Source
Tobiasson, Luo, Wolf & Koonin (2026), Nature 650: 141–149, LIT-1558, read in full in NOTE-1242 from the PubMed Central copy (PMC12872458). The sections “Origins of eukaryotic proteins” through “Sparse contributions from other bacteria”, the Discussion, Methods, Figures 1, 2 and 4, and Extended Data Figures 1–9, read from captions.
What was actually shown
- The data. 13,500 eukaryotic–prokaryotic orthologous clusters (EPOCs) built by HMM profile searches. The genomes are 47,545 complete prokaryotic ones plus 63 Asgard assemblies, and 993 eukaryotic species. Prokaryotic genes were filtered to those in at least 50% of a class’s species, to keep out late transfer.
- The statistic. For each eukaryotic clade, up to 12 nearby prokaryotic clades are each forced to be its sister, and expected likelihood weights (ELW) are compared. The core set keeps 4,290 EPOCs whose best sister has 0.4 < ELW < 0.99.
- Asgard leads. It has the highest average weight overall, and 50% of all weight in the core set. That held under tree-based re-clustering, under GTDB phyla in place of NCBI classes, and across pangenome cutoffs of 0–67%.
- Function. Information processing is “almost exclusively of Asgard descent”. Asgard also leads in trafficking, glycosylation, GPI anchors, sphingolipids and the mevalonate pathway. Alphaproteobacteria lead only in mitochondrial translation, oxidative phosphorylation, ubiquinone and 4Fe-4S assembly. Outside the mitochondrion even Fe–S handling is Asgard.
- Other bacteria. They are strongest for 600 of 2,100 KOGs and 1,580 of 4,290 EPOCs, spread thinly over most pathways. Only Myxococcota, for nucleotide-sugar phosphatases and transferases, beats Asgard in a pathway with more than 20 EPOCs besides Alphaproteobacteria in oxidative phosphorylation. In a strict subset, 33 of 130 EPOCs are bacterial, with “no clear functional trend”.
- The reading. “we interpret these diffuse associations with diverse bacteria as indications of several bacterial contributions, each with a limited functional scope, probably originating from isolated HGT events rather than additional endosymbioses.” And: “we obtained no indication of another symbiotic event contributing to eukaryogenesis.”
What could have failed: the Asgard lead could have depended on the clustering, the taxonomy or the pangenome cutoff, and by the paper’s checks it did not. The bacterial remainder could have shown pathway-level coherence for some donor, as the mitochondrion’s does, and apart from a narrow Myxococcota signal it did not.
What this does not say
- Not that most LECA genes are Asgard-derived by count. The 50% is a share of likelihood weight among families with prokaryotic homologues that pass a confidence window. Bernabeu et al., on a different measure, put Asgard at 3.28% of the whole proteome. The two figures are not comparable.
- Not that no other partner existed. Diffuse transfer is what the authors infer from the absence of functional coherence. A partner whose genes were mostly lost, or replaced, or spread thin, would also leave a diffuse trace. The inference is mine.
- Not a timing result. That the cytoskeleton and endomembranes evolved in the Asgard lineage before the mitochondrion is the scenario the composition is “most compatible with”. The paper times nothing, and argues that stem lengths cannot (THEORY-1101).
- Not robust to the sampling the authors flag. They allow that better sampling “might lead to a substantial modification of the model”. They also allow that the Myxococcota signal may be only “part of a principal signal”.
Connections
- THEORY-1105 (Bernabeu et al.): rival on two points, compatible on two. Rival on whether the non-alphaproteobacterial genes are diffuse or form a few major donors’ waves, and on the size of the Asgard share. On the first, Tobiasson et al. say the other’s larger bacterial signal may be late transfer let in by “more liberal criteria for core pangenome construction”. Compatible on a mitochondrion acquired late and on bacterial genes arriving before it. Neither can show what kind of interaction moved the genes.
- THEORY-1109 (Santana-Molina et al.): compatible, with a tension. Both find the alphaproteobacterial contribution mitochondrial. In central carbon metabolism, that paper counts more alphaproteobacterial than Asgard enzymes (17 to six), against this paper’s Asgard lead across “large parts of the core metabolic network”. The measures differ.
- THEORY-1101, from the same paper, is why this one makes no timing claim of its own.
- LIT-1566 (Kay et al., read in NOTE-1234). The record summarizes it as a complex archaeal host before a late mitochondrion. That is this paper’s scenario, not its rival.
- QUESTION-277. It bears on G: on this account the host lineage supplies nearly every essential function except the mitochondrion’s energy and Fe–S machinery. It does not bear on D, R, C or S.
Note of 2026-10-30: the Myxococcota signal
The old title said the other bacteria’s genes were scattered “with no pathway from one donor”. By the paper as read in NOTE-1242, Myxococcota beats Asgard in nicotinate and nucleotide-sugar metabolism, in pathways with more than 20 EPOCs, though the signal reduces to nucleotidases and phosphoribosyltransferases. The title now names it; the authors’ reading of the remainder as diffuse transfer is unchanged.
Note of 2026-10-30: promoted, and the wider thesis filed beside it
The twenty-second status review promoted this reading under ADR-065: it reports what Tobiasson and colleagues’ survey finds and how the authors read it, and NOTE-1242 read the paper in full. Whether eukaryogenesis had one host lineage, one endosymbiont and piecemeal transfer is THEORY-1197’s question, Proposed, which keeps this entry’s old promote_when. The record’s own critique, once in the title, is now CLAIM-508, Proposed.