Reading Methylation on Each Fish’s Own Genome

The question

DNA methylation is a chemical “annotation” layered on top of the DNA sequence — small tags (methyl groups) attached to certain letters that can turn genes up or down without changing the letters themselves. It’s one of the ways the same genome can behave differently in different animals, so it’s a natural place to look for what distinguishes lean and siscowet lake trout.

An earlier step in this project (14-diff-meth) found ~300 stretches of the genome that looked differently methylated between the two ecotypes. But there was a catch, the same one that has shadowed this whole project: those differences were all measured against a single reference genome built mostly from lean-type fish. When you measure a siscowet fish against a lean-based yardstick, some of what looks like a “difference” can be an artifact of the yardstick, not real biology. Now that we have each ecotype’s own assembled genome, we can finally check.

What we did

The idea is simple to state: read each fish’s methylation on its own genome, not on a borrowed one.

Align to self. Each fish’s PacBio reads (which carry the methylation tags directly) were aligned to its own ecotype’s genome — lean fish to the lean genome, siscowet to the siscowet genome (23.8).
Call methylation. We measured, at every CpG site, what fraction of reads were methylated, using the same modkit tool the project used before.
Compare fairly. Here’s the hard part: the two genomes have different coordinate systems, so you can’t just line them up position-by-position. Instead we used genes as the common language — every gene has the same ID in both genomes — and asked, for each gene, “how methylated is it in lean fish versus siscowet fish?” (23.9). That sidesteps the coordinate problem entirely.
Re-test the old findings. Finally, we checked the ~300 reference-based differences against this fresh, native measurement: do they hold up?

What we found

Two honest headline numbers:

Of the 10,838 genes we could test, 63 showed a nominal methylation difference between the ecotypes (about 37 more-methylated in siscowet, 26 less). Interesting candidates, but…
None of them survived a strict statistical correction, and none of the ~180 genes tied to the earlier reference-based differences were confirmed on the native genomes.

In plain terms: the earlier methylation differences did not reproduce when we measured each fish on its own genome. That fits the pattern from the previous step (23.7) — signals that lean heavily on a single reference genome tend to weaken once that crutch is removed.

The important caveat

This is a place to be careful rather than dramatic. The native test was underpowered, for reasons that have nothing to do with the biology:

Only four fish per group. With samples that small, even a real difference struggles to clear the statistical bar once you correct for testing thousands of genes at once — the math simply can’t reach significance.
Uneven coverage. Reading methylation reliably needs deep coverage at each site. Some fish had plenty; others had relatively little. Only about a fifth of genes — and only 45 of the ~180 previously flagged genes — had enough coverage in enough fish to test at all.
A coarser lens. Aggregating to whole genes is more robust across two different genomes, but it can blur a sharp, localized methylation difference that a position-by-position test would catch.

So the fair conclusion is “the reference-based methylation differences are not corroborated on native genomes at the current sequencing depth,” not “they are definitely artifacts.” Settling it properly would take deeper sequencing, more individuals, or a position-level comparison.

Why this matters

This closes the loop on the project’s central caveat for the three cheapest, highest-value lines of evidence. Across structural variants, presence/absence, and now methylation, the story is consistent: some ecotype differences are robust to removing the single-reference bias (notably the calcium/depth-adaptation signal in 23.7), while others — including much of the methylation signal — are not, or can’t yet be confirmed. Knowing which is which is exactly what a two-genome design was built to tell us, and it keeps the project’s conclusions honest.

Where to explore

All tables are hosted on gannet:

Per-gene methylation, every gene × every fish — native_methylation_by_gene.tsv
All genes tested, with statistics — dmg_results.tsv
The 63 nominal candidates — dmg_significant.tsv
Did the earlier reference-based differences hold up? — reference_dmr_confirmation.tsv
Plain-text summary — native_dmg_report.txt

What’s next

More power. The clearest next move for methylation is depth and sample size — the biology here may be real but currently invisible to a four-versus-four test.
Phase 4 — new genes. So far every analysis only sees genes that already exist in the reference. A from-scratch gene annotation (BRAKER3) of each ecotype’s genome would reveal genuinely ecotype-specific genes that the reference can’t show — the last major piece of the two-genome comparison.

The large intermediate files (the eight read-to-genome alignments and full per-site methylation calls) are too big for GitHub; the committed outputs are the gene-level tables above.