Late Mitochondrial Origin Is an Artifact

Abstract

The origin of mitochondria was a crucial event in eukaryote evolution. A recent report claimed to provide evidence, based on branch length variation in phylogenetic trees, that the mitochondrion came late in eukaryotic evolution. Here, we reinvestigate their claim with a reanalysis of the published data. We show that the analyses underpinning a late mitochondrial origin suffer from multiple fatal flaws founded in inappropriate statistical methods and analyses, in addition to erroneous interpretations.

Keywords: endosymbiosis; major evolutionary transitions; origin of mitochondria.

Fig. 1.—

Distribution of 1,078 stem length (sl) values. (a) sl as a function of sample size (eukaryotic sequence length). (b) Fit of the sl values to a five Gaussian mixture model (top), and to a log-normal model (bottom). AIC: Akaike information criterion, BIC: Bayesian information criterion. Note that the log-normal distribution is strongly preferred. (c) 1,000,000 random samples from a 5-Gaussian mixture model were generated with the parameters estimated from the observed stem length (sl) data obtained by Pittis and Gabaldón (2016). Top: BIC for the fitting of each of the 1,000,000 random samples (bars) and the observed sl data (dashed line) to a Gaussian mixture model from 1 to 7 components. Bottom: Difference between BIC for the fitting to a log-normal distribution and to a Gaussian mixture model from 1 to 7 components for each of the 1,000,000 random samples (bars) and for the observed sl data (dashed line). BIC, Bayesian information criterion.

Fig. 2.—

Phylogenetic tree and sl derivation for COG4178_01, an ABC transporter present in 25 eukaryotic taxa. Which eukaryotic branch length (ebl) should be used to calibrate the raw stem length (rsl)? The minimal, median, and maximal lineages are highlighted in magenta. Perchance it is a moot question, as in the absence of a LECA-to-present molecular clock, none of the resulting sl values convey meaningful information. The ratio of longest to shortest ebl is 2.15, a value representative of the data set as 579 other trees have larger ratios.

Fig. 3.—

Stem length (sl) distributions among eukaryotes and the fit to Gaussian mixture and log-normal models. (a–j) Histograms of group specific sl for the largest clade containing only group members with taxa from at least two taxonomic subgroups. Values in panel (a) are from Pittis and Gabaldón (2016), values in panels (b–j) were calculated from the trees in Pittis and Gabaldón (2016). In panels (a–j), the rightmost bin contains all values ≥3; AIC: Akaike information criterion, BIC: Bayesian information criterion. (k–l) Empirical cumulative distribution functions for the sl values in panels (a–j), in sl scale (k) and log(sl) scale (l). Colors match the colors used in (a–j).

Fig. 4.—

Comparison of stem length (sl) values in classification of the prokaryotic sister clade as α-proteobacterial or bacterial but non-α-proteobacterial. (a) Unfiltered: full data set analyzed in Pittis and Gabaldón (2016). (b–f) Data sets obtained by exclusion of questionable, low-quality, or nonindependent sample points. n: number of observations, U-test: Mann–Whitney U test; CV, coefficient of variation.