Contribution
It asks whether any donor other than the two accepted partners, Alphaproteobacteria and Asgard archaea, can be traced in the LECA proteome “with a level of support similar to those traced to these broadly accepted partners”. It answers yes for two bacterial phyla, Myxococcota and Planctomycetota. It adds viruses as a route: giant viruses (Nucleocytoviricota) as carriers of some prokaryotic genes into the eukaryotic lineage. It then orders the donors in time with stem lengths and infers each donor’s likely metabolism from its living relatives. It is the current statement of the Gabaldón group’s “premitochondrial symbioses” hypothesis, continuing Pittis and Gabaldón 2016 (LIT-1595).
Key insight
Calibrate on the partners everyone accepts, then see who else survives. The authors search over support thresholds for the setting that maximizes the share of trees assigned to Alphaproteobacteria and Asgard, on the assumption that those signals are real and the rest is mostly noise. At that setting two more bacterial phyla still hold a share as large as the accepted partners’. A negative control asks whether they could be misplaced alphaproteobacterial genes, and the answer is no. If their genes also cluster in time, each looks like a “transfer wave” from one donor in one period. That is what a sustained association would leave. The authors stop short of calling it symbiosis: “our data do not provide information on potential types of interaction (endosymbiotic, ectosymbiotic, parasitic, and so on)”.
Assumptions
- LECA families by spread. An orthologous group is ancestral if it has at least five species, both main eukaryotic stems, and three (relaxed) or five (strict) of nine supergroups; the donor analysis uses seven. Recent in-paralogues are collapsed to one representative, and contaminants removed by identity and taxonomy rules.
- Order-level prokaryotic pangenomes. One genome per GTDB genus; genes kept if present in 15% or more of genus representatives in an order, “the cutoff for the ‘cloud’ component of the pangenome”. Viral proteins from 1.3 million clusters are added.
- The first sister is the donor. Trees are rooted at the non-eukaryotic node farthest from the LECA clade. The donor is the most abundant taxon in the first non-eukaryotic sister, or, for viral sisters, in the first non-viral sister further back. A sister with over 25% eukaryotic sequences is not assigned.
- The accepted partners calibrate the thresholds. Thresholds (ultrafast bootstrap > 75%, “taxonomic bootstrap” > 95%, seven supergroups, donor certainty > 95%) are those that maximize the share of trees assigned to Alphaproteobacteria and Asgard, “assuming that the selected thresholds properly removed noise and kept only the reliable signals”.
- Stem length is relative time. Normalized stem length, as in Pittis and Gabaldón, with “longer normalized lengths … interpreted as indicating more ancient relatedness”. The mode of each donor’s distribution is estimated as a gamma mode by MCMC (Bernabeu, Armero & Gabaldón). Families with stems past the first minimum after the peak are dropped as later transfers, “heterotachy or ghost lineages”.
- Living relatives stand for the donor. Donor metabolism is the prevalence of pathways in extant genomes of the donor clade that share more than half of the KOs it gave to LECA. “Note that this reconstruction was different from standard ancestral state reconstruction.”
Key results
-
The LECA proteome. 12,907 (relaxed) or 7,751 (strict) monophyletic families on average; a consensus of 5,317 KEGG orthologs. 96% consistency of LECA groups across datasets, 79% after the monophyly refinement. LECA looks complex, heterotrophic and aerobic. It lacks pyruvate:ferredoxin oxidoreductase, [FeFe] hydrogenase and rhodoquinone synthesis, and its cell cycle “seemed to be incipient”. Only 3% of families might come from transfer between eukaryotic supergroups.
-
Origins (Fig. 2a). About 33% innovations (no non-eukaryotic homologue), about 53% acquisitions, about 13% unknown. 918 KOs are proposed as a revised set of eukaryotic signature proteins.
-
Donors (Fig. 2b), as shares of the LECA proteome at the calibrated thresholds:
donor share Alphaproteobacteria 3.99% Myxococcota 3.92% Asgard archaea 3.28% Planctomycetota 2.24% Chloroflexota and Gammaproteobacteria appear but fail in some datasets and are dropped. The four shares sum to 13.43% of the proteome, about a quarter of the acquisitions. The rest of the acquisitions are not assigned to a major donor at these thresholds.
-
Controls.
- Trees pointing to the new donors were not harder to resolve than alphaproteobacterial ones (Supplementary Fig. 6).
- The “verticality test”: rerunning the pipeline on alphaproteobacterial genomes with Alphaproteobacteria removed from the database assigns most trees to Gammaproteobacteria, with lower shares for more distant groups. Myxococcota and Planctomycetota do not follow that pattern (Supplementary Fig. 4).
