A total of 114 bats collected from their natural habitats from December 2006 through July 2007 (8 species from 10 locations; Table) were euthanized after being deeply anesthetized with 2% xylazine and 10% ketamine administered subcutaneously (in some cases after sedation with CO₂). The bats were then taxonomically classified on the basis of morphology, and the carcasses were stored at –70°C until used. For sampling, the carcasses were thawed at 4°C for 3–4 hours, then oropharyngeal and anal samples were taken with Dacron-tipped swabs that were then placed in RNAlater (Ambion, Austin, TX, USA) and stored at –20°C until used.

CoV detection and sequencing were conducted as previously described (4). Briefly, viral RNA was extracted from swabs by using the QIAamp viral RNA minikit (QIAGEN, Westburg, the Netherlands) and used as the template for reverse transcription–PCR (RT-PCR) detection of the CoV RNA–dependent RNA polymerase (RdRp) gene (9). Primers based on the RdRp gene, conserved for all known coronaviruses, were then used for RT-PCR detection. The RdRp PCR products were gel purified by using the QIAquick PCR purification kit (QIAGEN) and sequenced to confirm virus species. RNA from samples positive for coronavirus was then used for cDNA synthesis by using random hexamer, gene-specific, and oligo(dT) primers. The RdRp gene and 1b open reading frames, including the helicase (HEL) domain, were then sequenced, also as previously described (4). Sequences derived from this study were deposited in GenBank (accession nos. EU769557 and EU769558).

Sequences were aligned with previously published CoV sequences from GenBank by using ClustalX (http://bips.u-strasbg.fr/fr/Documentation/ClustalX) then manually aligned by using the Se-Al program (http://tree.bio.ed.ac.uk/software/seal). The GenBank accession numbers of all sequences used are noted in the taxon names in Figures 1 and 2. The RdRp sequences were trimmed to equal length, which created 2 datasets of 780 bp (n = 40) and 378 bp (n = 45). The latter included Bt-CoV sequences from North America and Germany that were too short to be included in the first dataset. A third dataset comprised an alignment of the HEL domain (n = 46) trimmed to 1,797 bp. Maximum likelihood (ML) phylogenies were inferred under a General Time Reversible (GTR + Γ₄ + I) model, which was identified as the best-fit model of nucleotide substitution using MODELTEST version 3.7 (10). Bootstrapping was performed to assess the robustness of tree topologies by using 1,000 replicate neighbor-joining (NJ) trees under the ML substitution model. All analyses were performed with PAUP* version 4.0b (Sinauer Associates, Inc., Sunderland, MA, USA).

CoV RNA was detected in 1 of 21 Glossophaga soricina and 1 of 5 Carollia perspicillata bats tested. The latter, designated Bt-CoV/Trinidad/1FY2B, was an adult male from the area of Fyzabad, and virus was detected in anal and oropharyngeal swabs. The infected G. soricina specimen (Bt-CoV/Trinidad/1CO7B) was an adult female from the area of Couva, and the virus was detected in the anal swab only. Subsequent sequencing was performed on virus from the anal swabs. A total of 3,905 bp of RdRp from Bt-CoV/Tri/1CO7B and 5,160 bp from Bt-CoV/Tri/1FY2B were sequenced, in addition to 1,782 bp from the HEL domain of both samples (full sequence data not shown). Across the shared 3,798-bp region of RdRp, divergence was 28.7% at the nucleotide level and 15.5% at the amino acid level. Similarly, the HEL domains of the 2 viruses showed 27.3% divergence at the nucleotide level and 13.3% at the amino acid level.

ML phylogenies inferred from the RdRp genes of 40 viruses (780 bp) and the HEL domains of 46 viruses (1,797 bp) are shown in Figure 1. In both cases, the 5 groups proposed by Tang et al. (4) on the basis of NJ trees, and the lineage containing the recently reported novel CoV sequences (11), were strongly supported with bootstrap values >95% in all cases. In each case the Trinidadian sequences clustered with group 1 CoVs within a clade containing all other group 1 bat and human CoVs as well as porcine CoV. To determine the phylogenetic relationship between the Trinidadian Bt-CoVs and North American Bt-CoVs, for which only relatively short sequences for RdRp were available, we inferred a second RdRp ML phylogeny based on a 378-bp fragment (Figure 2). When this shorter fragment was used, 5 groups were again defined, but bootstrap support for the putative group 5 (4) was lower. The level of divergence between Trinidadian sequences was notably higher than among the North American (6) and German sequences (7).

Our detection of RNA from group 1 CoVs in Trinidadian bats shows that Bt-CoVs have a wider distribution than previously suspected and is added support for bats as the original host species for these viruses. Group 1 CoVs form 2 well-supported clades designated 1a and 1b (12). The Trinidadian bt-CoV clustered within the latter clade, which contains all other group 1 Bt-CoVs, including those from Germany and North America, and the 3 known group 1 human CoVs associated with respiratory illness (13–15). Despite the geographic proximity of the bats from which the Trinidadian Bt-CoV sequences were derived—Couva and Fyzabad are 28 km (17 miles) apart, and Trinidad has an area of only 4,769 km² (1,864 square miles)—they are relatively highly divergent. This divergence might reflect virus adaptations to different host species; however, more data would be needed to confirm this. Given the mobility of bats, the possibility of the viruses having different geographic origins (perhaps even from outside Trinidad) cannot be ruled out. Further work on CoV diversity in Trinidad and the rest of the Americas, as well as on the ecology and behavior of susceptible bat species, is needed to understand the origins, evolution, and dispersal of these viruses.