Among mammalian phylogenies, those characterized by rapid species radiations have long been one of the most plaguing and challenging problems in species tree reconstruction . This is the first study utilizing data from whole mitochondrial genome sequences from ursids, an approach that allows increased phylogenetic resolution of the Ursidae family, whose origin can be traced back to the extremely recent mid-Miocene[6, 9] (15–20 MYA). Previous molecular studies relevant to Ursidae phylogeny provided either an inconsistent view on the issue or weak statistical support for discriminating alternative hypotheses (Figure 1). The branching event following the divergence of the spectacled bear has long been a large unresolved polytomy leading to six Ursus species, of which only the sister-relationship between the brown bear and polar bear was unanimously favored. The close relatedness of brown bear and polar bear, as well as the paraphyletic association between mtDNA of these two bears has been upheld in previous studies [29–31]. More sequences of brown bear and polar bear included in the future research will help test further the earlier observations. In sum, the long-standing lack of full resolution within Ursidae may be primarily due to the low level of variation harbored in much shorter sequences than those used in this study.
Based on the largest available mt data set from Ursidae, our genome phylogeny provides strong evidence that within genus Ursus, the sloth bear is the sister taxa of all the other five ursines, and that the latter group is divided into the brown bear/polar bear and the two black bears/sun bear assemblages, upholding and strengthening the hypothesis drawn by our previous analysis of the five fragments of mtDNA  (Figure 1E). Alternative hypotheses for a mitochondrial sequence based phylogeny are not supported when the entire mitochondrial DNA sequence information is utilized. In particular, when nucleotide base compositional bias introduced by the giant panda outgroup was removed from our analysis and the spectacled bear was used for rooting, the overall Ursidae relationships recovered have the largest statistical support in comparison to all other previously proposed hypotheses. To further examine the sensitivity of our tree topology to outgroup choice, we selected Pinnipedia. This superfamily in Carnivora includes the Otariidae, Phocidae, and Odobenidae families, members of which have been used as alternative roots in earlier studies of bears [8, 12].
Combined protein-coding genes (10888 aligned sites, 4522 variable and 3245 parsimony-informative) and all mt genes (15017 aligned sites, 5642 variable and 3926 parsimony-informative; control region was not included due to alignment difficulty) analyses of the eight bear species using available Pinnipedia mt genomes as outgroups (Accession No. AJ428576, AJ428578 and X63726; [32, 33]) gave an identical tree topology to that obtained using the giant panda as the outgroup. Support levels for most branches were similar to those estimated with the giant panda outgroup except for an increased ML BS (≥ 85) and PP (≥ 0.95 PP) of placement of the sloth bear (Figure 2A and 3). A chi-square test of composition stationary showed that both Pinnipedia and the giant panda have deviant base composition with respect to the other bears (p < 0.05), a circumstance that might have a negative impact on the branch supports in MP analysis. Thus, our genome phylogeny was robust with either outgroup, though the spectacled bear, exhibiting the least phylogenetic noise, appears a more favorable outgroup for Ursidae in terms of overall support levels. Taken together, the present genome result significantly resolved the conflict between those trees using partial mt genes [8, 11–13] and represents the most probable explanation of bear evolution.
Nevertheless, a more in-depth understanding of the Ursidae relationships will definitely benefit from the addition of independent sequence data, considering that the genome phylogeny obtained here is based on a single and haploid linkage unit. The necessity of including other unlinked genes for phylogenetic resolution of the Ursidae is also illustrated by the fact that our recent study on two nuclear genes, transthyretin (TTR) and interphotoreceptor retinoid binding protein (IRBP), has united the sloth bear and the sun bear as sister taxa with high statistical support  (Figure 1 nuA), a relationship not consistent with mt gene analyses, including the present study. In mt trees, the sloth bear was mostly placed as the earliest diverging taxa among the six ursine species [8,11,13, and the present result], and the sun bear closer to the American black bear , the brown bear/polar bear clade [MP phylogeny in 12; ML phylogeny in 8], or the clade including the two black bears (NJ phylogeny in 12; 13 and the present result). Notably, the sister relationship between the Asian and American black bears, previously proposed on paleontological and morphological grounds [34, 35] was reinforced by consistent recovery from both our mt genome and nuclear analysis. Such a grouping has also been retrieved previously with moderate support by mt analysis of complete CYTB and 2 tRNAs , as well as the addition of the partial D-loop region . Thus, the placements of the sun bear and the sloth bear represent are the most obvious discrepancy observed in the mt and nuclear trees comparisons. Our genome analysis has established a very useful benchmark that can be tested with future independent evidence.
Our genome analyses provide important insights into not only Ursidae phylogeny, but also the phylogenetic utility of different mt genes. Our data add to the well-studied performance of individual mt genes, mostly protein-coding genes, for estimating phylogeny of deep divergence [16, 36], we are interested to see their relative efficiencies, adding mt RNA genes and control region as well, in those of extremely recent split. Our results suggest that combined mt protein-coding genes are more informative than the other subsets of mt genes regarding the lower-level bear relationships resolution. Only by combining all genes is it possible to reach a fully-resolved tree with moderate to strong support from MP, ML, and Bayesian methods of analysis. Ranking single genes by their respective contribution to the total PBS values of the genome tree, as a rough indicator of phylogenetic utility, reveals that some genes, such as ND5, ND4, CYTB, ND2, and 16SrRNA are better indicators of Ursidae evolution than are other genes, such as ND3, ND4L, and ND1 (Figure 5). Our results add to previous findings from Zardoya and Meyer (1996)  and Russo, Takezaki, and Nei (1996)  that did not included concatenated tRNAs, 2 rRNAs, and control regions in the evaluation of phylogenetic performance, and also agree globally with their conclusions about the rough classification of 12 mt protein-coding genes into good, medium, and poor categories. These conclusions are upheld even though significantly different evolutionary time frames between our studies and theirs (i.e., distantly related vertebrates) (Figure 5) are involved. In this sense, general knowledge of phylogenetic values of the mt genes makes it possible to preselect subsets of mt genes for different-level phylogenetic questions in the case of mt genomes unavailable. In fact, in some previous studies, Ursidae phylogenies based on the combined analysis of a few mt genes have also, to a degree, demonstrated the potential valuable information as those based on complete genome analysis [8, 12].
Ursidae has one of the most extensive fossil records of extant Carnivora families [9, 26, 27, 35, 37]. Given a good fossil documentation and strongly supported phylogenetic relationships from the largest available data set, it is of interest to draw a comparison between the present mitogenomic dating results and those from previous paleontological and molecular evidence (Table 3). According to estimates based on our genome data, all the separation times within genus Ursus appear to have occurred in the Late Miocene or Early Pliocene (5–6 MYA), except that in the Early Pleistocene the most closely related bear species, the brown and polar bears diverged. The rather recent origins and rapid succession of these bear lineages are in line with the observation that most short mtDNA sequences used in previous studies lack sufficiently strong phylogenetic signals and provide limited resolving power for recovering a strongly supported Ursidae phylogeny. As is seen in Table 3, our estimates of the diversification of the Ursidae family was more in agreement with those obtained with partial mt genes analysis  and protein electrophoresis analysis  than those obtained with the fossil record [9, 26, 35] and nuclear sequence analysis , which are slightly younger than the present results.