Article activity feed

  1. Evaluation Summary:

    The authors aim to understand how certain transposable elements escape chromatin-based silencing. Focusing on variably methylated copies of IAP (VM-IAPs) in the mouse, the authors show that elements that can escape silencing share sequence variations that alter KRAB zinc finger protein (KZFP) binding and KAP1 recruitment, proximity to expressed genes and high CpG content. Analysis of human elements in human KZFP-free mouse cells recapitulates some of these observations. The authors propose that ZF-CxxxC proteins play a role in establishing permissive chromatin at transposable elements that harbor high CpG content and weak KZFP binding. The data are mostly correlative and open the path for further mechanistic analyses. The paper is of interest to readers in the field of epigenetics, genome biology and transposable elements.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #3 agreed to share their name with the authors.)

  2. Reviewer #1 (Public Review):

    The work aims to increase our understanding of the relationship between young transposable elements and repressive mechanisms. Specifically, it is focused on variable methylation patterns observed at a subset of elements called metastable epialleles. First, the authors provide an analysis based on sequence clustering and CpG content of mouse transposable elements. They then extend their analysis to human elements, recapitulating some of their observations. Finally, they showcase a new hypothesis suggesting that CFP1 plays a role in the mechanism of establishment of these methylation patterns.


    - The way it sometimes presents information about classification of VM-IAPs elements is interesting. Even if the findings have been recently published elsewhere, the ChIP signal overlaid on multiple alignment helps pinpoint the exact regions that are susceptible to be impactful. However it is puzzling why they didn't use the same approach for Figure 2.
    - The analysis of human transposons showing that those that are known to be variably methylated share some of the same patterns (High CpG and evolutionary recent) as VM-IAP is of interest.
    - The demonstration that Trim28 haploinsufficiency is reactivating many young transposable elements is informative, although I would have expected more detail about VM-IAPs and a breakdown per clade for IAPs to link it with the rest of the manuscript.


    - A lot of the first half of the article have been described before (in Bertozzi et al, eLife 2021, Elmer et al, Mobile DNA 2020 and Bertozzi et al., PNAS 2020. While this is acknowledged, I feel there could be more new insights presented from an in-depth analysis - for example with a zoom-in on precise motifs / binding sites being different between clades.
    - The use of public datasets of various strains is a source of potential concern in this specific case - VM-IAPs and their KZFP controllers are sometimes strain specific, and information should be provided about the precise strains for every public dataset used. The same strain should be used whenever possible.
    - Similarly, the use of cell line as proxies for liver human data when there is primary tissue datasets available (from GTEx for example) is less than ideal.
    - The absence of DNA methylation when KZFPs are not present, such as in their Tc21 model, is expected and previously described (Wiznerowicz, JBC 2007).
    - Most importantly, the CPF1 results do not demonstrate a causal relationship as the authors claim - all I see is that unprotected IAPs are accessible to DNA binding proteins, which is not surprising in the slightest - it is the case for CPF1, and I'm sure many dozens of other transcription factors.

  3. Reviewer #2 (Public Review):

    ERVs constitute around 10% of the mouse genome and are usually kept under control through host silencing mechanisms. However, a few ERV families are able to escape such control, as IAP elements. In "Mechanisms regulating interindividual epigenetic variability at transposable elements", Costello and colleagues address the establishment and maintenance of IAP epialleles by thoroughly searching IAP sequence variants that match activation signals (lack of silencing complexes and presence of permissive chromatin marks). More precisely, the authors take advantage of a previously published set of IAP loci that are variably methylated between individuals (VM-IAP), to perform multiple sequence alignment and hierarchical clustering of IAP LTRs, and IAP-ez internal regions, and search for VM-IAP enrichment in the obtained sequence clusters. The authors are able to demonstrate that VM-IAPs are enriched in specific clades, and such clades lack KZFB binding, and show decrease KAP1 enrichment. The authors also suggest that VM-IAPs are mainly found within 50Kb of a constitutively expressed gene, and conclude that this proximity might be necessary for variable IAP methylation. Finally, the authors broadly study CpG content and chromatin regulation, and suggest that permissive chromatin marks can be deposited at VM-IAPLTRs thanks to CpG rich regions. The final model suggests endogenous retrovirus that harbor high CpG content and also loss of KZFP binding, are the target of ZF-CxxC proteins, recruiting K4me3 to the IAP copy.

    In general, I find the article to be well written and the results robust. The conclusions are of course complicated as they illustrate the biology of IAP regulation: it depends on the sequence, it depends on the chromatin context, it depends on the genomic background. I find the authors could discuss such complexity, specially within the framework of already published research articles on IAP regulation.

