The genus Burkholderia was originally introduced to accommodate an assemblage of seven Pseudomonas species (Yabuuchi et al., 1992), two of which were later transferred to Ralstonia (Gillis et al., 1995; Yabuuchi et al., 1995). Since then, the number of Burkholderia species has grown substantially, to about 108 in 2015 (Estrada-de los Santos et al., 2016), spanning a range of human, animal and plant pathogens, as well as numerous strains with significant biotechnological potential (Depoorter et al., 2016; Estrada-de los Santos et al., 2016). The latter includes the so-called plant beneficial and environmental (PBE) species (Suárez-Moreno et al., 2012), many of which are plant-associated (e.g., those with plant growth promoting activities, the symbiotic diazotrophs and free-living species with diazotrophic, bioremedial and antibiotic activities) (Depoorter et al., 2016; Estrada-de los Santos et al., 2016). Because of this heterogeneity, new genera [e.g., ‘Caballeronia’ (Gyaneshwar et al., 2011) and ‘Paraburkholderia’ (Sawana et al., 2014)] has been introduced to accommodate most of the PBE species (Oren and Garrity, 2015a,b, 2017), while retaining the pathogens in Burkholderia sensu stricto. Most recently, a third genus, Robbsia was introduced to accommodate the phytopathogen previously referred to as B. andropogonis (Lopes-Santos et al., 2017).

Overall, the taxonomy of Burkholderia sensu lato remains in significant flux (Estrada-de los Santos et al., 2016). With their review of the group, Estrada-de los Santos et al. (2016) recognized two monophyletic groups [Groups A and B; A consists of Caballeronia and Paraburkholderia as circumscribed by Gyaneshwar et al. (2011) and Sawana et al. (2014), respectively, while Group B includes most of the notable human, animal and plant pathogens, as well as the so-called “B. cepacia complex”]. They showed that B. andropogonis (now Robbsia andropogonis) is separated into its own group, and they designated two so-called “Transition Groups” (i.e., 1 and 2; neither were supported as monophyletic and both contained mainly environmental species). Since then, Dobritsa and Samadpour (2016) have proposed the transfer of species in Transition Group 2 to a new genus. However, to complicate the issue, this new genus was named “Caballeronia” although its proposed usage is not synonymous with the one previously proposed by Gyaneshwar et al. (2011) for accommodating all the PBE isolates.

The proposals to split Burkholderia sensu lato were based almost entirely on evidence from 16S ribosomal RNA (rRNA) phylogenetic trees with limited and in some cases no statistical support (Gyaneshwar et al., 2011; Dobritsa and Samadpour, 2016; Eberl and Vandamme, 2016). Even phylogenies based on conventional multilocus sequence analysis (MLSA) using the combined sequence information for 4–7 genes (Gevers et al., 2005) produced phylogenies in which the major groups were not supported as monophyletic (Estrada-de los Santos et al., 2013). Also, the most comprehensive phylogenetic hypothesis to date (based on 21 conserved gene sequences) lacked sufficient representation across this diverse assemblage (Sawana et al., 2014). Thus, uncertainties remain regarding the genomic and evolutionary coherence of Burkholderia sensu lato and its lineages. This, in turn, blurs the boundaries of the Burkholderia sensu lato genera currently recognized and also casts doubt on the appropriateness and legitimacy of their taxonomic circumscriptions.

In this study, we aimed to resolve the relationships within Burkholderia sensu lato, particularly those pertaining to Paraburkholderia and Caballeronia, by making use of whole genome sequence data. For this purpose, we utilized all of the sequences for type strains (or appropriate representatives) available in the public domain. To increase representation of the so-called environmental species, we also determined the sequences for eight additional taxa via Phase III of the GEBA (Genomic Encyclopedia of Bacterial and Archaeal type strains) project (Whitman et al., 2015). These included the rhizobial species P. aspalathi (Mavengere et al., 2014) and P. diazotrophica (Sheu et al., 2013), and the soil bacteria P. hospita (Goris et al., 2002), P. phenazinium (Viallard et al., 1998), P. sartisoli (Vanlaere et al., 2008), P. terricola (Goris et al., 2002), as well as the plant-associated diazotrophic species P. caballeronis (Martínez-Aguilar et al., 2013) and P. tropica (Reis et al., 2004).

