الانتقال إلى مراجعة PREreview

مراجعة PREreview لChromatin boundary permeability is controlled by CTCF conformational ensembles

نُشرت
DOI
10.5281/zenodo.24323389
الترخيص
CC BY 4.0

Summary of the study

The authors ask how CTCF determines, within the few tens of milliseconds of an encounter, whether to stall or bypass an extruding cohesin complex. They propose that boundary permeability is encoded not by CTCF occupancy alone but by a rapidly interconverting ensemble of DNA-bound CTCF conformations, generated by the association and dissociation of blocks of zinc fingers. Using nanopore GpC footprinting in HEK293T cells, they find that cohesin-rich CTCF sites show a partially protected (“unwrapped”) footprint, supported by MD simulations. Using an optical tweezers DNA unzipping assay, they show that single CTCF molecules switch between four states on a millisecond timescale comparable to cohesin passage, and that motif sequence, CpG methylation, a nearby nucleosome, and the cohesin regulator PDS5A shift this ensemble. They further report that PDS5A docks onto CTCF and, together with cohesin, is associated with a subpopulation of CTCF-DNA complexes of unusually high rupture force. This leads them to propose a two-step model in which PDS5A first primes CTCF for cohesin capture and then helps convert the captured complex into a stable loop anchor. This question is important, and the shift from an occupancy-based to a state-based view of boundary function is novel, thought-provoking, and of clear interest to the genome-organization field.

Overall, the approaches are well suited to the question, and the observations are consistent with the proposed model. However, several key links in the model are not yet directly supported by the data: the direct interaction between PDS5A and the CTCF zinc fingers, the proposed order of events (PDS5A before cohesin), and the functional link between ensemble composition and cohesin stalling. These currently rest on correlation and inference, and testing them directly would considerably strengthen the conclusions.

Major points

Major point 1: the claim for a direct PDS5A-CTCF interaction would be strengthened by additional evidence.

The model proposes that PDS5A binds CTCF, including its zinc fingers, and shifts the ensemble toward the capture-competent state. The current evidence for this is limited, for several reasons:

· PDS5A raises conformation II occupancy only from 76% to 83% (Fig. 3G), a modest change.

· The docking experiment (Fig. 3D-E, S22) would be easier to interpret with a control, such as ΔN CTCF.

· The AFM data (Fig. 3F) localize PDS5A broadly (median 106 bp from the motif, 95% CI −160 to 297 bp). This is compatible with association near CTCF but does not show direct contact.

· The predicted PDS5A-zinc finger contacts rest on a single AlphaFold 3 model that the authors themselves describe as lower confidence. The PAE plot (Fig. S25) appears to show high expected error in this region, which makes the prediction difficult to weigh.

· The unzipping data (Fig. 4B, 4F) show that PDS5A alters the mechanics of CTCF-DNA complexes, and the loss of the C-terminal effect with ΔN CTCF (Fig. 4B) is consistent with the known IDR-mediated recruitment. However, the high-force N-terminal population (Fig. 4F) rests on 3 of 96 molecules with PDS5A alone, and a change in rupture force does not by itself demonstrate binding to the zinc fingers.

No direct binding measurement is currently presented. A quantitative assay with purified proteins (e.g., SPR or ITC), comparing full-length, ΔN, and ZF-only CTCF with and without DNA, along with the docking control mentioned above, would be an informative way to demonstrate the interaction. Mutating the predicted contact residues would test the interface more stringently, and reporting ipTM across seeds would help readers judge how much weight the structural prediction can bear.

Major point 2: the proposed order of events (PDS5A first, cohesin second) would benefit from direct experimental support.

The model (Fig. 5) proposes that PDS5A pre-binds CTCF and primes it before NIPBL-cohesin arrives, after which PDS5A exchanges with NIPBL to form a stable anchor. Even if direct binding were established, this sequence of events has not yet been tested:

· PDS5A can associate with CTCF without cohesin (Fig. 3D-F), but this does not show that it does so first, or that it changes the outcome of an encounter with NIPBL-cohesin. The alternative the authors raise, that PDS5A arrives after NIPBL-cohesin pauses at CTCF, is therefore not excluded.

· The ChIP-seq correlations (Fig. 3B-C) are static and cannot resolve temporal order.

· The loss of the ensemble shift when cohesin is present (72% conformation II occupancy; Fig. 3G) is read as PDS5A acting “upstream”, but it could equally mean that the effect is lost once cohesin is bound.

· The exchange of PDS5A for NIPBL is inferred from suppression of a rare high-force population (Fig. 4F) rather than observed directly.