    The usage of transchromosomic mouse model is elegant to show the equilibrium between KZFP targeting and K4me3 deposition, along with the Trim28 haploinsufficiency experiment. However, one of the main weaknesses relies on the material and methods section, where unfortunately it is difficult to understand how some of the analysis were performed: uniquely/multimapped ChIP-seq, alignment of LTR elements belonging to full-length copies, and VM vs non-VM comparisons.

  4. Reviewer #3 (Public Review):

    Building on previous studies that identified ~50 so called metastable alleles driven by transposable elements (TEs) that exhibit variable methylation (VM) between genetically identical mice, this paper aims to assess whether genetic sequence and sequence-specific TE-binding genetic modifiers contribute to VM in mammals. To investigate genetic determinants of metastable alleles, the authors focus primarily on mouse-specific TEs of the IAPLTR1 and 2 class known to constitute the majority of mouse metastable alleles, but also provide functional data on human VM- LTR12C elements.


    The paper identifies sequence variants that characterize IAPLTR1 and 2 elements enriched for VM-IAPs and makes use of a broad array of published sequencing data to characterize these sequence variants further. As sequence variants alone are not sufficient to establish VM-IAPs this bioinformatic analysis is important to advance the understanding of metastable alleles. The use of a diverse array of available datasets from different tissues is appropriate as VM-IAPs are generally retained across adult tissues. By looking into the known genetic modifiers of the KZFP family (previously implicated in metastability and TE repression) at IAP-LTR1 elements the authors show that sequence variants enriched for VM-IAPs correlate with diminished sequence-variant-specific KZFP binding and diminished binding of the KZFP co-repressor KAP1. This observation indicates escape of sequences enriched in VM-TEs from silencing by the KZFP/KAP1 system that evolved to repress TEs. The authors convincingly show that these VM-prone IAP-LTR1 variants are also the least conserved across mouse strains, a correlation that also holds true in other mammals, as previously identified human variably methylated loci are also shown to be enriched for young TEs.

    As diminished KZFP binding occurs across VM and non-VM IAPLTR1 elements alike, the paper uses annotated regulatory sequences and proximity to constitutively expressed genes as a proxy for a euchromatic context, which they find enriched at VM, but not no-VM-IAPLTR1 elements suggesting that this chromatin context contributes to establishing metastable alleles. Like others before them, the authors find mouse and human VM young TEs to have also a particularly high CpG density. As CpG density is also shared among VM and non-VM TEs in mice the authors assess the binding of the CxxC-domain containing protein TET1, known to bind to CpG rich regions, and the active histone mark H3K4me3, which they find both enriched specifically at VM-prone IAPLTR1 sequence variants. While expected, this correlation is novel in the context of metastable alleles and lets the authors propose the interesting hypothesis that competition between KRAB and CxxC proteins may contribute to establish metastable alleles. Notably, H3K4me3 at ERVs has also been found to be enriched upon KZFP or KAP1 depletion in mouse ESCs, consistent with their model.

    To functionally test the model build on these correlations, the authors use an elegant Tc1 mouse model that harbors human chr21 and with it human-specific variably methylated TE-derived CpG islands. The remaining mouse genome lacks the cognate human KZFPs encoded on other chromosomes not present in the mouse. This model is thus valid to test the hypothesis. The gain of the mouse ZF-CxxC protein CFP1 at a variably methylated LTR12C element in the Tc1 model, but not in human HepG2 cells (that encode both human TEs and cognate KZFPs) convincingly shows that the VM LTR12C element gains binding of the CxxC protein CFP1 when KZFPs are absent and thus supports the main claim of the paper. Supporting their claim, TE-derived sequences targeted by KZFP shared between mouse and human were highly methylated in this model and largely not bound by CFP1 regardless of their CpG content. Thus, their experimental data suggest that TEs that are VM can gain CFP1 binding when their cognate KZFP is missing.

    To support that the absence of CxxC domain containing proteins causes changes in methylation, the authors use available mouse DKO cells for the ZF-CxxC proteins TET2 and 3. In this model, the absence of ZF-CxxC proteins leads to increased methylation at VM-IAPs, but not non-VM IAPs thus supporting a causative role of CxxC proteins in maintaining a hypomethylated state at metastable alleles in mice.