To achieve our primary goal of resolving the generic boundaries and relationships within Burkholderia sensu lato, we endeavored to use as wide a taxon selection as possible. Therefore, to complement the genome data already in the public domain for 78 species in this assemblage, we determined the whole genome sequences for an additional eight PBE species. The genomes for these species exhibited similar characteristics as those of other members of Burkholderia sensu lato (see Table 2). This was particularly true in terms of genome size and total numbers of genes encoded. Some differences were observed in G+C content. As have been observed before (Gyaneshwar et al., 2011; Estrada-de los Santos et al., 2013; Sawana et al., 2014), the Burkholderia sensu stricto genomes were higher in G+C content than Paraburkholderia and Caballeronia, which were similar in G+C content. Future studies aimed at exploring genome architecture and the functions encoded on these genomes will undoubtedly reveal traits and processes that more clearly characterize the various lineages of this economically important assemblage of bacteria.

For inferring a robust phylogeny that are congruent with the evolutionary history of Burkholderia sensu lato, we attempted to avoid or limit the effect of factors known to negatively impact phylogenetic trees (Philippe et al., 2011). The criteria used for generating the respective datasets therefore focused on the use of orthologous loci and on limiting the effects of non-phylogenetic signal. The former was accomplished by using EDGAR to identify orthologous protein-coding genes (Blom et al., 2016). The orthologous nature of a large proportion of the genes included in our final dataset was also congruent with expectations of the so-called complexity hypothesis (Jain et al., 1999; Cohen et al., 2011). In silico functional analysis showed that about 44% of these genes represented “informational genes” with products that potentially participate in processes related to DNA replication and repair, transcription and translation. Due to the complexity of their interactions with different proteins and other cellular constituents, these genes are typically less prone to horizontal gene transfer (Jain et al., 1999; Cohen et al., 2011). Our approach for identifying suitable gene sequences from which to infer the phylogeny thus lessened the chances considerably of accidentally using paralogous or xenologous gene copies (Koonin, 2005).

To limit the amount of non-phylogenetic signal in the data, a three-tiered approach was used. [i] The final dataset was large, almost devoid of missing sites (i.e., where genes in some taxa were not sequenced in their entirety) and consisted of the sequences for 106 genes common to Burkholderia sensu lato and its Ralstonia and Cupriavidus outgroups. Such large datasets typically outperform smaller datasets that only contain the sequences for one or a few genes (Daubin et al., 2002; Coenye et al., 2005; Galtier and Daubin, 2008; Bennet et al., 2012; Chan et al., 2012). This is because the “true” phylogenetic signal inherent to orthologs included in such a large dataset will dominate the analysis and typically attenuate or dilute the effects of spurious non-phylogenetic signal associated with one or a few genes (Daubin et al., 2002; Andam and Gogarten, 2011). [ii] Lack of evolutionary independence among loci may contribute to non-phylogenetic signal during tree inference (Gevers et al., 2005). For example, genes that are clustered or whose products are involved in similar or linked processes typically experience similar evolutionary forces, which is accordingly also reflected in their phylogenies (i.e., these reflect the linked evolutionary history of the genes and not the evolutionary history of the species or genus). However, the 106 genes used for resolving Burkholderia sensu lato were not significantly clustered (see Supplementary Figures S1, S2), while their inferred products were predicted to participate in diverse functions (see Supplementary Table S3). [iii] Substitution saturation is another important source of non-phylogenetic signal (Philippe and Forterre, 1999; Xia et al., 2003; Jeffroy et al., 2006; Philippe et al., 2011), and to compensate for its limited occurrence in our datasets, all phylogenetic analyses utilized independent substitution models for each gene partition. This approach proved fairly successful as both the nucleotide and amino acid data supported congruent trees with highly similar topologies.

Our maximum likelihood analyses of the aligned amino acid and nucleotide sequences for 106 genes produced a highly supported phylogeny for Burkholderia sensu lato (see Figure 2). Most of the branches on this 92-taxon phylogeny received full (100%) bootstrap support. The generation of such a well-resolved phylogeny is, however, not unusual when large datasets containing the information of numerous genes are used. Various previous studies have shown the value of this approach for resolving systematic questions at taxonomic ranks from the genus level and up (e.g., Zhang et al., 2011; Richards et al., 2014; Ormeno-Orrillo et al., 2015; Rahman et al., 2015). Our study thus adds to the growing body of work demonstrating how genome-informed taxonomic decisions represent more robust solutions than those based solely on 16S rRNA or conventional MLSA.