- Reanalysis of one earlier study’s trees with this pipeline suggests that “Database completeness rather than methodological choices” drives differences between studies (Supplementary Fig. 7).
All three are in the unread Supplementary Information; the main text reports their outcome.
-
Function (Fig. 2d,e). Donor contributions are “broad and patchy”; most modules mix origins, “a high degree of chimerism”. Known trends return: archaeal information processing, alphaproteobacterial energy production.
-
Viruses. About 4.5% of trees have a viral first sister, 74% of them giant viruses (3.32% of the proteome). Behind the virus, the first prokaryotic sister includes all the major donors, Asgard most often (about 10%). Viral-route genes are enriched for signal transduction (kinases), chromatin and post-translational modification.
-
Timing (Fig. 3, Extended Data Figs 2–3). Asgard is “the most ancestral signal”. Among bacteria Planctomycetota has the longest stems, “followed by later contributions from the mitochondrial ancestor and Myxococcota, with largely overlapping transfer waves”. The order matches the depth of these taxa in one studied microbial mat (Gutiérrez-Preciado et al. 2018). Supplementary Fig. 13 is cited for functional classes giving “overlapping distributions with similar modes”, the authors’ answer to the worry that function, not time, sets stem length. It was not read here.
-
Donor metabolism (Fig. 4, Extended Data Figs 4–5), as prevalence in donor-like genomes:
- Asgard ancestor: no complete methyl-coenzyme M reductase, so “may not have a methanogenic nature”; fermentation (ethanol 0.63, butyrate 0.67).
- Mitochondrial ancestor: aerobic respiration strongly supported; photosynthesis 0.34–0.51 (“moderately supports”); no hydrogenases, “as implied by the hydrogen hypothesis and reverse-flow models”.
- Sulfate reduction in Myxococcota (0.44) and Planctomycetota (0.5); anammox in Planctomycetota weak (0.23).
-
Interpretation. The data “challenge models depicting eukaryogenesis as a one-off event resulting from a binary symbiotic interaction and support a paradigm shift towards more gradual scenarios”. Planctomycetota, with its cellular complexity and engulfment in one genus, “may lend some support” as an alternative bacterial host. The authors add at once that “our data do not provide information on the host or endosymbiotic nature of any partner”.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | The LECA proteome contains families traceable to Myxococcota and Planctomycetota with a share comparable to those traced to Alphaproteobacteria and Asgard | moderate | consistent across three datasets and both LECA criteria; verticality control; difficulty control; controls in the unread supplement |
| C2 | These signals are not misplaced alphaproteobacterial genes | moderate | the verticality test, which addresses vertical misplacement only, not transfer into the alphaproteobacterial or Asgard lineages before they reached eukaryotes |
| C3 | Genes from each donor were acquired in waves at distinct relative times: Asgard first, then Planctomycetota, then the mitochondrial ancestor and Myxococcota | weak–moderate | posterior modes of stem-length distributions; rests on stem length as a clock, which Tobiasson et al. dispute on their own data |
| C4 | Some bacterial transfers preceded mitochondrial endosymbiosis | weak–moderate | follows from C3 for Planctomycetota |
| C5 | Giant viruses mediated some gene transfers into the eukaryotic lineage, including of prokaryotic genes | moderate | tree topology with a viral first sister and prokaryotic sisters behind; strict filtering for direction |
| C6 | Eukaryogenesis involved serial interactions with several bacterial partners, not a single binary symbiosis | weak | an interpretation of C1–C4; the authors concede the data cannot tell symbiosis from other interactions |
| C7 | LECA was an aerobic heterotroph, and its Asgard ancestor was not a methanogen | weak–moderate | absence of anaerobic markers in the reconstruction; prevalence in donor-like genomes, which the authors call “incomplete and indirect” |
Concepts
- mLECA-OG: a monophyletic LECA orthologous group, after splitting groups whose eukaryotic sequences are not monophyletic in the tree.
- Innovation / acquisition / unknown: no non-eukaryotic homologue; LECA nested among non-eukaryotic sequences; direction unclear.
- Taxonomic bootstrap: the share of bootstrap trees that assign the same taxon to the LECA clade’s sister, regardless of which sequences form it.
- Transfer wave: “a set of genes potentially acquired from the same donor during the same period of time”, seen as a peak in a donor’s stem-length distribution.
- Premitochondrial symbioses hypothesis: Gabaldón’s 2018 proposal that the host had symbiotic partners before the mitochondrion.
- Donor-like genomes: extant genomes of a donor clade sharing at least 50% of the KOs that donor gave to LECA.