    The hypothesis that TEs prone to VM rely on escaping the KZP system is further substantiated by profiling of CFP1 binding and H4K4me3 in livers of mice lacking one copy of KAP1, a universal co-repressor used by KZFPs to repress TEs. In livers of these mice, which are known to affect metastability at the agouti coat color gene, ~70 loci gain H3K4me3. These loci susceptible to KAP1 levels are largely young TEs bound by CFP1. Analysis of upregulated TEs in these mice identifies them as particularly CpG-rich supporting a model in which reduced KAP1-mediated silencing would allow CpG-rich TEs to be bound by CFP1 to potentially become metastable alleles.


    The authors propose that specific sequence features of the subclass of IAPLTR1 and 2 elements that often become VM-IAPs lead to lowered KZFP recruitment and thus 'escape' from KZFP-KAP1 mediated silencing. However, the authors do neither describe the specific sequence changes they refer to, which seems relevant, nor call sequence motifs over these sequences, which could uncover whether it is indeed direct competition between KZFPs and CxxC domain containing proteins that leads to the establishment of VM-IAPLTRs.

    Similarly, decreased KAP1 recruitment is central to the conclusions of the paper that TEs prone to be VM escape KZFP repression, but KAP1 coverage heatmaps are only shown for IAPLTR1 regions. It thus remains unknown whether KAP1 recruitment is also decreased at internal IAPez sequences flanked by these LTRs that also show diminished mouse-specific KZFP binding or at VM-IAPLTR2s. The lack of the analysis of KAP1 binding seems particularly relevant for the latter IAPLTR2s, as these seem not to lack KZFP binding, at least for the KZFPs assessed (the KZFP ZFP429 (S1.6)). That these class1 IAPLTR2s are nevertheless often VM seems to contradict the model, unless the cognate KZFPs are less efficient at repressing the TE, which Kap1 density plots could reveal. Alternatively, class 1 IAPLTR2s could depend on diminished recruitment of another KZFP or be KZFP-independent. Currently, such analyses are however challenging, as the genomic binding sites of most mouse KZFPs (unlike human, which are better characterized) are unknown. Similarly, internal sequences flanked by class1 IAPLTR2s, unlike class 3 IAPLTRs, are not proximal to the mouse-specific KZFP Gm14419. Here, this lack of diminished KZFP binding ,may however reflect that most VM-IAPLTR2 sequences are solo-LTRs.

    While the human data in the Tc1 mouse model provides convincing evidence for the main conclusion of the paper, such evidence is not presented for mouse, which is surprising given the focus of the paper on mouse-specific VM-IAPs. The majority of the mouse data (except for the TET2/3 DKO and KAP1-/+ liver experiment) are correlative. While the correlations characterizing the identified sequence variants in mice support the main hypothesis, they e.g. do not demonstrate that sequence contributes to establish VM-IAPs, that genomic context influences variable methylation or that high CpG density is important to establish variable methylation and also does not show that mouse TET1 is recruited when mouse KZFPs are absent.

    This concern also applies to conclusions made on some of the human data. While the human data does demonstrate CFP1 recruitment in the absence of human KZFPs at a LTR12C element, it does not demonstrate that CpG density establishes LTR12C elements as variably methylated. Rather it shows that VM LTR12C elements are also CpG dense, which is a correlation.

    The authors suggest that VM-IAPs have unique sequence variants that allow them to have diminished KZFP recruitment and recruit CxxC-domain containing proteins, but only if the genetic context permits it (as sequence variants characterize both VM and non-VM IAPs). Other than being euchromatic it remains open, what exactly this genetic context entails and if it indeed causes VM.

    Like others before, the authors observe high CpG content at VM-IAPs and propose that this hypomethylation is caused by the recruitment of CxxC domain containing proteins such as TET1 and CFP1 that can induce active histone marks or DNA demethylation, respectively. It would require further functional studies to understand the mechanism of diminished KZFP recruitment in establishment of VM despite identical sequence to non-VM-IAPs that also have diminished KZFP recruitment.

    The conclusions of this work will likely stimulate and impact the further study of KZFP and CxxC proteins in the establishment of metastable alleles. The elegant use of the Tc1 mouse model could be exploited by the field to dissect the mechanism that establishes metastable alleles further. Given the recent discovery of abundant species-specific KZFP binding to species-specific TEs in human and mice, this work adds to the significance of this binding not just for mere TE repression but gene regulation with relevance to biological processes, in this case the establishment of metastable alleles. As metastable alleles in mice control phenotypes such as coat color and obesity, these results are relevant for our understanding of phenotypic variation in genetic identical individuals. It will be exciting to elucidate the mechanism by which KZFP and CxxC proteins confer metastability to some but not other species-specific TEs.