Based on our results, boundaries can for the first time be confidently demarcated for Burkholderia sensu stricto, Caballeronia and Paraburkholderia. These three genera, respectively, represent three of the five distinct lineages recovered among the Burkholderia sensu lato species. Burkholderia sensu stricto is represented by a large clade that includes the B. cepacia complex as well as the B. pseudomallei group, and consists primarily of pathogenic species, as suggested previously (Gyaneshwar et al., 2011; Sawana et al., 2014; Estrada-de los Santos et al., 2016). The Caballeronia clade includes environmental species that initially formed part of Transition Group 2 of Estrada-de los Santos et al. (2016) and that were transferred to the genus Caballeronia by Dobritsa and Samadpour (2016). This clade also includes all 13 of the recently described and validated Burkholderia glathei-like species (Oren and Garrity, 2016; Peeters et al., 2016). Based on these findings, we propose the formal inclusion of these species in the genus Caballeronia (sensu Dobritsa and Samadpour, 2016) (see Table 3 for details of the proposed new combinations). The inclusion of these taxa into Caballeronia raises the number of species to 25. Based on our analyses of their genomes, these species do not encode common nod or nif and fix loci, suggesting that none of the current Caballeronia species represent rhizobia or diazotrophs.

The Paraburkholderia clade is represented by diverse species, including both free-living and symbiotic diazotrophs, as well as environmental species. Although most of the taxa in this clade have already been formally transferred to Paraburkholderia (Sawana et al., 2014) and the novel combinations have been validated (Oren and Garrity, 2015a,b), this genus should also clearly include ‘P. acidipaludis’ (Aizawa et al., 2010) isolated from water chestnut as suggested by Sawana et al. (2014). This novel combination, however, still awaits validation. Interestingly, Paraburkholderia separates into two fully supported sub-clades, one including at least 23 species (spanning from P. caledonica to P. hospita in Figure 2) and the other including 11 species (P. kururiensis to P. sacchari in Figure 2). Although we could not identify any obvious reason for this split, future studies should explore its possible biological and taxonomic significance.

The two remaining lineages of Burkholderia sensu lato is represented by R. andropogonis [a pathogen of sorghum (Lopes-Santos et al., 2017)] and P. rhizoxinica [a member of Transition Group 1 of Estrada-de los Santos et al. (2016)]. Various previous studies have pointed out that these species should be excluded from Burkholderia sensu stricto, Caballeronia and/or Paraburkholderia (e.g., Estrada-de los Santos et al., 2013, 2016; Dobritsa and Samadpour, 2016). In fact, they have been suggested to represent new genera (Estrada-de los Santos et al., 2013; Dobritsa and Samadpour, 2016). This debate ultimately culminated in the introduction of the new genus Robbsia to accommodate R. andropogonis (Lopes-Santos et al., 2017). Based on our findings, the taxonomy of P. rhizoxinica requires similar revision. This species is definitely not a member of Paraburkholderia despite having been moved there from Burkholderia by Sawana et al. (2014). Both R. andropogonis and P. rhizoxinica currently represent the only members of their respective lineages for which whole genome sequences are available. Future studies should therefore seek to identify their respective congeneric species [some of which will likely include those in Transition Group 1 (Estrada-de los Santos et al., 2016)] and to understand the biological and evolutionary properties underlying these two lineages.

In addition to allowing unambiguous demarcation of the genera in Burkholderia sensu lato, this study also revealed, for the first time, the relationships among these taxa. Burkholderia sensu stricto has a well-supported sister group relationship with the clade containing Caballeronia, and Paraburkholderia. P. rhizoxinica is sister to the Burkholderia sensu stricto+Caballeronia+Paraburkholderia clade, while R. andropogonis occupies the most basal position in the tree. Knowledge about these relationships could inform hypotheses regarding the biology and evolution of these bacteria, especially in terms of virulence and pathogenicity. For example, Burkholderia sensu stricto primarily includes human and animal pathogens, while P. rhizoxinica and Robbsia are also represented by pathogens (Estrada-de los Santos et al., 2016; Lopes-Santos et al., 2017). Moreover, certain Caballeronia and Paraburkholderia species have also been isolated from clinical samples [e.g., ‘C. consitans’ and ‘C. turbans’ (Peeters et al., 2016) and P. fungorum (Coenye et al., 2001), and P. tropica (Deris et al., 2010), respectively]. The availability of a robust phylogenetic framework for these taxa would thus be invaluable for deciphering the processes and mechanisms involved in the evolution of these species.

CB, MPa, PM, SV, and ES: Original concept; analyses; interpretation of results, writing and proofreading. WC, JA, and EvZ: Analyses, writing and proofreading. MH, AC, MPi, KP, NV, NM, DS, TR, CD, NS, VM, NI, NK, and TW: Genome analyses; interpretation of results; proofreading. JB and WW Genome analyses; interpretation of results; proofreading.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.