Does radiation move the same chemistry a magnetic field does?

Multi-omics profiling reveals ethylene signalling as a key pathway underlying both genetic and epigenetic responses to low-dose ionizing radiation in Arabidopsis — NASA OSDR OSD-782 / GLDS-679. Two doses, four timepoints, Arabidopsis, GeneLab-processed differential expression.

Ionising radiation is not a magnetic-field regime and this page does not treat it as one. It is here because radiolysis makes radicals and ROS directly, and the magnetic-field literature converges on the same redox machinery. So the question is overlap: which nodes does radiation move, and are they the ones a magnetic field moves? A shared locus is a lead, not a mechanism.

• 123/125atlas loci measured
• 2×4doses × timepoints
• 4/11loci respond to both
• 2agree in direction

Why this page works at locus level, not node level

The obvious way to ask the question is per node: does the AOX node respond to both? Done that way the answer looked clean — three of six shared nodes, two agreeing in direction. It does not survive looking at the underlying genes.

At node level AOX appears to respond to both radiation and near-null field in the same direction, which reads as one clean shared response. At locus level it is two stories. AOX1D (AT1G32350) rises under both and genuinely agrees. AOX2 (AT5G64210) also responds to both, but FALLS under radiation while rising under near-null field. The two family members move in opposite directions under radiation, so the node's single trace is the aggregator choosing a different gene at each timepoint - it swung from -2.43 at 24 h to +3.96 at 72 h for that reason, not because anything reversed. A gene family is not a unit of response, and the node-level agreement here is partly real (AOX1D) and partly an artefact (AOX2 cancelling into it).

So the maps below draw multi-locus nodes as a per-locus heatmap rather than a single trace. One row per locus, one column per timepoint, same diverging colour scale as everything else — a family that splits shows as a warm row above a cool one. On these maps 17 node(s) are drawn that way: ACONITASE, ALT_NADH_DH, AOX, ASCORBATE_PEROXIDASE, ASC_GSH_CYCLE, CATALASE, CIA_SYSTEM, COMPLEX_I, COMPLEX_II, CYTOCHROME_C, FES_CLIENTS, HSCB_GRXS15, NFU_CARRIERS, PLASTID_ROS, RIESKE_FES, SUF_SYSTEM, SUPEROXIDE_DISMUTASES. A node standing for a single locus keeps its sparkline, which reads better when there genuinely is one trace.

The AOX heatmap is the case above made visual: the 1g32350 row is warm across all four timepoints while 5g64210 turns blue at 24 and 72 hours. No aggregate could have shown that, and it took a separate investigation to find before the heatmap existed.

So the atlas gained a check it did not have: qbio.project.project_series now records, per timepoint, whether a node's loci disagreed in sign, and the caption says so. Under radiation almost every multi-locus node diverges — a strong perturbation splits gene families, and a node aggregating them is summarising genes doing different things. That qualifies every multi-locus value on every page in this atlas, not just this one.

Radiation against near-null field, locus by locus

The near-null comparison is Parmagnani 2022, the only near-null time course in Arabidopsis available. It reaches 11 of the atlas's loci, because its supplementary table was filtered by its authors to oxidative-reaction enzymes. A locus counts as responding at |log₂FC| ≥ 0.5, applied identically to both sides — the point of a fixed threshold is that it cannot be tuned to make the overlap more interesting.

Scroll sideways for every column. Values are the most extreme timepoint in each experiment.

And against a strong field, where a test is possible

The OSD-8 16.5 T contrast shares 123 loci with this study — enough to test. 3 respond to both, against 2.112 expected by chance (p = 0.352, 10,000 permutations), and 0 agree in direction.

That is no overlap at all, and it is consistent with the strong-field study itself: 16.5 T moved none of the six shared nodes past threshold. The near-null regime — the one the review is actually about — is where the leads are.

Are the atlas's loci especially radiation-responsive? No.

The same permutation test the OSD-8 page runs, from the same function in qbio.compare with the same seed and permutation count — otherwise the two numbers could not be placed beside each other. At every dose and timepoint the atlas's loci move less than the background, p between 0.87 and 1.00.

But they are not inert, and that matters. They move up to 0.299 mean |log₂FC| under radiation against 0.166 under the 16.5 T field — roughly 1.8× more. The background moves more still (0.423), which is why they stay under-represented. So the magnetic null on the OSD-8 page cannot be explained by these genes being unable to move. They move perfectly well when something perturbs them; they simply are not preferentially field-responsive.

The study, and the contrast that needed its own rule

Three declared factors: Ionizing Radiation, Absorbed Radiation Dose, Time of Sample Collection After Treatment. cesium-137 gamma radiation at 0.1 Gy and 1 Gy against non-irradiated controls, sampled at 1 hour, 3 hour, 24 hour, 72 hour.

The dose comparisons differ in two declared factors, not one. Radiation source and dose are coupled by construction — there is no such thing as cesium-137 at zero dose — so the single-factor rule the OSD-27 page uses to isolate a magnetic field rejects every one of them. qbio.osd782.isolates_dose treats source and dose as one coupled axis and requires time to match. 8 contrasts qualify, and the build asserts that none of them is single-factor, so the reason the rule exists stays encoded.

Contrasts are kept irradiated-first, so a positive log₂ fold change means "higher after irradiation".

The maps

Each node is tinted by its most extreme timepoint and carries its trajectory. Read them with the divergence warning above in mind: where a node's loci disagree in sign, its trace is the aggregator choosing between them, and the caption says which nodes those are.

What this does not show

• Not evidence about magnetic fields. Radiation is a different perturbation axis. QBO's field_regimes vocabulary is not used for it and was not extended; the record carries its own perturbation block.
• Overlap is not shared mechanism. Two perturbations converging on one locus says they converge, not why.
• Different labs, tissues and platforms. Callus versus seedling, microarray ratio versus RNA-seq differential expression. Direction is the robust comparison; magnitude is not, and no magnitude comparison is made across experiments.
• Eleven loci is not a test. The near-null arm is reported as a table and explicitly not tested.

Reproducing it

PYTHONPATH=src python3 scripts/build_osd782_showcase.py
PYTHONPATH=src python3 scripts/build_overlap.py
PYTHONPATH=src python3 scripts/build_osd782_page.py

The differential-expression table is ~255 MB across 132 contrasts; qbio.osd27.slice_contrasts streams it to a few hundred KB. Every number here is written to results/osd782/record.json and results/overlap/record.json, and this page is generated from them.