NOTE-1242: Tobiasson, Luo, Wolf & Koonin — Dominant contribution of Asgard archaea to eukaryogenesis

Contribution

It re-asks the old question of where the genes of the last eukaryotic common ancestor (LECA) came from, with far more genomes than earlier surveys and a different statistic. Earlier surveys found more bacterial than archaeal genes, and archaeal ones mostly in information processing. This one finds Asgard archaea the largest single source in nearly every functional class, including much of metabolism. It also tests the stem-length method that underpins “mitochondrion-late” timing and reports that, on its data, stem length does not behave like a clock.

Key insight

The bacterial share of the eukaryotic gene set is real but has two different shapes. The alphaproteobacterial part is small and functionally coherent: it is the mitochondrion’s energy and Fe–S machinery. Everything else bacterial is diffuse, a little in every pathway, with no pathway derived from any single donor. The authors read coherence as the signature of an endosymbiont and diffuseness as the signature of piecemeal gene transfer. Hence: one host lineage (Asgard), one endosymbiont, and a continuous trickle of transferred genes, partly carried into the lineage by the Asgard ancestor itself. In their words, “we obtained no indication of another symbiotic event contributing to eukaryogenesis”.

Assumptions

• Widespread means ancestral. A family is taken to trace to LECA if its eukaryotic clade has more than five taxonomic labels, with at least one from Amorphea and one from Diaphoretickes. The authors say they “did not attempt to explicitly reconstruct the LECA gene set”.
• Soft-core pangenomes exclude late transfer. Prokaryotic genes are kept only if present in at least 50% of species in their class, and eukaryotic ones if in at least 20% (Extended Data Fig. 1d varies this from 0 to 67%). The aim is to drop lineage-specific transfer. The authors concede it “might be insufficient to compensate for extensive and continuous HGT”.
• The nearest sister is the donor. For each eukaryotic clade, up to the 12 topologically closest prokaryotic clades are each forced to be its sister in a constrained tree. The expected likelihood weight (ELW) of each is taken as “proportional to the probability of a sampled prokaryotic clade being the best estimate of the closest sister clade”. The average ELW over families (aELW) is called “association”.
• Only mid-range weights count. The core set keeps sister clades with 0.4 < ELW < 0.99. Weights of 0.99 or more are dropped as “disproportionately eukaryotic-like” and probably late transfer, except for alphaproteobacterial oxidative phosphorylation, where ELW of 1 is kept.
• Trees are unrooted in the biological sense. They are midpoint-rooted by weight: “no trees in our current study have been ‘phylogenetically rooted’”. Each tree has at most 30 eukaryotic and 70 prokaryotic sequences.
• Taxon labels are coarse and unequal. Prokaryotes are binned into 26 groups, mostly NCBI classes. All Asgard sequences share one label, “Asgard”, because their internal taxonomy is unresolved. Bacteria are split by class (for example Alpha-, Beta- and Gammaproteobacteria). A re-run under GTDB phyla is the check on this.
• Stem length as clock (tested, not assumed). Following Pittis and Gabaldón, normalized stem length is the FECA-to-LECA branch divided by the median LECA-to-tip branch length.

