By combining information derived from the phylogram of bat CoVs, together with data on the geographic origin of viruses, we were able to describe the phylogeographic distributions for known CoVs from bats in China (Figures 1, 2; Table). Bat SARS-like CoVs formed a monophyletic clade. Species-specific host restriction was found for CoVs in 4 of 7 bats species (Myotis ricketti, M. magnater, P. abramus, and T. pachypus) sampled from >1 geographic location, and these clustered with high Bayesian posterior probability. Overall phylogenetic relationships between virus lineages were similar across our analyses, and well-supported genetic structure was observed within some CoV lineages. For example, CoVs isolated from M. magnater were monophyletic but formed 3 well-supported clades with no evidence of geographic structure (Bayesian posterior probability [PP] = 1.0 for each). A similar pattern was apparent in CoVs from Myotis ricketti, which formed 2 geographically overlapping independent clades (PP = 0.99 and 1.0, respectively). One T. pachypus was infected by a virus that clustered with moderate statistical support (PP = 0.91) within the larger clade associated with P. abramus, which indicated a potential interspecies transmission event or recent evolutionary host shift (defined as interspecies transmission followed by establishment and long-term persistence in the new host species [18]) (Figure 1).

Phylograms of host sequences were also constructed and were essentially of the same topology with high support whether derived by using MrBayes version 3.1.2 or MEGA3 (data not shown). When we mapped host phylogram to virus, virus phylogeny did not always track host phylogeny (Figure 3). When separate host-virus phylograms were constructed for the 2 bat families (Verspertilionidae and Rhinolophidae), different corresponding relationships were evident. Verspertilionids and their CoVs showed phylogenetic congruence, and rhinolophids and their CoVs showed incongruous phylogenies (Figure 4).

We found evidence for evolutionarily divergent relationships for some vespertilionid viruses and their hosts when analyzed at the family scale (Figure 4, panel A). For example, divergence between viruses harbored by P. pipistrellus and P. abramus is congruent with their hosts. The divergence among other viruses was incongruent with divergence of host species, e.g., those from S. kuhlii and Myotis ricketti.

Rhinolophid bats and their viruses were analyzed at a different taxonomic scale (within genus). In this co-phylogeny, viral host shifts were the evident virus-host feature (Figure 4, panel B). Except for R. macrotis, all rhinolophidae bats had 2 distinct lineages of CoVs, and host shifts were found among viruses carried by R. ferrumequinum, R. pearsoni, and R. sinicus.

Genetic diversity of CoVs harbored by rhinolophids and vespertilionids was similar (vespertilionids π = 0.27 ± 0.13; rhinolophids π = 0.25 ± 0.13). In contrast, genetic diversity of cyt b sequences from bats was much higher among the vespertilonids (π = 0.17 ± 0.007) than among the rhinolophids (π = 0.09 ± 0.006).

CoVs sequenced from different bats of the same species clustered together, even when bats were collected in locations 1,000–2,000 km apart. This pattern was found for CoVs from P. abramus, T. pachypus, Myotis ricketti, and M. magnater. Bats of the genus Miniopterus are known to migrate long distances (30), which explains why the phylogeny of viruses isolated from M. magnater sampled in distant places (Guangxi, Anhui, Fujian, and Hong Kong) lacks geographic structure. In nonmigrating species such as bats of the genera Pipistrellus and Tylonycteris, intimate physical contact of bats in same cave or the same bamboo roost site, as well as periodic exchange of bats among neighboring colonies, may facilitate virus transmission among populations.

Despite the co-roosting of many bats species, we found little evidence of host shifts for some viruses. For example, CoVs from M. magnater and Myotis ricketti sampled in the same cave in Guangxi were divergent, although sample size was limited. Although Myotis ricketti has a closer phylogenetic relationship with T. pachypus, P. pipistrellus, and P. abramus than with M. magnater and S. kuhlii, its behavior and habits are closer to those of the last group. For example, Myotis ricketti and S. kuhlii bats roosts in caves (although S. kuhlii also roosts under palm leaves), whereas T. pachypus roosts inside bamboo and P. abramus roosts almost entirely in old buildings. Thus, it seems plausible that the close phylogenetic relationship between viruses harbored by Myotis ricketti and S. kuhlii reflects the similar behavior and ecology of their hosts.

The phylogenetic and phylogeographic associations we found suggest that there may be a coevolutionary relationship between some bat CoVs and their hosts. For example, sister taxa within the genus Pipistrellus independently maintained 2 distinct viruses that share a most recent common ancestor. A similar relationship was apparent among the viruses of some closely related genera (e.g., Pipistrellus and Tylonycteris), whereby divergence of each genus was mirrored by divergence in viral phylogeny. However, viruses are usually thought to have evolved more recently than their hosts (31). Thus, apparent coevolutionary patterns may reflect either a high frequency of host shift among closely related bat species or simultaneous lineage splitting of hosts and viruses. Host shifts among related bats might be favored by a variety of mechanisms, including preadaptation to overcome immune defenses or greater rates of interspecific contact relative to distantly related bat species. Phylograms with better resolution would enable statistical comparison of phylogenetic congruence and estimation of divergence times.

