ARTICLES https://doi.org/10.1038/s41559-020-1239-x Phylogenetic analyses with systematic taxon sampling show that mitochondria branch within Alphaproteobacteria Lu Fan 1,2,3,7 ✉ , Dingfeng Wu4,7, Vadim Goremykin5,7, Jing Xiao4, Yanbing Xu4, Sriram Garg 6, Chuanlun Zhang1,3, William F. Martin 6 ✉ and Ruixin Zhu 4 ✉ Though it is well accepted that mitochondria originated from an alphaproteobacteria-like ancestor, the phylogenetic relation- ship of the mitochondrial endosymbiont to extant Alphaproteobacteria is yet unresolved. The focus of much debate is whether the affinity between mitochondria and fast-evolving alphaproteobacterial lineages reflects true homology or artefacts. Approaches such as site exclusion have been claimed to mitigate compositional heterogeneity between taxa, but this comes at the cost of information loss, and the reliability of such methods is so far unproven. Here we demonstrate that site-exclusion methods produce erratic phylogenetic estimates of mitochondrial origin. Thus, previous phylogenetic hypotheses on the origin of mitochondria based on pretreated datasets should be re-evaluated. We applied alternative strategies to reduce phylogenetic noise by systematic taxon sampling while keeping site substitution information intact. Cross-validation based on a series of trees placed mitochondria robustly within Alphaproteobacteria, sharing an ancient common ancestor with Rickettsiales and currently unclassified marine lineages. he origin of mitochondria is one of the defining events in the his- convergence artefacts8,11,14. Recently, Martijn et al.17 revisited the tory of life. Gene-network analyses1–4 and marker gene-based topic, reporting that when compositional heterogeneity of the pro- phylogenomic inference have generally reached a consensus tein sequence alignments was reduced by excluding sites from the T 5 that mitochondria have an alphaproteobacterial common ancestor , amino acid alignment, the entire alphaproteobacterial class formed yet the specific relationship of mitochondria to alphaproteobacte- a sister group to mitochondria17. Their conclusion is at odds with rial taxa remains an important evolutionary issue. Phylogenetic the long-standing phylogenetic consensus that mitochondria origi- placement of mitochondria within the tree of Alphaproteobacteria nated within Alphaproteobacteria18. While excluding composi- has been hampered by variation in mitochondrial DNA nucleotide tionally heterogenous sites might reduce phylogenetic noise and composition and substitution rates as well as strong phylogenetic mitigate systematic errors, it will also necessarily lead to loss of artefacts associating mitochondria with some fast-evolving alpha- phylogenetic information (Supplementary Note 3). At what point proteobacterial lineages such as Rickettsiales and Pelagibacterales, does the exclusion of sites exclude signals of the true evolution- resulting in erroneous branching patterns (Supplementary Note 1). ary connection between mitochondria and Alphaproteobacteria? To minimize the possible influence of long-branch attraction cou- This question was not adequately addressed in their publication, pled with convergent compositional signals, various strategies have and a similar concern has been voiced by Gawryluk19. To explore been applied, such as the use of nucleus-encoded mitochondrial this important evolutionary issue, we systematically examine the genes4,6,7, site or gene exclusion8–10, protein recoding10 and the use impact of site-exclusion methods on the phylogenetic affiliations of heterogeneity-tolerant models6,11 (Supplementary Note 2). These of mitochondria to Alphaproteobacteria. The results uncover hith- attempts have not converged but have instead generated contradic- erto unrecognized pitfalls of site-exclusion approaches. Subsequent tory hypotheses including (1) mitochondria root in or are sisters of taxon sampling and verification approaches robustly place mito- Rickettsiales7,12, which are all obligate endosymbionts (but see ref. 13); chondria within Alphaproteobacteria. (2) mitochondria are sisters of free-living Alphaproteobacteria such as Rhodospirillum rubrum9, Rhizobiales and Rhodobacterales4; (3) Results mitochondria are neighbours to a group of uncultured marine bac- We used several approaches to systematically investigate the teria14 and (4) mitochondria are most closely related to the most effects of data exclusion on determining phylogenetic relation- abundant marine surface Alphaproteobacteria, SAR11 (referred to ships between mitochondria, Alphaproteobacteria and out- as Pelagibacterales in this study)15,16. The first hypothesis has been groups. First, different site-exclusion methods were investigated reported most frequently so far, while the last has been identified by cross-validation to see if alternative trends in tree topologi- by several independent groups as being the result of compositional cal change were observed. Specifically, five metrics with different 1Shenzhen Key Laboratory of Marine Archaea Geo-Omics, Department of Ocean Science and Engineering, Southern University of Science and Technology (SUSTech), Shenzhen, China. 