Key results

• Scale. 47,545 complete prokaryotic genomes plus 63 Asgard assemblies; 993 eukaryotic species (EukProt v.3) plus 72 GenBank genomes. 13,500 eukaryotic–prokaryotic orthologous clusters (EPOCs) over 3,950 KEGG orthology groups (KOGs); a core of 4,290 EPOCs over 2,100 KOGs with maximum ELW of at least 0.4.
• Asgard leads overall. In all 13,500 EPOCs, Asgard has the highest aELW, followed by Cyanobacteria, Actinomycetota, Betaproteobacteria and Alphaproteobacteria. In the core set, Asgard takes “50% of all likelihood weight”. The result held under the annealed (tree-partitioned) clusters and under GTDB taxonomy (Extended Data Figs 1e, 3b, 10a), and paralogy “alone cannot account for” it (Extended Data Figs 1d, 5).
• Information processing is “almost exclusively of Asgard descent”: replication, transcription, repair, recombination, ribosomal proteins, translation factors, protein targeting.
• Beyond information processing. Asgard associations for nucleocytoplasmic transport, protein sorting, N-linked glycosylation in the endoplasmic reticulum, GPI-anchor synthesis, sphingolipid metabolism, the mevalonate pathway, Ras-like GTPases, ubiquitin and the proteasome. Fatty-acid synthesis and decay show “a higher-than-average degree of mosaicism”, with Actinomycetota and Alphaproteobacteria alongside Asgard. Bacterial-type membrane lipid synthesis remains “unresolved”.
• Alphaproteobacteria: small and mitochondrial. Mitochondrial translation, oxidative phosphorylation, ubiquinone synthesis, and the 4Fe-4S part of Fe–S cluster assembly. ISCU is mosaic with “a weaker but clearly detectable Asgard association”; NFS1, the sulfur amino-acid pathways and the cytosolic CIA components CIAO1 and CIAO2B are Asgard. “Outside the mitochondria, alphaproteobacterial associations with Fe–S metabolism cease.”
• Other bacteria: a third, diffuse. About one third of KOGs (600 of 2,100) and EPOCs (1,580 of 4,290) have their strongest association with neither Asgard nor Alphaproteobacteria (28.6% and 36.8%, checked here). Cyanobacteria and Actinomycetota have higher global aELW than Alphaproteobacteria, but from “several small contributions across disjoint pathways”.
  • Only Alphaproteobacteria (oxidative phosphorylation) and Myxococcota (nicotinate and nucleotide-sugar metabolism) beat Asgard in any KEGG pathway with more than 20 EPOCs (Fig. 2). The Myxococcota signal reduces to nucleotidases and phosphoribosyltransferases (Extended Data Fig. 8b), plus a narrow diterpenoid signal.
  • In a strict subset (eukaryotic clade of at least 15 taxa, sister of at least 20 sequences, ELW > 0.7), 33 of 130 EPOCs point to diverse bacteria, mostly Actinomycetota and the FCB group, and 97 to Asgard. They show “no clear functional trend”.
  • The cyanobacterial signal involves no photosynthesis genes, so the authors place it pre-LECA and distinct from the plastid.
• Stem lengths (Fig. 3). 5,850 stems, peaking near 0.05.
  • For stems below 0.3 (70% of all stems), alphaproteobacterial stems are significantly longer than Asgard ones; above 0.35 they are shorter. Only the second regime reproduces Pittis and Gabaldón, “who analysed primarily stems within that range”.
  • Oxidative phosphorylation stems, “assumed to have been acquired simultaneously”, vary by more than two orders of magnitude.
  • Information-processing genes, presumably the oldest residents, have shorter than average stems.
  • Conclusion: “even normalized stem lengths are determined primarily by factors other than the gene acquisition time”. The authors suggest post-acquisition adaptation as one such factor: genes from distant bacteria had to adapt to a new cell, lengthening their stems.
• The scenario (Fig. 4). An Asgard first eukaryotic common ancestor (FECA) that “already possessed many signature features of eukaryotic cells, including the cytoskeleton and the endomembrane system”; mitochondrial acquisition “a relatively late event” with “a limited, but functionally crucial, contribution”; bacterial genes captured “at all stages”. Some genes inherited from Asgard “might ultimately be of bacterial origin”. Membrane replacement “must have occurred at an early stage”, possibly before the endosymbiosis. The origin of the nucleus as a compartment stays “unresolved”.

