Matteo P. Ferla; J. Cameron Thrash; Stephen J. Giovannoni; Wayne M. Patrick

“ We focused on rRNA genes, rather than protein-encoding genes, in order to recover phylogenetic signals that may have been obscured due to the genomic AT-richness and ancient divergence events that are hallmarks of the Alphaproteobacteria. We focused on concatenated 16S-23S rRNA gene sequences to give better support than 16S alone, while also allowing the inclusion of many more taxa than what is currently feasible with available genomes for concatenation studies. ”

“ The problematic internal organisation of the Rhodospirillales is even more aggravated in the datasets that include species represented only by 16S rRNA gene sequences (Figure S20) . Some species classified as Rhodospirillales actually belong to the separate orders Kiloniellales, Kordiimonadales and Sneathiellales, while several Rhodospirillales species, such Tistrella mobilis (whose genome was published too recently to be included in this study, [33] ) , form clades that are basal to the other Rhodospirillales. ”

“ However, when the species sampling was increased or when maximum likelihood was used instead of Bayesian inference, the resulting trees supported the membership of Pelagibacterales within the clade Rickettsiales. In that study, mitochondria, Odyssella thessalonicensis and Magnetococcus marinus were omitted. On the other hand, Georgiades et al. found that the Pelagibacterales and the mitochondria formed a sister clade to the Rickettsiales, whereas Odyssella thessalonicensis was found to be basal to the clade composed of the remaining alphaproteobacterial orders [16] . ”

“ One of the subsequent clades has the Pelagibacterales subtending the Anaplasmataceae, Midichloria mitochondrii, the Rickettsiaceae and the mitochondria (if present) . The other clade has the Holosporaceae at the base, the Rhodospirillales as the next clade branching out, followed by the Sphingomonadales, then the remaining orders. The Holosporaceae is represented by Odyssella thessalonicensis and Caedibacter caryophilus, and is monophyletic in six out of the eight trees, but in two it resolves paraphyletically at the same location. ”

“ GC bias
The debate over alphaproteobacterial phylogeny focuses mainly on the location of the Pelagibacterales (SAR11 clade) , and particularly whether it is a sister clade to the Rickettsiales or whether they are artefactually attracted due to shared low GC content [26] - [18] . Despite concerns about the effects of such compositional biases on tree topologies [25] , little is known about the effect of genomic GC composition on 16S and 23S sequences. ”

“ For example, the primary clade composed of Rickettsiaceae, Anaplasmataceae, Midichloria mitochondrii and Pelagibacterales – in which mitochondrial sequences also cluster – and the clade with a basal Holosporaceae and various orders are supported much more strongly in the datasets with mitochondria, especially the Arb-Sina-aligned datasets. ”

“ We also introduce RYMK-recoding and show that it avoids some of the information loss in RY-recoding. We demonstrate that the topology of the Alphaproteobacteria is sensitive to inclusion of several groups of taxa, but it is less affected by the choice of alignment and rate methods. The majority of topologies and comparative results from Approximately Unbiased tests provide support for positioning the Rickettsiales and the mitochondrial branch within a clade. This composite clade is a sister group to the abundant marine SAR11 clade (Pelagibacterales) . ”

“ The values in the boxes are the bootstrap support percentages for the given bipartitions from each of the eight inferences for each set (with or without mitochondria, aligned with Arb-Sina or Muscle, GTRΓ or GTRCAT model) . In the cases where the final tree did not agree with the proposed topology on the location of a clade the bootstrap support for the proposed bipartition is represented in red, despite being absent in the final tree. ”

“ Additional evidence against compositional bias-driven attraction comes from the recoded datasets. The RY-recoded trees were less supported on many nodes, which may be due to the presence of two instead of four character states. We introduced RYMK-recoding to help overcome the limitation caused by the reduced number of character states and we found that the resulting trees concurred with the regular dataset, regarding the clustering of the Pelagibacterales with the mitochondria, Rickettsiaceae and Anaplasmataceae. ”

“ This approach has allowed us to update the canonical 16S rRNA gene tree of the Alphaproteobacteria with additional important taxa that were not previously included, and with added resolution provided by concatenating the SSU and LSU genes. We investigated the topological stability of the Alphaproteobacteria by varying alignment methods, rate models, taxon selection and RY-recoding to circumvent GC content bias. ”

“ In 2007, Williams et al. used a thoroughly selected set of 104 protein-encoding genes and found that Candidatus Pelagibacter ubique (P. ubique) was basal in the Rickettsiales and that mitochondria were sister to the Rickettsiaceae and Anaplasmataceae [3] . ”

“ The trees inferred from the RY-, MK- and RYMK-recoded datasets had even more diverse topologies (bootstrap summaries in Figures S6–S8) . The Rhodospirillales were either paraphyletic with respect to the clade composed of Sphingomonadales, Rhizobiales, Rhodobacterales and Caulobacterales, or paraphyletic or polyphyletic in respect to the Holosporales. The Pelagibacterales was found basal to the remaining Alphaproteobacteria (except for Magnetococcus marinus) or basal to the clade containing the Caulobacterales in all but three trees. ”

“ In the case of marker selection, many studies choose highly conserved housekeeping genes (e.g. [3] ) , but in some cases subsets of these genes with particular properties are chosen (e.g. [13] ) , and the criteria for inclusion vary between studies. ”

“ Full genome sequences of these species may therefore increase the resolving power of future studies.
Approximately Unbiased tests
Trees from the 16S, 23S, complete with regular encoding, complete RY-recoded and complete RYMK-recoded datasets were assessed against the various alignments with the Approximately Unbiased (AU) test [29] . A p-value of 0.05 was used as the cut-off, such that trees with p-values below this number could be rejected based on an alignment. Generally, trees from a given dataset were unable to be rejected by their respective alignments from either Arb-Sina or Muscle. ”

“ Bootstrap values (n = 1,000) are indicated at the nodes. Red arrows indicate how a taxon or clade differs in the other trees. Insets show clades which are particularly rich in unsequenced genomes (black text) compared to sequenced genomes (green text) . The internal topology of the Rhodospirillales differs between trees, but the six constant subdivisions are highlighted. (EPS)
File:Pone.0083383.s010.epsClick here for additional data file. Figure S11 16S and 23S rRNA gene trees for the complete dataset, with and without mitochondria. ”