A robust species tree for the alphaproteobacteria

Abstract

The branching order and coherence of the alphaproteobacterial orders have not been well established, and not all studies have agreed that mitochondria arose from within the Rickettsiales. A species tree for 72 alphaproteobacteria was produced from a concatenation of alignments for 104 well-behaved protein families. Coherence was upheld for four of the five orders with current standing that were represented here by more than one species. However, the family Hyphomonadaceae was split from the other Rhodobacterales, forming an expanded group with Caulobacterales that also included Parvularcula. The three earliest-branching alphaproteobacterial orders were the Rickettsiales, followed by the Rhodospirillales and then the Sphingomonadales. The principal uncertainty is whether the expanded Caulobacterales group is more closely associated with the Rhodobacterales or the Rhizobiales. The mitochondrial branch was placed within the Rickettsiales as a sister to the combined Anaplasmataceae and Rickettsiaceae, all subtended by the Pelagibacter branch. Pelagibacter genes will serve as useful additions to the bacterial outgroup in future evolutionary studies of mitochondrial genes, including those that have transferred to the eukaryotic nucleus.

FIG. 1.

Basal branching patterns from various analyses. ML bootstrap and Bayesian (underlined) support values are shown when <100%, with nodes collapsed when support was <50%. Taxa: Z, Rhizobiales (b, Bradyrhizobiaceae; x, Xanthobacter; o, other Rhizobiales); B, Rhodobacterales (r, Rhodobacteraceae; h, Hyphomonadaceae); P, Parvularculales; C, Caulobacterales; S, Sphingomonadales; L, Rhodospirillales (r, Rhodospirillum; m, Magnetospirillum; a, Acetobacteraceae); K, Rickettsiales (r, Rickettsiaceae; a, Anaplasmataceae; p, Pelagibacter); M, mitochondria; O, outgroup Proteobacteria.

FIG. 2.

Topology convergence for trees from subsets of all single-molecule families. For each masked sequence alignment, the Robinson-Foulds symmetric difference was used to compare topologies between the Bayesian tree and either (A) the corresponding ML tree or (B) the final tree (Fig. 3). The data for the protein alignments were fit to the power law curves (A) y = 1,110x^−0.70 (R² = 0.61) and (B) y = 1,150x^−0.55 (R² = 0.74). The values (zero in both cases) for the final tree were not used in the curve fitting, but their positions are marked.

FIG. 3.

Tree for the Alphaproteobacteria. This tree was the most likely found by Bayesian analysis of the concatenation of masked alignments for 104 selected protein families (33,730 characters) and had topology identical to that of the ML bootstrap consensus tree. Note that unusual support values are displayed. Traditional support values were extremely high, with 100% posterior probability Bayesian and ML bootstrap support values for each node, except for the nodes marked by values in parentheses, which show bootstrap support when <100%. The main support values presented here instead show concordance with the 106 single-gene trees (104 proteins and two rRNAs), given as the percentage of single-gene trees with very high (≥95%) Bayesian support for the node. Node concordance is an extremely stringent criterion for support that should not be interpreted as the probability that the bipartition is true. The point at which the mitochondria branch in Fig. 5 is indicated. The dashed arrow is an edit (grouping the Caulobacterales/Parvularculales/Hyphomonadaceae with the Rhizobiales rather than with the Rhodobacterales) that would increase by 1 the total concordance, with highly supported nodes from single-gene trees. The taxon used to root the tree was Geobacter sulfurreducens from the Deltaproteobacteria.

FIG. 4.

Links between nodes on species tree. The top 34 links, from an analysis of edit paths between single-protein family trees and the final tree, are mapped onto the final tree of Fig. 3. Noninteger scores for links were allowed because nonobligate edits were weighted and added to the count of obligate edits.

FIG. 5.

Mitochondrial branch. The portion for Rickettsiales is shown for the most likely tree found by Bayesian analysis of the concatenation of masked alignments for 16 selected protein families (4,830 characters), in which each node received 100% Bayesian support, except those indicated with underlined values. Identical topologies for bacterial strains arose as the ML bootstrap consensus, from which all nodes received 100% support, except those indicated in parentheses. The inset shows the topology for the mitochondrial branch from the ML bootstrap consensus, collapsing nodes with <50% support. The outgroup and alphaproteobacterial portions of the tree that are collapsed in this depiction had the same topology as those shown in Fig. 3.