Written and maintained by CASRAI Editorial Board
Last updated
ATAC-seq — assay for transposase-accessible chromatin with sequencing — asks a narrow question well: which parts of the genome are physically open right now. It answers it with a single enzymatic step, needs far less input than the methods it replaced, and produces a library in an afternoon. It also fails in characteristic ways that are invisible until the sequencing comes back, which is why the quality-control section below matters more than the protocol itself.
What the assay actually measures
A hyperactive Tn5 transposase, pre-loaded with sequencing adapters, inserts those adapters into DNA it can physically reach. Nucleosome-bound and tightly packed chromatin is protected; open regulatory DNA is not. Because Tn5 fragments the DNA and attaches adapters in the same reaction, the step is called tagmentation — tagging plus fragmentation.
The readout is therefore a map of accessibility, from which you can infer regulatory regions, transcription-factor binding sites and nucleosome positioning. It is not a map of transcription, and it does not tell you which factor is bound — only that something is holding a stretch of DNA open.
Compared with the earlier DNase-seq and MNase-seq approaches, ATAC-seq is more convenient to run and recovers DNA more efficiently, which is why it became the default genome-wide accessibility assay.
The mitochondrial problem
This is the single most important practical fact about ATAC-seq, and the one that surprises people running it for the first time.
Mitochondrial DNA is not chromatinised. It is naked, abundant, and utterly accessible to Tn5 — so the transposase attacks it enthusiastically. Depending on cell type, roughly 20% to 80% of sequencing reads can map to the mitochondrial genome. Those reads are pure cost: you pay for them, and then discard them in analysis.
A library that looks like it sequenced deeply can therefore be shallow where it counts. Always judge depth by non-mitochondrial reads, never by raw read count.
Approaches to the problem differ in how well they work:
- Omni-ATAC — an improved protocol that reduced mitochondrial reads and raised signal-to-background across cell lines and tissues, and crucially made frozen tissue workable. This is the usual modern starting point.
- CRISPR/Cas9 depletion of mitochondrial fragments after the fact. In a direct comparison, the original protocol followed by CRISPR treatment gave the highest peak count and the best data quality.
- Detergent-based approaches reduce mitochondrial reads too, but removing detergent increased background and yielded fewer peaks — a reminder that mitochondrial percentage alone is a bad optimisation target.
That last point is worth stating plainly: driving the mitochondrial fraction down is only an improvement if peak count and signal-to-background hold up. Optimising one number in isolation can produce a cleaner-looking library with less real signal in it.
The quality metrics that decide whether a library is usable
Three metrics carry most of the diagnostic weight, and established pipelines — ENCODE, PEPATAC and nf-core among them — all report them.
- Library size / complexity — how many distinct fragments you actually have, as opposed to PCR duplicates of a few. Over-amplification is the usual culprit when complexity is low.
- Percentage of reads mapping to mitochondrial DNA — see above. Interpret it alongside peak count, not on its own.
- TSS enrichment — the degree to which signal piles up at transcription start sites. Promoters are constitutively open in essentially every cell type, so they act as a built-in positive control. Flat TSS enrichment means the assay did not work, whatever else the numbers say.
The fragment-length distribution is the fourth thing to look at and the most immediately readable. A working library shows a periodic pattern: a large population of short, sub-nucleosomal fragments from open regions, then progressively smaller populations at mono-, di- and tri-nucleosome spacings, because Tn5 cuts on either side of intact nucleosomes. Losing that periodicity — a smooth, featureless distribution — usually means over-tagmentation or degraded input rather than a sequencing problem.
Where it goes wrong upstream
- Too much or too little Tn5 relative to input. Over-tagmentation shreds everything and destroys the nucleosomal periodicity; under-tagmentation gives long fragments and few usable reads. The enzyme-to-nuclei ratio is the parameter to control, which means cell counting has to be accurate.
- Poor nuclei preparation. Incomplete lysis leaves intact cells that never see the transposase; over-harsh lysis releases mitochondria and raises the very contamination you are trying to avoid.
- Dead and dying cells. Compromised membranes expose DNA that was never open in a living cell, adding background that looks like signal.
- Over-amplification during library PCR. Extra cycles convert a low-complexity library into a high-duplicate one without adding information.
Choosing between accessibility assays
ATAC-seq is the right choice for genome-wide accessibility with limited material, and for frozen tissue if you use a protocol validated for it. It is the wrong choice when you need to know which protein occupies a site — that is a ChIP-based question — or when you want expression rather than regulation, which is RNA-seq territory. Accessibility and expression are correlated but not interchangeable: a promoter can be open and transcriptionally quiet.
Frequently asked questions
What does Tn5 actually do?
It is a hyperactive transposase carrying sequencing adapters. It inserts those adapters into DNA that is physically accessible, fragmenting and tagging in one step — tagmentation. Regions wrapped in nucleosomes or otherwise compacted are protected.
Why are so many of my reads mitochondrial?
Because mitochondrial DNA has no chromatin to protect it. Twenty to eighty per cent is the reported range across cell types. It is expected, not a mistake — but it means you must size your sequencing depth on non-mitochondrial reads.
What is Omni-ATAC and should I use it?
An improved protocol that lowers mitochondrial background, improves signal-to-background, and enables work on frozen tissue. For most new projects it is the sensible default, particularly if your input is anything other than fresh cultured cells.
What does the fragment-size ladder tell me?
That the assay worked. Short sub-nucleosomal fragments plus periodic peaks at nucleosome spacings indicate Tn5 cut around intact nucleosomes. A featureless smear points to over-tagmentation or degraded material.
How much TSS enrichment is enough?
Thresholds are pipeline- and genome-specific, so take the number from the pipeline you are running rather than from a remembered rule of thumb. What is universal is the direction: promoters should be clearly enriched, and a library without that enrichment is not rescuable by deeper sequencing.
Can ATAC-seq tell me which transcription factor is bound?
Not directly. It shows a protected footprint within an accessible region, which supports inference about binding, but identifying the protein requires an orthogonal assay.
References
- ATAC-seq: a method for assaying chromatin accessibility genome-wide — PMC4374986
- An improved ATAC-seq protocol reduces background and enables interrogation of frozen tissues (Omni-ATAC) — Nature Methods
- Chromatin accessibility profiling by ATAC-seq — PMC9189070
- Reducing mitochondrial reads in ATAC-seq using CRISPR/Cas9 — PMC5446398
