NOTE-1240: Pittis & Gabaldón — Late acquisition of mitochondria by a chimaeric host

Contribution

It turned the mito-early against mito-late debate into a measurement on the whole inferred LECA proteome. For each LECA family it measures how far the family sits from its closest prokaryotic relatives, corrected for the family’s own rate. It then asks whether that distance differs by origin, function and cell location. The answer is that it does, in a consistent order: archaeal longest, other bacterial middle, alphaproteobacterial shortest. The paper reads the order as the order of acquisition.

Key insight

If all bacterial genes came in with the mitochondrion, as in “a simple fusion model”, their stems should be alike. If they came in waves, the stems should differ. They differ: alphaproteobacterial families have the shortest stems, and other bacterial families longer ones. So the other bacterial genes were, on this reading, acquired earlier, and the host that took in the mitochondrion already carried them. In the authors’ words, “the host that engulfed the mitochondrion was already a complex cell, whose genome already harboured pathways and processes of diverse bacterial origins”.

Assumptions

• Stem length measures time. Raw stem length (substitutions per site from the prokaryotic sister to the LECA family) is divided by the median branch length inside the eukaryotic family. Schematically, sl = (R_s·T_s)/(R_e·T_e). Since T_e (time since LECA) is the same for every family, the correction removes rate only if R_s is proportional to R_e: “Under the assumption that rates pre- and post-LECA are correlated (i.e. not necessarily constant)”. The authors “cannot discard that major rate shifts … occurred in some cases”, and argue such shifts are unlikely to track origin rather than function.
• The sister group marks the origin. A family’s origin is the narrowest NCBI taxon that holds every prokaryote in its sister partition. Families whose sister mixes bacteria and archaea are assigned “cellular organisms”.
• Presence in LECA. A family counts if it has Unikont sequences and sequences from at least one Bikont group, a stricter rule than earlier work, to exclude transfers by secondary plastids.
• The sample. eggNOG v4: 238 eukaryotic, 3,318 bacterial and 130 archaeal genomes; one sequence per taxon level from 37 eukaryotes and 692 prokaryotic levels. FastTree for a first pass, then RAxML with consensus alignments, keeping families with SH-like support above 0.5 on the stem. One Lokiarchaeum metagenome is added in a separate analysis.
• Present localization and function stand for the ancestral ones. GO cellular components with experimental evidence, assigned if present in more than 10% of a family’s sequences, and COG categories.

Key results

• Four components (Fig. 1b, Table 1). A Gaussian mixture on stem lengths gave five components. The fifth (14 extreme values, enriched in archaeal origins) was dropped as outliers. Of the remaining 1,064 families:
  • Component 1, shortest (452 families): enriched in Bacteria (388), Alphaproteobacteria (49, P = 1.1e-4), Chlamydiae/Verrucomicrobia (19) and Deltaproteobacteria (29); in the mitochondrion (150); in amino acid metabolism and energy production.
  • Component 2 (284): no enriched origin; enriched in ER, Golgi and extracellular space.
  • Component 3 (234): Archaea (80), several archaeal groups and Actinobacteria (16); nucleus, nucleolus and chromosome; replication, translation and transcription.
  • Component 4, longest (94): Archaea (80); ribosome and cytosol; translation.
• Origins compared (Fig. 2a,b). Bacterial families have shorter stems than archaeal (P = 1.38e-25, two-sided Mann–Whitney). Alphaproteobacterial families are shortest, “together with families pointing to the Verrucomicrobia/Chlamydiales group”, and shorter than other bacterial families at P = 4.4e-2.
• Function and location (Fig. 2c,d). Informational families have the longest stems and metabolic the shortest. Nuclear families are longest, endomembrane families next, mitochondrial shortest.
• Origin, not function (Extended Data Fig. 2; Supplementary Information section 1, not read). Archaeal and bacterial families matched for function, dN/dS, connectivity or expression still differ in raw stem length. Raw stems give the same order as normalized ones (Fig. 2e,f).
• A positive control (Extended Data Fig. 3). In plants, families of cyanobacterial origin have shorter stems than alphaproteobacterial ones, as the later plastid predicts.
• Robustness (Extended Data Figs 4–6). The alphaproteobacterial result holds under stricter LECA definitions, other pipelines and Rochette et al.’s trees, and support thresholds up to 0.9. Halving alphaproteobacterial sampling does not change their stems. Removing Rickettsiales does not either. Three scenarios of transfer to and from the proto-mitochondrion are argued not to produce the non-alpha signal.
• Lokiarchaeota (Extended Data Fig. 7). With one Lokiarchaeum genome added, 30 families change their inferred origin to Lokiarchaeota, and these have the shortest stems among archaeal families, still longer than bacterial ones.
• Origin against location (Fig. 3, Extended Data Fig. 8). Alphaproteobacterial families associate with the mitochondrion and with energy production (P ≤ 1e-6). Archaeal families associate with the nucleus. Other bacterial families associate with endomembrane compartments. Mitochondrial families of alphaproteobacterial origin have shorter stems than mitochondrial families of other origins (P = 6e-3).
• Interpretation. Mitochondria came late. The non-alpha bacterial component came earlier, by “Serial symbiotic associations with different partners, the existence of prokaryotic consortia, or gradual waves of HGT”, which the data cannot tell apart. “Mitochondrial engulfment marked an end to massive bacterial HGT in LECA”. The mitochondrion “brought about the definitive selective advantage”.

