A plant ontology on a fly, in a 16.5 tesla magnet

Transcription profiling of Drosophila exposed to a levitation magnet for different lengths of time — NASA OSDR OSD-27, doi:10.1186/1471-2164-13-52. This is the page where the cross-species machinery does real work, and where it is easiest to see what that machinery costs.

• 63/125atlas loci reach the fly
• 5contrasts isolate the field
• 15nodes received data
• 0move consistently

The original authors report an effect. This page does not contradict them.

Herranz et al.'s own abstract states that “a strong magnetic field, of 16.5 Tesla, had a significant effect on the expression of these genes, independent of the effects associated with magnetically-induced levitation and hypergravity”. That is a finding about the fly transcriptome. This page reports a null. Both can be true, and the difference is worth being precise about:

• A different gene set. They analysed the transcriptome. This page looks only at the handful of fly genes that QBO's Arabidopsis loci reach through orthology — a small, non-random slice chosen for quantum chemistry, not for responsiveness.
• A different question. They asked whether the field had a significant effect. This page asks whether individual QBO nodes move the same way in all five matched contrasts. A gene can contribute to a real overall effect while its direction varies between sexes and durations.
• A different multiple-testing position. Significance here is the adjusted p-value GeneLab computed across the whole array, read per gene. Nothing on this page re-tests anything.

So the honest summary is narrow: the quantum-chemistry nodes this atlas cares about are not where that reported effect shows up consistently. Nothing here argues the effect is absent.

What this study is, and what it is not

It is a strong-field study. Roughly 16.5 T, against the ~50 µT geomagnetic field the review is about. Those are opposite ends of the same axis, about six orders of magnitude apart. QBO's field_regimes vocabulary spans both, so the data is in scope — but nothing here should be read as evidence about near-null fields, and a node that does not respond at 16.5 T has not thereby been shown insensitive at 30 nT.

Diamagnetic levitation confounds field with gravity. Inside the bore, position determines effective gravity, so 0g*, 1g* and 2g* all occur at high field. Most of the study's 420 contrasts change field and gravity together. Only these 5 compare the magnet against Earth at matched effective gravity, duration, temperature and sex:

Every other comparison in the file was excluded. Five contrasts are five replications of one question, not five results — so the evidence on this page is whether a node moves the same way across them, not whether it moves at all in one.

The bridge, and where it lands

The ontology is anchored on Arabidopsis. Reaching the fly means asking, for each locus, what the corresponding gene is — here through Ensembl Compara pan_homology, the only division that crosses kingdoms. 63 of 125 atlas loci map (50%), 12 of them one-to-one.

Only one backbone could run. The committed OrthoDB matrix is human-anchored and has no Drosophila column, so this projection has no second method to corroborate it. That is stated rather than left to look like agreement: ensembl_pan_homology.

Where it lands is worth seeing on its own. Arabidopsis CRY1 (AT4G08920) projects onto Drosophila FBgn0025680 — cry, the canonical animal magnetoreceptor — in a dataset where the flies were sitting in a superconducting magnet. A plant-anchored quantum ontology reaches the one animal gene the radical-pair literature is built on.

What the projection destroys

An ortholog call is not an identity, and a finished figure gives no sign of how much of the map's own structure survived the crossing. Two collapses happen here, both real and neither an error in the orthology — they are true statements about how plant and fly gene families correspond. They are errors only if left unsaid.

Distinct nodes landing on the same fly gene. When this happens the nodes carry identical numbers, and a reader sees agreement between independent measurements that are one measurement drawn several times.

The cryptochrome map is the severe case: CRY1, CRY2 and the Trp-triad node all project onto the single fly gene cry. On QBM-03 the three are distinct entities with different evidence; in the fly they are one number shown three times. The complex I pair is benign by comparison — the FMN site is part of complex I, so sharing subunits is what should happen.

One node averaging very many fly genes. A node standing for one gene and a node standing for forty are not measuring the same kind of thing, and nothing on the map distinguishes them.

• ASC_GSH_CYCLE on QBM-07 averages 39 fly genes, reached from AT3G24170 AT1G19570 AT3G52880 — largely the glutathione S-transferase family. Its value should be read as 'something in a large family', not as a measurement of the ascorbate–glutathione cycle.

The maps

One panel per map, showing the first contrast, on a colour scale fixed across all 5 so the panels are comparable rather than each self-normalised. The full per-contrast values are in the table below — the panel is an illustration, the table is the evidence.

Every node across every contrast

The actual evidence. ▲ is up in the magnet, ▼ is down, and * marks significance at adjusted p ≤ 0.05. A node with five arrows the same way would be a finding; none has.

QBM-01 — 7 of 10 identifier-bearing nodes (70%)

Scroll the table sideways to see every contrast.

QBM-03 — 3 of 4 identifier-bearing nodes (75%)

Scroll the table sideways to see every contrast.

QBM-07 — 5 of 8 identifier-bearing nodes (62%)

Scroll the table sideways to see every contrast.

How to read a null like this

• It does not show the field had no effect. It shows that these 15 nodes, measured on a 2012 microarray through a cross-kingdom ortholog projection, did not move consistently. Each of those three steps loses signal.
• It does not transfer to near-null fields. 16.5 T and 30 nT are different regimes, and the mechanisms proposed for each are different.
• The confound is handled, not removed. The 5 contrasts match effective gravity, but every one of them is still inside a 16.5 T bore, where the flies were also being levitated.
• No power analysis is offered here. The atlas's pre-registered test reports its own minimum detectable effect; this page does not, so "no consistent change" should not be read as "no change larger than x".

Reproducing it

PYTHONPATH=src python3 scripts/build_osd27_showcase.py
PYTHONPATH=src python3 scripts/build_osd27_page.py

The differential-expression table at OSDR is ~623 MB across ~1,690 columns. qbio.osd27 streams it a row at a time and keeps only the 5 contrasts' columns, so the cached slice is a few hundred KB. Coverage, per-contrast values and the collapse analysis are written to results/osd27/record.json, and this page is generated from that record.