2Academy for Advanced Interdisciplinary Studies, Southern University of Science and Technology (SUSTech), Shenzhen, China. 3Southern Marine Science and Engineering Guangdong Laboratory, Guangzhou, China. 4Department of Bioinformatics, Putuo People’s Hospital, Tongji University, Shanghai, China. 5Research and Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy. 6Institute of Molecular Evolution, Heinrich Heine University, Düsseldorf, Germany. 7These authors contributed equally: Lu Fan, Dingfeng Wu, Vadim Goremykin. ✉e-mail: [email protected]; [email protected]; [email protected] NATURE EcOLOgy & EVOLUTION | www.nature.com/natecolevol ARTICLES NATURE ECOLOGY & EVOLUTION a b 200 220 Untreated 100 (R) Modified24 99 (R) 200 180 Stuart’s score Modified18 80 (R) Bowker’s score 160 2-score 180 59 (R) 89 (R) χ Mito-in -score Mito-out 140 160 63 (R) Evolving-rate score Unresolved 99 (R) 140 120 65 (R) Mito-in 99 (R) Mito-out 120 100 Unresolved 79 (R) -score 100 -score 100 80 Z Z 99 (R) 80 (R) 80 Less taxa: 77 (R) 60 Less taxa: 95 (R) 100 60 100 (R) 96 (R) 40 100 56 (R) 100 40 100 98 (R) 20 53 (T) 100 91 (R) 20 72 (R) 0 100 73 (T) 100 100 69 (T) 0 –20 7,000 6,000 5,000 4,000 3,000 2,000 7,000 6,000 5,000 4,000 3,000 2,000 Sites Sites c d 240 170 100 (F2) Modified24 Modified24-AlphaII 220 Modified18 99 (R) Modified24-AlphaII-MoreTaxa 150 Modified18-AlphaII 200 Mito-in Mito-out 130 100 (R) Mito-in Alpha IIb 180 97 (F2) Unresolved 100 (R) 160 110 100 140 100 (R) 90 120 Less taxa: 100 (R) 50 (AIIb) 61 (AII + AIII) 99 (R/MA9) -score -score Z 100 Z 70 100 (R) Less taxa: 100 80 100 (R) 97 (R/MA9) 85 (F2) 56 (AII + AIII + GT) 50 60 30 100 (R) 52 (R) 40 63 (R) 98 (R) 100 100 100 100 (R) 99 (R) 20 10 100 (R) 55 (R) 0 100 100 (R) –10 7,000 6,000 5,000 4,000 3,000 2,000 7,000 6,000 5,000 4,000 3,000 2,000 Sites Sites Fig. 1 | Relationships between alignment sites, the phylogenetic position of mitochondria and model fit (mean square heterogeneity across taxa test) based on different datasets, site-exclusion approaches and taxon-selection approaches. a–d, Bayesian inference with model CAT+GTR was conducted. The x axis shows the number of sites for phylogenetic inference in each dataset. The y axis shows the Z-scores of the ‘mean square heterogeneity across taxa’ posterior predictive test. Numbers beside markers show node support values (posterior probability support values) of the consensus trees. Values ≥95 are in bold. Labels in parentheses show the closest relatives of mitochondria in the tree (R, Rickettsiales; T, T. mobilis; F2, FEMAG II; AII, Alpha II; AIII, Alpha III; GT, Geminicoccus roseus and T. mobilis; MA9, MarineAlpha9). Mito-in (filled circles) means mitochondria branch within Alphaproteobacteria; mito-out (open circles) means mitochondria branch outside Alphaproteobacteria and mito-in Alpha IIb means mitochondria branch within the Alpha IIb clade of Alphaproteobacteria. a, Site-exclusion methods applied to the ‘24-alphamitoCOGs’ dataset17. Trees are shown in Supplementary Figs. 1–22. b, χ2-score-based site-exclusion and taxon-reduction methods applied to subsets of the ‘Modified24’ and the ‘Modified18’ datasets containing only the backbone, Rickettsiales and mitochondrial sequences. Trees are shown in Supplementary Figs. 35 and 37–42. c, χ2-score-based site-exclusion and taxon-reduction methods applied to subsets of the ‘Modified24’ and the ‘Modified18’ datasets containing only the backbone, FEMAGs and mitochondrial sequences. Trees are shown in Supplementary Figs. 36 and 42–48. d, χ2-score-based site-exclusion method applied to the subsets of the ‘Modified24-AlphaII’, ‘Modified24-AlphaII-MoreTaxa’ and ‘Modified18-AlphaII’ datasets. Trees are shown in Fig. 4 and Supplementary Figs. 50–58. principles—Stuart’s score, Bowker’s score, χ2-score, -score and The more sites excluded, the greater the improvement. When the evolving-rate score—were implemented (Supplementary Table 1). same number of sites were excluded, Stuart’s score, Bowker’s score Site-excluded subsets of the ‘24-alphamitoCOGs’ datasetɀ 17 were and χ2-score methods were more efficient in improving model fit generated using the five methods with a series of cut-off values than were -score and evolving-rate-score methods. We found two (Supplementary Table 2). Bayesian trees were reconstructed, and notable cases of data irreproducibility in comparison to the origi- posterior