1c,e: Would be nice to know the absolute levels of DNAme instead of just the delta (maybe as a supplement).

Would also be nice to display the p-value between “other domains” and “H3K27me3 only domains” in Fig 1c.

Are the DNMT3B W263A and 266A mutations found in human diseases?

Are major satellite II / pericentromeric DNA regions transcribed in the various cell lines?

Similarly, how come H3K27me3 colocalizes with H3K9me3 in the absence of DNAme? See (Saksouk et al., 2014 Molecular Cell). Could the authors speculate?

It would be interesting to test whether any of the mutations generated and tested here modify the cooperative binding of DNMT3B with other proteins frequently mutated in human ICF, such as ZBTB24 (ICF-2), CDCA7 (ICF-3) or HELLS (ICF-4).

Competing interests

The author declares that they have no competing interests.