In the vespertilionids, close phylogenetic concordance between host and virus suggests a close, possibly evolutionarily divergent relationship. However, there are different scales of comparison between the Vespertillionidae, in which all but 1 CoV came from separate genera, and the Rhinolophidae, in which we examined a co-phylogeny of multiple species within 1 genus. Genetic diversity in the vespertilionids sampled was nearly double that of the rhinolophids, which was probably due to the greater number of species sampled and their broader taxonomic range. Despite this greater genetic diversity among vespertilionid bat hosts, the genetic diversity of CoVs did not differ between vespertilionids and rhinopholids. This diversity suggests that vespertilionids may maintain undiscovered CoVs or that rhinolophids might harbor disproportionate CoV diversity relative to diversity of their genus. We propose that future work may identify more vespertilionid bat CoVs, which would enable an accurate comparison of propensity for host shifts within this group.

In the rhinolophids, the host phylogram demonstrated genetic divergence between R. ferrumequinum and other species, as shown by the division of Rhinolophus bats into 2 groups. Each of these groups harbors CoVs from 2 clusters (SARS-like CoVs and other CoVs), which suggests multiple introductions of CoVs into these species.

Lack of concordance between phylogenies of rhinolophid bats and their CoVs can be interpreted as evidence for host shifts between bats of the genus Rhinolophus. Different species of Rhinolophus are often observed roosting inside the same cave, which facilitates virus transmission between species. However, the degree of host shifting of rhinolophid bat CoVs may not be particularly high relative to other genera of bats. This observation will be clarified when a greater diversity of CoVs from other bat genera is reported and the sequences are analyzed. These requirements support the need for further research on bat viruses (32,33).

Host-shifting within the genus Rhinolophus would likely be promoted if these bats shed CoVs in a way that makes them more available to other Rhinolophus spp.; had behavioral traits that lead to increased contact with other Rhinolophus spp.; or if CoVs harbored by these bats have structural, biologic, or other traits that make them more readily transmitted to other Rhinolophus spp. Two lines of evidence suggest that host traits are the most parsimonious explanation for host shifts within the genus Rhinolophus. First, SARS-like CoVs and other rhinolophid CoVs (RfV1 and RpV1) show evidence of interspecies transmission. Second, CoVs from other bat groups that are phylogenetically much closer to RfV₁ and RpV₁ than to the SARS-like CoVs do not show evidence of successful host shifts. Thus, the ability to jump hosts is unlikely to be a strictly viral trait.

The phylogeography of bat CoVs suggests that the bat SARS-like CoVs form a monophyletic clade that is both phylogenetically distinct from other bat CoVs and geographically isolated. Although we acknowledge that this interpretation may be limited by sample size, it may also indicate that rhinolophid bats, the hosts of a cluster of SARS-like CoVs within which human and civet SARS CoV nestle phylogenetically (6,7), are more likely to foster the host shifts of CoVs than are other bat species. The potential for close contact between bats, civets, and humans in the wildlife trade in southern China, coupled with a possible propensity of these bats to foster CoV host-shifts, could explain SARS-like CoVs as the source of SARS-CoV. This potential supports molecular results on bat CoVs that suggest a recent host shift from bats to civets or other animals and humans (34). Such host shifts may indicate a risk posed by other bat CoVs for novel disease emergence. Finally, the ability of SARS-like CoVs to be transmitted between and establish in new species (i.e., to undergo host shifts) is consistent with other CoVs. This has been shown for several CoVs of livestock species (35) and has been used to support their inclusion as 1 of the groups of viruses most likely to be responsible for emerging zoonoses, even before the emergence of SARS (36).

The total diversity of CoVs (including SARS-like CoVs) in bats has likely not been fully described. The genus Rhinolophus (8) contains 77 bat species distributed in Asia, Europe, and Africa. The recent discoveries of bat CoVs in the United States (37) and SARS-COVs in African bats (38) support the hypothesis that CoVs are diverse and widespread in bat species. Given the diversity of CoVs in this group, and their propensity for host shifts, further viral discovery in rhinolophids may assist in understanding and ultimately controlling the emergence of zoonotic viruses. Bats are increasingly recognized as reservoirs of many highly lethal zoonotic agents (32). Understanding their diversity, behavior, and mechanisms of virus transmission may play a key role in preventing future outbreaks of both known and unknown zoonotic diseases of bat origin.