Claims

id	claim	strength	support
C1	Asgard archaea are the largest single prokaryotic source of the widespread eukaryotic gene set, about half of likelihood weight in the best-resolved families	moderate	aELW over 4,290 core EPOCs; robust to clustering, taxonomy and pangenome cutoffs (Extended Data Figs 1, 3, 10); depends on the 0.4–0.99 window and the single Asgard label
C2	Core information-processing systems are almost exclusively of Asgard descent	strong	confirms and extends prior work; consistent across categories (Fig. 1c, Extended Data Fig. 4)
C3	The alphaproteobacterial contribution is small, coherent and almost entirely mitochondrial	moderate	Extended Data Figs 4c, 6; agrees with prior Fe–S and carbon-metabolism work; under-sampling of the mitochondrial ancestor’s relatives acknowledged
C4	Contributions from other bacteria are diffuse, with no pathway from a single donor, and are best read as isolated transfers, not further endosymbioses	weak–moderate	the diffuseness is shown (Figs 2, Extended Data Figs 7–9); the reading of diffuseness as “not symbiosis” is an interpretation; Myxococcota possibly under-detected, by the authors’ account
C5	Normalized stem length is determined mainly by factors other than acquisition time	moderate	the reversal at 0.3–0.35, the spread in oxidative phosphorylation stems, short information-processing stems, 20 technical replicates (Extended Data Fig. 10b); the alternative factor is suggested, not shown
C6	Key features of eukaryotic organization (cytoskeleton, endomembranes) evolved in the Asgard lineage before the mitochondrion was taken up	weak–moderate	inferred from gene ancestry; the paper does not time events, having argued its own timing tool unreliable
C7	The dominant Asgard contribution “is here to stay”	weak	the authors’ expectation, against their own caveat that sampling could “lead to a substantial modification of the model”

Concepts

• EPOC: a eukaryotic–prokaryotic orthologous cluster, one eukaryotic cluster plus all its significant prokaryotic HMM hits.
• ELW / aELW: expected likelihood weight of a constrained topology in which a given prokaryotic clade is sister to the eukaryotic clade; aELW is its average over EPOCs, called “association”.
• Soft-core pangenome: the genes present in at least a set fraction of the species of a taxonomic class.
• FECA and LECA: the first eukaryotic common ancestor (here, the last common ancestor of eukaryotes and their prokaryotic sister) and the last.
• Normalized stem length: the FECA-to-LECA branch divided by the median LECA-to-tip branch, proposed by Pittis and Gabaldón as a relative age of acquisition.
• Binary vs. two-endosymbiosis scenarios: in the paper’s framing, “binary symbiogenesis” has an Asgard host take up the alphaproteobacterium and later switch to bacterial lipids. The alternatives have a bacterium engulf an Asgard archaeon first, then a second endosymbiosis give the mitochondrion.

Connections

• Pittis and Gabaldón 2016 (LIT-1595, unread here). The paper calls it “a pivotal study” and reproduces its result only for long stems. Its own conclusion, that stem length mostly does not track time, undercuts the method, though the paper still endorses a late mitochondrion on other grounds.
• Bernabeu et al. (LIT-1561, read in NOTE-1237). Cited as ref. 65 in its first preprint version, under the title “Diverse ancestries reveal complex symbiotic interactions during eukaryogenesis”. The paper says that report, “using more liberal criteria for core pangenome construction, suggested a far greater bacterial contribution than observed here, potentially due to late HGT”. The two groups thanked each other for comments.
• Santana-Molina et al. (LIT-1564, read in NOTE-1239). Cited as ref. 58 for the alphaproteobacterial association with mitochondrial carbon metabolism.
• Vosseberg et al. 2021 (ref. 26) on pre-LECA duplications of Asgard genes, cited for “their prominent role in eukaryogenesis”. Not in the record.
• Syntrophy. The paper invokes “the syntrophic lifestyle of many Asgard archaea” as a route for bacterial genes into the Asgard lineage, citing López-García and Moreira (2020). Its introduction cites Imachi et al. (LIT-1591, unread here) for Asgard archaea “found in consortia with bacteria, particularly Myxococcota”. It treats the two-endosymbiosis syntrophy models (hydrogen–sulfur transfer, with Myxococcota as partner) as “less parsimonious than binary scenarios” but says it may “capture only part of a principal signal from Myxococcota”.

