Changelog

A living log of corrections and findings

Syntropa is an ongoing research project: as the work advances, results are refined and scope is sharpened. This log records the dated changes that matter (material corrections and substantive findings), so any earlier piece (the preprint, the one-pager, an outreach note) can be read alongside them. It deliberately does not list routine refinements, wording, reference fixes, or work still in progress; each entry records the settled result, not every step. Tags: [Correction] (a number or claim a reader may have relied on has changed), [Finding] (a substantive new result), and [Scope] (a clarification of what a claim does and does not cover).

2026-09-21

[Correction] The corpus is not dominated by hydrogenotrophic methanogens. It holds four gut species of them. The atlas scope section described the corpus as dominated by gut and host-associated bacteria and hydrogenotrophic methanogens, and described acetogens and many dissimilatory sulfate reducers as underrepresented or absent. A taxonomic census of the discovery-live set now puts numbers on both, and both were wrong, in opposite directions.

Twelve models of 15,664 carry a methane exchange at all. Eight of those are named to species, and those eight are four species, each a gut or rumen organism duplicated across AGORA and AGORA2: Methanobrevibacter ruminantium, M. smithii, Methanosphaera stadtmanae and Methanomassiliicoccus luminyensis. The remaining four are not named to species: one is a coarse lumped aerobic methanotroph, one is an APOLLO pan-model whose identifier names it a Methanosphaera, and two are BiGG models this corpus does not identify by name. Three of those four do carry a domain, so for them the gap is a missing species label rather than an absence of taxonomy; only the lumped methanotroph carries no taxonomy at all. A domain does not name a species, so knowing it does not narrow the range below. So the honest species count is at least four and at most six, and the corpus cannot currently say which, which is itself part of the finding. This entry originally continued that the environmental hydrogenotrophic methanogens were absent from the corpus; see the amendment below, which corrects that. The correction matters more than a wording fix, because Syntropa's flagship result is a hydrogen syntrophy: a reader choosing the tool for that work would have read the earlier sentence as a deep partner pool, and it is four gut species.

In the other direction, two guilds were sold short. Acetogens reach 23 species across Acetobacterium, Sporomusa and Clostridium. Dissimilatory sulfate reducers reach 18 species of Desulfovibrio alone, and 24 models across the sulfate and sulfur reducing genera. Carboxydotrophs and reverse-beta-oxidation chain-elongators remain genuinely thin, and anammox is absent entirely. Overall, between 485 and 1,080 of 15,664 models fall in a chemolithotroph, autotroph or phototroph guild by taxonomy, so between 3.1 and 6.9 percent. The single figure of about 3 percent published here until 2026-09-25 was the lower bound only. It came from a curated list of genera that was documented as an upper bound and measured to be acting as a lower one, missing 132 of the 152 cyanobacteria genera actually in the corpus. The upper figure adds whole phyla whose members are predominantly of that guild, which is generous by construction. Both are taxonomic bounds rather than verified counts: neither genus nor phylum membership is the same as a retained pathway.

Amended 2026-09-22, after running that functional check. This entry originally said Methanobacterium, Methanothermobacter, Methanococcus, Methanospirillum and Methanoculleus were absent from the corpus. That was wrong in letter and too kind in substance. CarveMe reconstructions of Methanococcus maripaludis, Methanothermobacter marburgensis, Methanosarcina baltica, Methanoculleus chikugoensis and Methanothrix thermoacetophila are all present. They carried no taxonomy at the time, which is why a taxonomy-based query could not see them. That has since been fixed: on 2026-09-23 taxonomy was backfilled for 4,325 models from their genome accessions, taking coverage from 69 to 95 percent, and every one of those genera is now labelled and visible. The count of archaea in the corpus did not grow from 14 to 198 because anything was added; it grew because the corpus could finally see what it already held.

What is missing is the function. Each of those five carries no metabolite with the formula CH4 and no metabolite named for any of the four methanogenesis cofactors, at 42 to 46 percent gap-filled and roughly 2,000 reactions apiece. The two curated gut methanogens carry methane, all four cofactors, and a methyl-CoM reductase with a real gene rule, in 690 to 753 reactions. The auto-reconstructions are more than twice the size and carry none of the machinery that defines the organism. So the gap is not acquisition alone: five of the genomes are already here, and a heterotroph-biased universal reaction database rebuilt each as a generic heterotroph, so more genomes through the same pipeline would add more shells. Separately, the count of twelve methane-carrying models over-counts, because E. coli iML1515 carries a methane exchange and transport without a methyl-CoM reductase.

