How modifications shape transcript fate
A modification is usually described by its name. What matters experimentally is where in the transcript's life it acts — before the transcript leaves the nucleus, or after it has to be translated and kept alive.
01What this article answers
The same modification can be reported at four different levels of a transcript’s life: how it is spliced, whether it is exported, how efficiently it is translated, and how long it survives. Which level a study can speak to depends on the readout it chose, not on which mark it picked.
That is why two papers on the same modification can reach different conclusions without contradicting each other: one describes a change in modification level, the other a change in transcript behavior. The useful comparison is between them, and it needs the two measurements to come from the same samples.
02The four levels, and what each one needs
A modification result is a property of a position. The functional claim is a property of a transcript. Aligning the two is the actual work of a study, and the four levels differ in how far apart those two things sit.
| Level | What has to be measured on the modification side | What the study adds |
|---|---|---|
| Splicing | Site-level modification calls annotated to exons, introns and splice-adjacent regions | Isoform-level or junction-level quantification from the same samples |
| Nuclear export | Site-level calls with transcript identity, so nuclear and cytoplasmic pools can be compared | Fractionation of the two pools before or alongside the modification assay |
| Translation | Site-level calls in coding sequences and untranslated regions | A translation readout such as ribosome profiling, run on matched material |
| Stability | Site-level calls plus a level estimate per position | A decay or time-course design that separates synthesis from turnover |
03How to set the comparison up
- Start from the level you want to claim. If the conclusion is about translation, the modification readout alone will not support it — the translation measurement has to be planned in from the beginning.
- Keep the two readouts on the same material. A modification difference and a functional difference measured on different samples are two results, not one.
- Prefer site-level modification data. A region-level profile can point at a transcript; only a resolved position can be tied to a specific feature of it.
- Check abundance underneath the modification. A change in how much transcript exists will move the modification signal as well, so the two have to be read together.
04Where this breaks down
- A resolved position does not by itself explain a phenotype. The functional link is a study design, not something the modification assay can demonstrate.
- Correlation across conditions is not causality: perturbing the modifying enzyme is what separates the two, and it brings its own confounders.
- For marks whose detection is enrichment-based, the readout is a region, so the level of the claim has to be stated at that level.
06FAQ
Can one experiment cover all four levels?
No. Each level needs its own readout, and the modification assay is only one of them, so the translation, export or decay side has to be planned in from the beginning. What one experiment can do is give the modification data at a resolution the other readouts can be compared against.
Why insist on site-level modification data?
Because the four levels are all position-dependent: a splice junction, a coding sequence, a UTR. A region-level profile can point at a transcript, but it cannot be tied to any of those features specifically, which is why the site-level readout is what the comparison starts from.
What if the mark has no single-base method?
Then the study works at profile level, and the functional claim has to be stated at that level too, because a region-level readout is what the modification side actually measured. The modification types page lists which marks have a single-base route and which do not.
Does the direction of the effect matter for the design?
It matters for interpretation rather than for measurement. A stabilising and a destabilising effect need the same comparison, and they differ only in what the expected sign of the change is, so the direction is settled when the result is read rather than when the experiment is designed.