Four Signals in Twenty Tests

A candidate countermeasure can look decisive in a dish, remain encouraging in an animal model, weaken in a human analog, and never reach flight. Yet its public description may barely change. “Promising” survives each rung even when the evidence beneath it does not.

Gastronaut put that familiar translation problem through a quantitative test. Across 20 prespecified comparisons of reported benefit between evidence tiers, four were statistically significant, nine were null, and seven could not be estimated. All four significant results occurred in oxidative stress and nutrition. The distribution is not a footnote. It is the argument.

What matters is not whether ground models are useful. They are indispensable. The question is how much confidence a program should carry forward when the exposure, organism, endpoint, dose, timing, and intervention can all change as a study moves closer to humans or flight.

Two numbers initially made the answer look dramatic. In one oxidative-stress comparison, the odds ratio for reported benefit was 31.0. When Gastronaut broadened the selection frame to account for records missed in the first pass, the provisional sensitivity estimate was 12.40. In nutrition, a registered estimate of 8.14 produced a provisional sensitivity result of 3.34 under the same bias probe. All three finite significant odds ratios decreased by about 60 percent; the fourth significant comparison remained infinite before and after the sensitivity check. The displaced records were provisionally coded from accessible abstracts and still require full reading and final tier confirmation.

The candidate signal remained, but its magnitude was sensitive to the selection frame. That is precisely what a sensitivity analysis should be allowed to say. A program that reports only 31.0 creates one expectation. A program that also reports the provisional 12.40 result reveals how strongly the answer depends on what entered the evidence set. The sensitivity results are bias probes, not replacement estimates.

The odds ratios require another boundary. They compare the probability that studies reported benefit. They are not pooled physiological effect sizes. A positive cell-culture marker, a rodent functional outcome, and a human nutrition measure can all be coded as reported benefit without becoming biologically interchangeable. The result detects a difference in the literature's pattern. It does not measure how many units of protection a crew member would receive.

The nine null comparisons matter for the same reason. They prevent a selective story in which every research area loses its signal at a higher tier. A null result may reflect genuine consistency, limited statistical power, mixed interventions, or coarse coding. It cannot be presented as proof of equivalence, but it also cannot be erased because four comparisons were more exciting.

The seven unestimable comparisons impose a different discipline. They lacked the populated benefit and non-benefit cells needed for the planned odds ratio. Unestimable is not another word for null. It means the literature structure could not support that particular comparison. Calling those cases failures would be as misleading as calling them successes.

Nrf2 illustrates why tier labels alone cannot carry the conclusion. Shimizu and colleagues flew six wild-type and six Nrf2-knockout male mice to the International Space Station for 31 days, with six wild-type and six knockout mice in matched ground groups. Several inflammatory, immune, coagulation, and fibrinolytic changes were more pronounced in the absence of Nrf2 [1]. That is actual-spaceflight evidence that pathway status affected a mouse response to the integrated flight environment.

It is not a trial of sulforaphane, microgreens, or ORCA-grown food. A knockout comparison asks what happens when a pathway is absent. It does not establish that dietary activation will improve the same outcomes. Huyan and colleagues add a separate simulated-microgravity retinal-cell model involving oxidative stress, apoptosis, and Nrf2 signaling [2]. Gómez and colleagues describe the difficulty of designing antioxidant combinations for crewed missions, including altered metabolism and individual susceptibility [3]. Together these sources justify a research question, not a ready countermeasure.

The strongest skeptical reading is that Gastronaut compared heterogeneous studies and therefore found more about publication and model design than about biology. Grant that objection. One useful conclusion survives: a claim supported mainly by simulation should carry visible translation risk until the same candidate, measured exposure, primary assay, and functional endpoint persist into a more relevant tier.

That standard changes experiment design. A cell study and a crew study should not use different compounds and then be treated as one evidence chain. Crop composition and the amount consumed must be measured, because a nominal serving is not a biological dose. At least one primary assay should survive across tiers. Molecular markers should be paired with an organ-level or performance endpoint. Sampling should occur during exposure when recovery can alter the signal. Most importantly, the program should declare in advance what result advances, modifies, or stops the candidate.

Prespecification will not remove biological uncertainty. It will stop the program from changing the meaning of success after each tier produces a different kind of answer. That discipline is especially valuable when a short mission can support only a small sample and a narrow set of measurements.

Gastronaut has published its Tier-Transfer Methods and 20-comparison result table so NASA investigators can inspect the tier definitions, 2 by 2 counts, Fisher tests, sensitivity estimates, and coding limits. The useful next exchange is not acceptance of the conclusion. It is reproduction, followed by the choice of one nutrition or oxidative-stress candidate for a matched-endpoint protocol across a ground analog and a later flight study.

ORCA remains a ground-stage cultivation system at approximately TRL 3 to 4 and has not flown. Its internal growth record can help set cultivation parameters and plan assays, but it cannot establish human efficacy or flight performance. Reconciliation of specific edited plant lines with individual ground cycles remains incomplete, so no edited-line performance claim belongs in this evidence chain.

The four significant comparisons identify where translation deserves attention. The nine nulls stop the result from becoming a universal law. The seven unestimable cases show where the record cannot answer. A credible countermeasure program should carry all twenty outcomes up the ladder, because the evidence that resists the headline is often the evidence that improves the experiment.

Research foundation and evidence boundaries

Gastronaut owns the tier coding, prespecified comparisons, sensitivity analysis, and cross-study interpretation. The analysis is exploratory and has not been independently peer reviewed. The odds ratios concern reported benefit rather than pooled physiological effect, heterogeneous interventions and endpoints limit causal inference, and significant attenuation appeared only in the oxidative-stress and nutrition comparisons described. The cited studies retain ownership of their underlying findings. Any matched-endpoint ORCA study remains proposed, and ORCA has not flown.

References

1. Shimizu et al., “Nrf2 alleviates spaceflight-induced immunosuppression and thrombotic microangiopathy in mice,” Communications Biology (2023). DOI: 10.1038/s42003-023-05251-w
2. Huyan et al., “Simulated microgravity promotes oxidative stress-induced apoptosis in ARPE-19 cells associated with Nrf2 signaling pathway,” Acta Astronautica (2022). DOI: 10.1016/j.actaastro.2022.05.012
3. Gómez et al., “Key points for the development of antioxidant cocktails to prevent cellular stress and damage caused by ROS during manned space missions,” npj Microgravity (2021). DOI: 10.1038/s41526-021-00162-8