The cause is source composition rather than a choice. AGORA and AGORA2 are human gut model collections, so the only archaea they contribute are gut archaea, and no environmental or anaerobic-digester archaeal collection is represented. Closing that gap is an acquisition question and it is open. What this correction does not touch is the flagship positive control itself, which was run on a named pair rather than found by a corpus search, so it never depended on the depth of the partner pool. What it does retire is any reading in which a corpus-scale search could have discovered that partnership. The flagship's own methanogen is not merely rare in the searchable set, it is excluded from it, so the search could not have returned that organism at all. The correction was prompted by an external reviewer's question about taxonomic and trophic bias, which turned out to be right.

2026-09-08

[Correction] The cross-genus electron-disposal survey is a necessary-condition screen, not a test that distinguishes obligate syntrophs. Syntropa reported that across 35 syntroph genome-scale models spanning 14 genera, every mappable hydrogen/formate electron-disposal reaction is thermodynamically blocked alone, 39 of 39, with the feasible-alone side empty. Those numbers are unchanged and still stand. What was missing was a negative arm: every organism in the survey had been selected as a syntroph, so the result measured sensitivity and said nothing about specificity. Syntropa has now run the missing control. Four organisms that grow perfectly well in pure culture, Thermotoga maritima, Zymomonas mobilis, Ruminococcus albus and Clostridium kluyveri, return the same verdict: 4 of 4 scoreable, 100 percent, identical to the syntroph arm. The screen therefore separates nothing. It detects a mappable disposal route, which is a necessary condition for interspecies hydrogen transfer and not evidence of obligate dependence. The claim is corrected to that narrower reading, which is what the preprint's own Figure 2 caption already stated.

The instrument was corrected alongside the claim, so the error cannot recur. The survey tool now ships a negative arm, reports sensitivity, specificity and abstention beside each other, and derives its closing interpretation from the two arms instead of asserting one. A positive-only run now states plainly that it carries no specificity information and declines to draw a conclusion; previously it printed a fixed "robust cross-genus pattern" summary whatever the data showed. The thermodynamic machinery itself, the shared-hydrogen coupling that produces the obligate result on canonical reaction chemistry, is not affected by this correction, and neither is the separately reported finding that a whole-network in-situ demonstration on gap-filled reconstructions remains blocked by the defined-medium catch-22.

2026-08-31

[Finding] The thermodynamic obligate test now runs on the flagship glucose-to-methane relay. The second-law (thermodynamic) test now runs on the flagship glucose-to-methane relay itself, not only the canonical secondary-oxidation classes, and confirms the obligate middle member. With a medium leak closed (below), the community-FBA result is correspondingly literal: the lone finisher produces zero of the target without a partner. Scope: "obligate" here is a property of the interspecies-hydrogen-transfer chemistry on canonical reactions, not of the discovered fermenter's own glucose-fermentation step.

Amended 2026-09-23: that step has now been tested, and the scope above holds. This entry originally added that the fermenter's glucose-fermentation step "was not itself thermodynamically tested". It has been. A glucose-fermentation class was added to the class-level gate and run through the same solo-FBA, solo-thermodynamic and thermodynamic-plus-finisher comparison as the other classes. The fermenter grows alone at full rate even with the thermodynamic constraint applied, and the methanogenic partner changes nothing, against a validated obligate class whose solo growth falls to zero. The reaction reads -140.1 plus or minus 12.1 kJ per mole, comfortably exergonic.

So the scope statement above is confirmed rather than overturned: "obligate" is a property of the interspecies hydrogen transfer, and the fermenter's own half runs perfectly well unaided. What changes is the standing of that sentence. It was an assertion readers were asked to take on trust, and it is now a measurement. The flagship result is untouched, since it was a named-pair positive control and the partnership still produces methane the finisher cannot make alone.

[Correction] Positive-control enrichment: 22 of 30 becomes 15 of 30. A medium-currency fix closed a small leak that let a lone finisher import the hand-off intermediate (hydrogen and CO2) from the auxotrophy-support medium rather than only from a partner. On the corrected engine the out-of-sample candidate enrichment is 15 of 30 (was 22 of 30); the true-negative decoys are unchanged at 0 of 60. Syntropa reports these as counts, not a significance test: the positive and decoy classes are separated on the very hydrogen/methanogenesis signature the score keys on, so a p-value would quantify a separation that is true by construction rather than an empirical result. The informative number is that only half of the structurally-eligible front-ends actually reach emergent, feedstock-dependent methane. The fix removed leak-inflated borderline hits. The separation, and the honest framing (a positive control that recovers a known partner class, not a novel discovery), are unchanged.

2026-07-18

[Finding] Syntropa found and reported an upstream reproducibility-tool bug. While stress-testing the corpus's reproducibility, Syntropa found that a widely-used FROG gene-knockout curator never applies the knockout, returning wild-type growth for every gene. The defect was confirmed on the textbook E. coli core model and reported upstream (issue #108) with a one-line fix (PR #109). Syntropa's own corpus results reproduce at the flux level; the defect is in the comparison tool, not the corpus.