Claims

id	claim	strength	support
C1	Among LECA families, normalized stem lengths order archaeal > other bacterial > alphaproteobacterial	strong for archaeal against bacterial, weak for alphaproteobacterial against other bacterial	P = 1.38e-25 and P = 4.4e-2; Fig. 2, Table 1
C2	Origin, not function, drives the stem differences	moderate	matched comparisons (Extended Data Fig. 2), raw stems; Supplementary section 1 not read
C3	Stem length orders acquisition in time, so the mitochondrion was acquired last	moderate to weak	rests on proportional rates before and after LECA; the plastid control supports the method’s direction
C4	The non-alphaproteobacterial bacterial component was acquired before the mitochondrion, so the host was already chimaeric	weak	follows from C1 and C3 for a difference at P = 0.044; HGT controls argued in Extended Data Fig. 6
C5	Alphaproteobacterial genes associate with the mitochondrion and energy production, archaeal with the nucleus, other bacterial with endomembranes	strong as an association	permutation tests, P ≤ 1e-6 for the first
C6	The mitochondrion ended massive bacterial transfer and gave the “definitive selective advantage”	not supported here	asserted in the closing paragraph; no analysis addresses it

Concepts

• LECA family: a monophyletic group of eukaryotic sequences, with Unikont and Bikont members, inferred to descend from one LECA gene.
• Raw stem length (rsl): the branch from the LECA family’s common ancestor to its common ancestor with the closest prokaryotic group.
• Stem length (sl): rsl divided by the median branch length from LECA to the family’s tips (ebl).
• Mito-early and mito-late: whether mitochondrial endosymbiosis drove eukaryogenesis in a simple host, or came after significant complexity.
• Chimaeric host: a host whose genome already held genes from several bacterial sources before the mitochondrion.

Connections

It builds on Rochette et al.’s inference of LECA’s mixed-origin proteome and re-analyses their trees. It cites the Lokiarchaeota genome for the host lineage. The relaxed-clock study of Kay et al. (LIT-1566, read in NOTE-1234) describes this kind of analysis (“gene tree branch lengths … normalized by relative evolutionary rates post-LECA”) and argues that clock models handle rate variation more flexibly. It agrees that the mitochondrion came late. It argues against serial endosymbiosis, one of the three explanations offered here for the chimaeric host, because most non-alphaproteobacterial duplications it dates come after the mitochondrion. Bernabeu et al. (LIT-1561) and Tobiasson et al. (LIT-1558) are not cited here; the paper predates them.

Bearing on the record

• QUESTION-277. The question names this work for “complexity before the mitochondrion”. The reading qualifies that.
  • What is shown is chimaerism, not complexity. The paper measures when genes were acquired, not when structures existed. “Complex cell” in its conclusion means a genome with bacterial pathways already in it, and the endomembrane association of those genes.
  • It leaves open whether the chimaera came from partners or from transfer. Serial symbioses, consortia and waves of HGT are all allowed. On the question’s choice between one merger and a sequence of associations, the paper is consistent with either a sequence of partnerships or one partnership after many transfers. For the routes, that is the gene-transfer route acting on the host before any endosymbiosis, with no dependence implied.
  • It says nothing on D, R, C or S. Stem lengths cannot show when partners came to depend on one another.
• The record’s description of the work. The LIT’s “Why this is here” says Kay et al. continue this paper’s argument. They continue its mitochondrion-late order, and depart from it on the bacterial component. The curation entry of 2026-10-27 (second pass) says the same; the correction goes on the LIT.
• “Its branch-length method has been disputed”. The dispute itself is not in this paper. The reading finds where the method is exposed:
  • the normalization needs rates before and after LECA to be proportional, which the authors assume;
  • stem length depends on how close the sampled sister is. The Lokiarchaeota control shows this: adding one closer genome shortens 30 archaeal stems. Archaea were sampled with 130 genomes against 3,318 bacterial ones, and Asgard archaea not at all in the main analysis (one Lokiarchaeum genome in a separate one). Part of the archaeal–bacterial gap may therefore be sampling distance, not time. The inference is mine; the authors’ halving of alphaproteobacterial sampling tests the bacterial side only.
• Filed as THEORY-1102. No instruction for machine-learning practice.

Limitations

• Length is not time. The method’s direction is checked once, on plastids. Its calibration is not.
• The key bacterial comparison is marginal. Alphaproteobacterial against other bacterial families is P = 4.4e-2 in the main test.
• Closest-sampled relatives. Stem length is measured to the closest sampled prokaryote, so uneven sampling across groups can produce differences by origin. Archaea were thinly sampled in eggNOG v4, and Asgard archaea were absent from the main analysis.
• Origin labels are coarse. Many families are assigned to broad taxa. A mixed sister partition gives “cellular organisms”.
• Present localization for past role. As the authors note, substantial retargeting has happened since.
• The closing claims are not tested. That mitochondrial entry ended massive transfer, and gave a “definitive selective advantage”, are stated without analysis.
• Not read here: the Supplementary Information, where most controls are given in full.

Open questions

• Does the stem-length order survive with dense Asgard sampling, now that many genomes exist?
• Can a method with an explicit rate model, such as a relaxed clock, reproduce the alphaproteobacterial-against-other-bacteria difference? LIT-1566 dates duplications, not acquisitions, so it does not answer this.
• Were the non-alpha bacterial genes brought by partners or by transfer? The paper names the options and cannot choose.