· According to the Methods, CTCF, PDS5A, and cohesin appear to have been present together (with PDS5A pre-incubated with cohesin), and in the confocal assay PDS5A was added to DNA with pre-bound CTCF. No experiment tests PDS5A pre-binding to CTCF followed by NIPBL-cohesin.

Order-of-addition experiments would be a valuable way to address this, for example in an in vitro loop extrusion assay with CTCF pre-incubated with or without PDS5A before NIPBL-cohesin is introduced. In the absence of such data, presenting the sequence as a hypothesis in the title, abstract and Fig. 5 would more accurately reflect what the experiments show.

Major point 3: the link between the unwrapped conformation and cohesin capture could be strengthened with a functional test.

The central conclusion is that ensemble composition controls cohesin capture and boundary permeability, with conformation II as the capture-competent state. At present this rests on indirect evidence: a correlation between the GpC footprint and RAD21 ChIP-seq (Fig. 1E), the high occupancy of conformation II (Fig. 2F), and ensemble shifts under perturbations (sequence, methylation, nucleosome, PDS5A) that are assumed to alter boundary strength. No experiment measures cohesin stalling or bypass as a function of ensemble composition.

In vitro loop extrusion assays with CTCF barriers exist (PMID: 37076620), so the central prediction appears to be testable: if ensemble composition governs capture, the probability of cohesin stalling should follow the ranking predicted by the model across the substrates used here, including the effect of PDS5A pre-binding. Without this, the causal wording of the title and abstract (“is controlled by”, “gates cohesin capture”) goes somewhat beyond what the correlative evidence supports, and a slightly more tempered phrasing could be considered.

Two additional points could further help the reader. First, it would be valuable to understand why conformation II is capture-competent whereas the similarly unwrapped conformation III is not, ideally with an explanation grounded in which zinc fingers are engaged in each state. Second, a kinetic model showing that the measured rates and the 15-50 ms encounter window can account for the reported stalling efficiency would make the link between single-molecule dynamics and boundary function much more convincing.

Major point 4: the quantitative difference between the in-cell and in vitro conformational distributions would benefit from discussion.

In cells, ~60% of CTCF-bound molecules are protected and ~40% partially protected (Fig. 1C). In the tweezers experiments, the fully bound state I accounts for only ~10% of the time and states II-IV for ~90% (Fig. 2F, Table S2). The study treats the two datasets as reporting the same wrapped/unwrapped ensemble, but this difference is not addressed, and it bears on how directly the in vitro states can be mapped onto the in-cell footprints.

Several factors could plausibly contribute: averaging over many motifs of varying strength in cells versus mainly one sequence in vitro, the lack of chromatin context and cofactors in vitro, and the dependence of the partially protected fraction on the number of ZFs covered per read (Fig. S1C). I appreciate that identifying the origin experimentally may be challenging. An explicit discussion would nonetheless be a helpful addition, ideally supported by an analysis such as comparing the in-cell footprint at the PDGFRA locus itself, and in motif-strength-stratified subsets (including consensus-like motifs), with the in vitro data for the same sequences. This would help readers see whether the difference reflects sequence heterogeneity in the in-cell average or a genuine difference between the systems.

Minor points

Minor point 1: clearer, consistent labeling would help readers follow the experiments.

The wild-type CTCF condition on the PDGFRA insulator is labeled “Insulator + WT CTCF” in Fig. 2H but “CTCF only” in Fig. 3G. It would help readers if the labeling were consistent across figures.

Minor point 2: some details on the protein reagents would help readers interpret the data.

The Methods indicate that both human CTCF (full-length, ΔN, ΔC) and mouse CTCF (with an alternative consensus motif) were used, but the figure legends do not state which was used in each main-figure experiment. Evidence, shown or cited, that the Halo/JF646 label and tags do not alter DNA binding or dynamics would increase readers’ confidence in the measurements. A comment on how the concentrations and equimolar stoichiometries used (3-50 nM) compare to estimated nuclear levels, particularly for the PDS5A and cohesin experiments, would help readers judge physiological relevance.

Minor point 3: the terminology for conformational states could be made more consistent.

“Wrapped”, “unwrapped” and “conformation” are used for different measurements without definitions. In the GpC data (Fig. 1), “unwrapped” refers to dissociation of peripheral ZFs (ZF1-2 and ZF9-11), whereas in the tweezers data (Fig. 2) it is applied to state II (N-terminal ZFs released) and elsewhere to state III (C-terminal ZFs bound only). As a result, the correspondence between the footprint classes and the four HMM states is not clear. Defining each term by the ZFs bound, with a schematic or table showing how the GpC categories to the HMM states and consistent labels throughout, would make the article easier to follow.

I am one of 2 participants from the same research lab who reviewed this preprint as part of the Transparent and Accountable Peer Review training program. This report was written with the support of my research mentor.

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.