Bearing on the record

• QUESTION-277, and the framing of the “live disagreement”. The question’s summary sets three programmes side by side: late mitochondrion after a complex host (LIT-1566), mostly Asgard inheritance (this work) and waves of bacterial transfer (LIT-1561). The reading changes that picture.
  • This paper is itself mitochondrion-late. It favours a complex Asgard FECA and a mitochondrion acquired “relatively late”. On timing it sits with Kay et al. (as the record summarizes them; LIT-1566 is unread), not against them.
  • It is two-partner on symbiosis, not on genes. It accepts many bacterial donors, “both before and after endosymbiosis”. What it denies is that any of them was a further symbiotic partner.
  • Its disagreement with Bernabeu et al. is real but narrow. It is about whether the non-alphaproteobacterial genes cluster by donor (a few major donors, in waves) or are diffuse; whether stem lengths can time them; and whether a liberal pangenome filter lets late transfer in. Neither paper’s data can show what kind of interaction moved the genes. Bernabeu et al. say so of their own data (NOTE-1237).
  • For the dimensions. The paper bears on G, how essential functions are split across partner genomes: on its account the host lineage supplies nearly everything except the mitochondrion’s energy and Fe–S machinery. It says nothing about D, R, C or S, and nothing about how the partners came to depend on each other.
• NOTE-159 reports the SEP entry’s suggestion that eukaryogenesis may be “first and foremost a story about diversification within Archaea”. This paper is the strongest version of that view so far in the record. But it still calls the mitochondrion “functionally crucial”, so the endosymbiosis remains in the story.
• THEORY-141 and THEORY-418. Nothing here tests them. The paper models gene origins, not selection or organization.
• Filed as THEORY-1104 (the composition result and its reading) and THEORY-1101 (stem length is not a clock). No instruction for machine-learning practice.

Limitations

• Sampling. The authors’ “above all” caveat. Most Asgard genomes are incomplete metagenomic assemblies. Sequenced Alphaproteobacteria are biased to reduced symbionts and parasites, and the mitochondrial ancestor “apparently belongs outside of the currently known diversity”. Other phyla are “even more severely under-sampled”.

• What “dominant” measures. It is a share of likelihood weight among families that pass a 0.4 confidence floor and a 0.99 ceiling, in trees of at most 100 sequences, among at most 12 candidate sisters. It is not a count of genes in a reconstructed LECA proteome. Families with no prokaryotic homologue are not in it.

• An unequal label. All Asgard sequences share one label while bacteria are split into classes. The GTDB phylum re-run is the authors’ control and the Asgard lead held. The rest of the ranking was less stable: “the correlations for other taxa were lower”.

• The ELW window is asymmetric. ELW ≥ 0.99 is excluded as late transfer except for alphaproteobacterial oxidative phosphorylation. This favours the alphaproteobacterial count in that one pathway; it is not shown what it does elsewhere.

• The scenario is not tested. That the cytoskeleton and endomembranes were Asgard-borne before the mitochondrion is read off gene ancestry. The paper has no timing method it trusts.

• The stem-length alternative is untested. “Post-acquisition adaptation” is “at least partially supported” by long stems from diverse bacteria, and no more.

• Slips in the PMC copy.

  • In the Discussion, refs 64–66 seem misnumbered. “As posited in some previous models26,64” cites Koonin & Yutin on giant DNA viruses (64). “Mosaic gene composition of prokaryotic genomes65,66” cites the Bernabeu preprint and an archaeal-lipid engineering paper.
  • Extended Data Fig. 9b’s caption excludes Asgard, Euryarchaeota, TACK, DPANN, Cyanobacteriota and Alphaproteobacteria from the “diverse” category. The text splits the same 130 EPOCs into 33 diverse and 97 Asgard, leaving no room for the other excluded groups. It is unclear whether none occurred or the split is reported loosely.

  None of these changes a result.

Open questions

• Does the Asgard share hold when Asgard genomes are complete and when the mitochondrial ancestor’s closer relatives are sampled? The authors expect it will.
• Is the Myxococcota signal larger than seen here? A closer look at the steroid pathway downstream of diterpenoids, cut by the paper’s scope criteria, would bear on the syntrophy models.
• What, if not acquisition time, sets stem length? A test would be stem lengths for gene sets of known common origin and different functional constraint, beyond the oxidative phosphorylation case.
• How did an Asgard cell come to have bacterial-type membranes? The paper names this, with the origin of the nucleus, as the open problem.