Connections
- Pittis and Gabaldón 2016 (LIT-1595, unread here). Its stem-length method is reused here, now in a Bayesian frame, with a study of taxon sampling (Bernabeu, Manzano-Morales & Gabaldón 2024) cited in its support. Tobiasson et al. (LIT-1558, read in NOTE-1242) test the same method and find that stem length behaves mainly like something other than time.
- Tobiasson et al. The two papers thank each other’s authors. Tobiasson et al. cite this work’s first preprint and attribute its “far greater bacterial contribution” to “more liberal criteria for core pangenome construction” letting in late transfer. This paper’s pangenome cutoff is 15% of genera in an order, against their 50% of species in a class. This paper does not answer that critique in the main text.
- Syntrophy models. It names the hydrogen–sulfur (“HS”) syntrophy hypothesis (López-García & Moreira) as supported in its sulfate reduction, but with Planctomycetota as an earlier contributor than the Myxococcota host that model proposes. It finds no hydrogenases in the mitochondrial ancestor’s relatives, against the hydrogen hypothesis (Martin & Müller, LIT-1597, unavailable here) and the reverse-flow model. It notes that both cultured Asgard archaea grow in syntrophy with sulfate-reducing partners. That is the subject of LIT-1591 (Imachi et al.), unread here and not cited by number in the passage.
Bearing on the record
- QUESTION-277 and the “live disagreement”. This is the record’s
“distributed” side, and the reading supports that placement with three
qualifications.
- The distribution is of gene donors, not of partners in a collective. The authors grant that “different forms of interactions beyond symbiosis could explain this genetic exchange”. A2’s caution, kept on LIT-1561, is the authors’ own.
- It is mitochondrion-late, like Tobiasson et al. Both papers have bacterial genes arriving before the mitochondrion. They differ on how the non-alphaproteobacterial genes are grouped (a few major donors in waves, here; diffuse transfer, there), on whether stem lengths can be read as times, and on how large the Asgard share is.
- The Asgard share is the starkest difference, but the two numbers are not the same quantity. Here Asgard is 3.28% of the reconstructed proteome under calibrated thresholds. In Tobiasson et al. it is half of likelihood weight among the best-resolved families with prokaryotic homologues. The inference is mine: neither paper’s figure can be converted into the other’s, and the record should not set them against each other as rival percentages.
- For the dimensions. Like Tobiasson et al., this bears on G (whose genes carry which functions) and not on D, R, C or S. If any of the extra donors was a partner, the question’s routes would apply to a partnership the record has no case for. The paper does not supply one.
- Gene transfer by viruses is a route to a chimeric genome that involves no partnership at all. It sits outside QUESTION-277’s five routes, which are routes to dependence between partners.
- Filed as THEORY-1105. No instruction for machine-learning practice.
Limitations
- The calibration is partly circular. Thresholds are chosen to maximize the accepted partners’ share, assuming those signals are the true ones. Any donor that survives is real only relative to that assumption. The authors state the assumption.
- The controls are in the supplement, which was not read here: verticality, tree difficulty, reanalysis of an earlier study, functional classes in the stem-length analysis, and the effect of OrthoFinder splitting.
- A liberal pangenome filter. At 15% of genera per order, more lineage-specific genes enter the database than under Tobiasson et al.’s filter. Whether that admits late transfer into eukaryotes is the other group’s objection and is not tested here.
- Stem length as time is assumed, defended by citation, and contested by Tobiasson et al.’s data. If it fails, the order of waves fails with it, and with it the claim that some transfers preceded the mitochondrion.
- The donor metabolisms are indirect. They are prevalences in living relatives, not ancestral reconstructions. No statistical test was used for functional enrichment, “as the sample sizes were non-homogeneous”.
- The mat comparison is an analogy with one mat (Fig. 3b).
- A slip. The viral share is given as “approximately 4.5% of the mLECA-OG phylogenies” in the results and as “at least 4.5% of all acquisitions” in the virus section. If acquisitions are about 53% of families, those cannot both hold. The first is the better supported, by the arithmetic in the history above.
Open questions
- Does the Myxococcota and Planctomycetota signal survive Tobiasson et al.’s stricter pangenome filter, or do theirs fail under this paper’s thresholds? Running each pipeline on the other’s database would settle which filter drives the difference.
- What would distinguish a symbiotic donor from a source of transferred genes in phylogenomic data? The paper does not say; nor does Tobiasson et al.’s.
- Is stem length a clock? The answer decides the ordering of waves and the premitochondrial claim.
- Were the viral-route genes picked up from extinct eukaryote-like lineages, as the authors suggest? That would make some “acquisitions” transfers between stem eukaryotes.