Do changes in 3D contact explain enhancer regulation after cohesin depletion?

Widespread contact loss, few expression changes

Cohesin is a ring-shaped protein complex that extrudes DNA loops; depleting it erases loops and contact domains across the genome. The puzzle the paper starts from is quantitative: after 6 hours of auxin-induced cohesin depletion in HCT-116 colorectal cancer cells, only 146 genes changed expression by more than 1.75-fold, even though 3D architecture was broadly disrupted. If enhancers work by physically contacting promoters, why does losing those contacts leave most transcription intact?

The authors' answer is that contact loss is not uniform. It is concentrated at long-range enhancer-promoter pairs, and only enhancers that actually lose contact lose regulatory potency. That reframes the question from whether cohesin matters at all to which enhancer-promoter pairs depend on it.

Activity-by-contact predicts which pairs are lost

To map enhancer-gene pairs before and after depletion, the study used the Activity-by-Contact (ABC) model. ABC scores each candidate element-gene pair as enhancer activity multiplied by enhancer-promoter contact frequency, normalized against all other elements within 5 Mb of the gene. Activity came from DNase-seq accessibility and H3K27ac ChIP-seq; contact came from matched ENCODE Hi-C at 5 Kb resolution. A pair was called an enhancer interaction if its ABC score passed 0.027, a threshold set to reach 70% recall against prior K562 CRISPRi data.

Because activity stayed largely stable between conditions while contacts changed, ABC score changes closely tracked contact changes. That makes ABC a useful readout of contact-dependent regulatory loss, but it is still a prediction: the model assumes an enhancer's effect scales with activity times contact, and the paper tests that assumption experimentally rather than taking it as given.

Contact loss concentrates beyond 50 Kb

The distance dependence is the clearest quantitative result. Enhancer-promoter pairs more than 50 Kb apart lost 62.2% of contact frequency on average after cohesin depletion, while pairs under 50 Kb lost 18.6%. The authors treat 50 Kb as the point where contact frequency changes by roughly twofold, and use it to define distal versus proximal elements.

That asymmetry reshapes the predicted regulatory landscape. In the presence of cohesin, ABC called 28,575 enhancer-gene pairs across 10,393 genes; after depletion, 21,523 pairs across 9,489 genes remained. Of pairs more than 50 Kb apart, 84.0% were lost, versus 23.7% of pairs under 50 Kb, and the 90th percentile interaction distance fell from 156.3 Kb to 69.2 Kb. The ENCODE-rE2G model, a logistic-regression alternative built on the same data, reproduced the same preference for losing distal enhancers.

Distance is not the whole story. Contact reduction varied among distal enhancers and did not collapse to a single distance-decay curve, and pairs near CTCF sites and away from them both lost contact. The authors interpret this as evidence that loop extrusion sets long-range contact, while other locus-specific mechanisms, including compartmentalization, help maintain some contacts without cohesin.

CRUDO tests enhancer effects with and without cohesin

Predictions needed a direct test, so the authors built CRUDO: CRISPRi of regulatory elements upon degron operation. They combined pooled CRISPRi screens, which repress candidate elements, with auxin-inducible degradation of RAD21, the major cohesin subunit, in the same cell line. Each element within 1 Mb of a target gene was perturbed, and effects on gene expression were read out by FlowFISH in both the cohesin-present and cohesin-absent conditions.

The design yields a per-element measure the paper calls cohesin dependence. If repressing an enhancer cuts expression by, say, 15% with cohesin and by 7.7% without it, part of that enhancer's effect required cohesin. Formally, the metric compares the enhancer's effect in the two conditions, assuming enhancer activity combines multiplicatively with the promoter; significance was assessed across gRNAs targeting the same element with a paired t-test and Benjamini-Hochberg correction below 0.05.

A pilot at CCND1 showed the logic. The promoter and two proximal enhancers had similar effects in both conditions, with cohesin dependence of −4.9%, −13.3%, and 2.3%. A third, distal enhancer (E1) dropped from a 15.3% effect to 7.7%, giving 49.7% cohesin dependence, and it was also the only one of the three whose contact with the promoter changed.

Thirty enhancer-gene pairs, sixteen cohesin-dependent

The full screen covered 1,039 candidate enhancers around five cohesin-sensitive genes: CCND1, FAM3C, KITLG, SSFA2, and MYC. These genes were chosen for high expression, which aids detection, and for having ABC-predicted enhancers with a range of contact changes. The screen identified 30 significant enhancer-gene pairs, 16 of which had cohesin-dependent effects.

Two boundaries matter for reading that count. First, the 30 pairs come from five loci, not the genome, so the proportions describe this selected set of genes rather than all enhancers. Second, the effects being measured are often small: the discussion reports changes of 5–24% relative to baseline expression, detected with two readouts, FlowFISH and DC-TAP-seq. The paper attributes detection of such modest effects to CRUDO's sensitivity.

All sensitive enhancers were distal, but not all distal enhancers were sensitive

The headline pattern is an asymmetry. Every cohesin-dependent enhancer in the set was distal, more than 50 Kb from its target promoter. But the converse did not hold: many distal enhancers kept their regulatory effect after cohesin loss. Distance therefore sets a precondition for sensitivity without determining it.

The variable that tracked sensitivity better was the change in contact frequency itself. Across the 30 elements, cohesin dependence correlated with the percent change in enhancer-promoter 3D contact, and the paper's regression comparison supports contact change over distance alone: a distance-only model fit worse than a model including contact change. Among the 22 pairs whose contact fell by more than 50%, 15 were classified cohesin-dependent, two were not, and four were underpowered.

This is a correlation across elements, and the authors present it as such: contact change predicts effect change, with locus-specific factors such as compartmentalization shaping which distal pairs retain contact. The practical reading is that a distal position alone does not flag an enhancer as cohesin-sensitive; a measured contact drop does.

Why most genes barely notice cohesin loss

The genome-wide ABC comparison supplies the explanation for the original puzzle. Most predicted enhancer-gene interactions did not show substantial contact change upon cohesin depletion, largely because most predicted pairs sit at shorter distances where contacts are maintained. If an enhancer keeps contacting its promoter, repressing it should still reduce expression, so the gene stays cohesin-insensitive.

On this account, the small number of cohesin-sensitive genes is not evidence that 3D contact is irrelevant to enhancer function. It reflects that only a minority of enhancer-promoter pairs depend on cohesin for their contact. The authors connect this to earlier work at engineered loci and synthetic activators, which emphasized distance, and extend it by showing that endogenous distal enhancers vary widely in how much contact they lose.

Contact change, not distance, is the better predictor

For readers weighing whether a given enhancer will respond to cohesin perturbation, the paper's priority is explicit: look at the measured change in enhancer-promoter contact frequency, not at genomic distance alone. Distance above 50 Kb was necessary for cohesin dependence in this dataset but not sufficient, and the 22 pairs with more than 50% contact loss split into 15 dependent, two not dependent, and four underpowered cases rather than a clean yes.

The scope is five cohesin-sensitive genes in one HCT-116 cell line after 6 hours of auxin treatment, with effects measured by CRISPRi repression of candidate elements. Within that scope, the study's contribution is a quantitative link: contact loss predicts loss of enhancer effect, and the modest size of most effects explains why genome-wide expression changes stay limited.

Sources used in this answer