Linking ATAC-seq Changes to Gene Expression Causality

Three things turn a correlation into a causal claim:

1. Perturb the element (not the gene), and measure both accessibility and expression.
2. Show the accessibility change precedes or is independent of the expression change — for example, by perturbing chromatin directly with an epigenome editor.
3. Show a dose relationship, or reverse the perturbation and see the phenotype reverse.

Correlation across conditions, samples, or cells is the weakest form of evidence and should be labelled as such.

Accessibility and expression correlate because both are downstream of the same regulators. That shared upstream cause produces strong correlations even when accessibility has no causal role in the gene's transcription.

Two concrete failure modes:

• • A locus opens because it is transcribed. Accessibility at gene bodies and promoters can be a consequence of transcription, not a cause.
• • A third factor changes both. A differentiation state alters both chromatin and expression genome-wide, generating correlations that are about the state rather than about the element.

Neither failure is detectable from correlation alone, which is why the causal step has to be a perturbation.

The second row is the one that specifically addresses the accessibility–expression link: if depositing acetylation at an element changes expression, chromatin state is doing causal work there. If it does not, the accessibility change may be a marker rather than a mediator.

If accessibility mediates the effect of a perturbation on expression, then three conditions should hold:

• • The perturbation changes accessibility at the element.
• • The perturbation changes expression of the target gene.
• • The accessibility change and the expression change track each other quantitatively across conditions or dose levels.

Mediation analysis can be formalized statistically, but in practice reviewers respond better to a dose series: titrate the perturbation (multiple guides of differing efficiency, or inducible systems at different induction levels), measure both readouts, and show the relationship. A monotonic relationship across a dose series is far more convincing than a single knockout.

• • Non-targeting perturbation with the same machinery.
• • Catalytically dead effector for epigenome edits.
• • A neighbouring element that does not regulate the gene, to show site specificity.
• • Second gene measurement, ideally one in the same domain but not expected to respond.
• • Matched clones from the same editing campaign.
• • Reversion where feasible.

The third control is the one that distinguishes "this element matters" from "chromatin in this region matters".

The second row is the strongest available statement about chromatin causality, and it is achievable with existing editors — which is why epigenome editing has become the standard tool for this specific question.

EDITGENE builds element-targeted perturbation models: CRISPRi/a lines for necessity and sufficiency, epigenome-editing lines for chromatin-level causality, isogenic deletion panels, and reporter knock-ins for quantitative readouts.

Tell us the element and the gene. Contact EDITGENE to request a project consultation or quote.

Accessibility–expression correlations have multiple explanations, including shared upstream regulators and transcriptional feedback; perturbation is required for causal claims.

Effect sizes and the feasibility of each design vary by locus and cell type, and conclusions should name the perturbation, its validation, and the readout used.

• • Fulco CP, et al. Systematic mapping of functional enhancer–promoter connections with CRISPR interference. Science. 2016;354:769–773.
• • Hilton IB, et al. Epigenome editing by a CRISPR-Cas9-based acetyltransferase activates genes from promoters and enhancers. Nature Biotechnology. 2015;33:510–517.
• • Klemm SL, et al. Chromatin accessibility and the regulatory epigenome. Nature Reviews Genetics. 2019;20:207–220.