Distance Window or TAD Boundary: Which Constraint Matters More?

Distance is the stronger statistical predictor; the boundary is the stronger mechanistic hypothesis — and both are weak enough that neither should decide a target-gene assignment on its own. In practice, distance decays regulatory influence smoothly and predictably, whereas boundaries are crossed often enough that a boundary-based exclusion is frequently wrong.

If you must choose one prior for ranking candidates, use distance plus activity. If you must choose one for a mechanistic story, use the boundary — then test it.

Contact frequency between an element and a promoter declines with genomic distance, and so does the frequency of validated enhancer–gene pairs. This produces a usable prior:

• • Most validated enhancer–gene pairs sit within a few hundred kilobases.
• • The probability of interaction falls roughly monotonically, with substantial scatter at every distance.
• • Long-range pairs are real but rare — which makes them interesting and also makes them statistically easy to over-claim.

The scatter is the important part. At any given distance, the distribution of true interaction probabilities spans orders of magnitude, because activity and architecture matter as much as spacing. Distance ranks; it does not classify.

Beware the degenerate version of this prior: the nearest gene is not the target gene. A substantial minority of validated pairs skip the nearest gene entirely, and a cluster of elements can regulate a gene several positions away.

Boundaries constrain contact statistically, so an enhancer and a promoter in different domains interact less often — but "less often" is not "never".

• • Cross-boundary regulation is well documented, particularly for strong enhancers and for genes near boundaries.
• • Boundary strength varies by cell type and by resolution of the contact map used.
• • A boundary annotated in one cell type may be weak or absent in another.

So the boundary offers a mechanistic story with a statistical tendency attached, not a hard partition. Papers that treat it as a partition produce two predictable errors: dismissing a real long-range pair because a boundary intervenes, and attributing an effect to boundary loss when the effect was mediated by something else in the deleted interval.

The fourth row is worth noting for screening: if the effect is small, neither prior helps, because the readout cannot resolve the pairing at all.

The Activity-by-Contact framework is the most widely used synthesis: it weights element activity, promoter activity, and contact frequency, and calibrates the combination against CRISPR perturbation data. Its practical value is in ranking, and its authors are explicit that predictions require experimental confirmation.

A defensible workflow:

1. Rank candidates with activity plus contact plus distance (ABC or equivalent).
2. Annotate whether a boundary intervenes, and record it as a caveat rather than a filter.
3. Test the top few candidates by perturbation of the element with the target gene measured, plus neighbouring genes as controls.
4. Report disagreements between prediction and perturbation, because those are the cases that refine the models.

Step 4 is where most of the field's useful information comes from, and it is usually omitted.

If you have budget for three perturbations and ten candidate genes: rank by activity-plus-contact, take the top two candidates and the single strongest candidate across any boundary, and measure all ten genes in each perturbation. The redundancy of readouts costs little and converts an assignment problem into a data problem.

If the boundary is the hypothesis you actually care about, the design shifts: perturb the boundary motif, measure contact, then measure expression. Our protocol for confirming a boundary is functional covers that route.

EDITGENE builds element-perturbation models with multiplexed expression readouts — element-level CRISPRi or deletion with neighbouring and distal genes measured in the same experiment — so assignments rest on data rather than on distance or domain membership.

Send us the locus and the candidate gene list. Contact EDITGENE to request a project consultation or quote.

Distance-decay relationships and boundary effects are population-level tendencies that vary by locus, cell type, and resolution; neither constitutes a rule for individual assignments.

Computational predictions, including activity-by-contact outputs, are prioritization tools and require experimental confirmation before being reported as regulatory relationships.

• • Fulco CP, et al. Activity-by-contact model of enhancer–promoter regulation from thousands of CRISPR perturbations. Nature Genetics. 2019;51:1664–1669.
• • Nasser J, et al. Genome-wide enhancer maps link risk variants to disease genes. Nature. 2021;593:238–243.
• • Rao SSP, et al. A 3D map of the human genome at kilobase resolution reveals principles of chromatin looping. Cell. 2014;159:1665–1680.
• • Gasperini M, et al. A genome-wide framework for mapping gene regulation via cellular genetic screens. Cell. 2019;176:377–390.