Since 2004, the Ministry of Agriculture and Rural Affairs of China has implemented the Rabies Surveillance Plan with a focus on free-roaming and stray dogs and cats, especially those showing abnormal behaviors such as biting humans. The plan also requires the monitoring of suspected rabies outbreaks in livestock and wild animals. During 2004–2018, brain tissues of 185 animals suspected of having rabies (dead dogs, dogs behaving strangely or biting humans, livestock showing rabieslike clinical signs, dead foxes, wolves, and raccoon dogs) were submitted (Appendix Table 1). In addition, 10,118 brain tissues were collected for active surveillance from 3 types of apparently healthy dogs, mostly from rabies-endemic rural areas: free-roaming and ownerless dogs, slaughtered dogs (for meat consumption), and dogs killed during emergencies (culled in rabies outbreak areas to prevent further transmission) (Appendix Table 2). All specimens were collected and submitted to the reference laboratory by the regional Centers of Animal Disease Prevention and Control.

The brain tissues were examined by the direct fluorescent antibody test (FAT) using FITC-conjugated anti-rabies monoclonal antibody (Fujirebio Diagnostics Inc., https://www.fujirebio.com) (13). To obtain the complete coding sequence of the N gene, rabies-positive specimens were subjected to RNA extraction using TRIzol Reagent (Invitrogen, https://www.thermofisher.com), followed by reverse transcription PCR with the SuperScript III First-Strand Synthesis System and the Platinum Taq DNA Polymerase High Fidelity kit (Invitrogen) (6).

Amplified N gene sequencing was performed commercially by the Sanger method and submitted to GenBank (see Appendix Table 3 for accession numbers). Phylogenetic analysis of the complete N gene was performed on these sequences and on representative sequences retrieved from GenBank, covering samples collected in China and its neighboring countries from the 1940s through 2018 (Appendix Table 3). The MEGA 7 program package was used to construct the phylogenetic trees using the neighbor-joining method with 1,000 bootstrap replicates (14). Trees were visualized using Figtree version 1.4.2 (http://tree.bio.ed.ac.uk/software/figtree).

To rank the prevalence of the different RABV phylogroups and to analyze their transmission trends in China, we retrieved the sequences of all RABV strains from China deposited in GenBank. After we removed duplicate sequences and those without clear time information, we phylogenetically classified the remaining sequences, along with those obtained during this study, by the procedure described previously and chronologically sorted them by collection date.

To investigate the temporal signal and clock likeness of molecular phylogenies based on the N gene dataset, the linear evolutionary rates of different RABV clades were estimated using the Bayesian Markov chain Monte Carlo in BEAST version 1.8.2 package (15,16). For these analyses, we selected the general time reversible model as the substitution model and gamma plus invariable sites as the site heterogeneity model based on the calculations of Model Generator (17,18). An uncorrelated log normal relaxed molecular clock model and the constant size model as a coalescent tree prior were also selected for the analyses, which were run for 100 million steps with sampling at every 10,000 states (19). The BEAGLE parallel computation library was used to enhance the speed of the likelihood calculations (20). Finally, the resulting log file was checked using TRACER version 1.5 (http://tree.bio.ed.ac.uk/software/tracer) to confirm that all effective sample sizes were >200. The tree file was obtained using TreeAnnotator version 1.8.2 with a burn-in of 10%, and the maximum clade credibility tree was visualized by FigTree version 1.4.2 (16). Based on the analyses, estimations were made of the rates of nucleotide substitution and the time to most recent common ancestor (tMRCA) for each RABV clade.

To investigate the phylogeographic spread of RABVs in China and neighboring countries, we used a Bayesian stochastic search variable selection (BSSVS) approach to analyze transmission routes of RABV subclades, in which we applied a Bayes factor to determine the best supported transmission event between 2 geographic locations. Bayes factors were calculated by SpreaD3 software with a value >3 as cutoff (21,22).

During 2004–2018, animal brain tissues collected from 185 animals with suspected rabies in 17 provinces were submitted to our laboratory; 144 of them (77.8%) were confirmed by FAT as rabies virus positive (Appendix Table 1). Among the positive species, dogs were the main infected animals, accounting for 68.8% of total cases (99/144), followed by cattle (12.5%), sheep (9.7%), camels (4.2%), foxes (2.1%), pigs (1.4%), raccoon dogs (0.7%), and donkeys (0.7%) (Appendix Table 1). Concurrently, 33 (0.33%) of 10,118 brain tissue samples taken during an active surveillance of apparently healthy dogs from 7 provinces across China were found to be FAT-positive. Of the 33 positive specimens, 31 were from free-roaming and ownerless dogs (including stray dogs) and 2 were from dogs killed during an emergency (Appendix Table 2). For livestock rabies, 29 cattle, sheep, and camel cases were reported in Inner Mongolia and Xinjiang during 2013–2018, all of which were caused by fox bites (Appendix Table 1).

A total of 108 complete N genes were amplified from 177 positive brain tissue samples. Of these, we selected 78, representing different years, animal species, outbreaks, and locations, together with 222 reference sequences from China, as well as from neighboring and other countries, to determine their phylogenetic characteristics (Appendix Table 3). Animal RABVs in China were clustered within 4 major clades: Asian, Cosmopolitan, Arctic-related, and Indian Subcontinent, together with different subclades (Figure 1). The Asian clade, the most prevalent one, widely distributed throughout China and Southeast Asia (SEA) countries, shows abundant genetic diversity and is transmitted mainly by dogs. This lineage was further divided into 5 subclades and different lineages. SEA1, 2, 3, and 5 subclades circulate mainly in populous areas within China; however, some lineages and strains in SEA1 and 3 were also found circulating in some countries in Southeast Asia, whereas SEA4 was restricted to the Philippines (Figure 2). Reported only in China, ferret badger RABVs were found to have abundant genetic diversity and were segregated into different lineages within SEA1, 2, and 5 (Figure 1) (23). Within the Cosmopolitan clade, which includes some vaccine strains, a steppe-type subclade was fox-transmitted and distributed along north and northwest border areas and neighboring Mongolia, Russia, and Kazakhstan, forming 2 major lineages (Figure 1). Other Cosmopolitan strains were dog-transmitted and mainly distributed in populous inland areas (Figure 2). Arctic-related RABVs in China segregated within the AL2 subclade and formed 2 lineages, one in northeastern China and far eastern Russia, Mongolia, and South Korea with dogs and raccoon dogs as the major hosts, and the other in southwestern China with dogs as the major transmission source (Figures 1,2). The Indian Subcontinent clade had not been identified in China until the first human rabies case caused by this clade was identified in 2017 in the border area of Tibet close to Nepal (24). That human case was caused by the bite of a local stray dog and remains the only Indian Subcontinent clade RABV confirmed so far in China.

For the chronological sorting of different subclades, we retrieved all 2,486 RABV sequences from China deposited in GenBank. After removing repeated sequences and those without date information, 1,118 eligible sequences remained, representing 1,118 Chinese strains isolated during 1969–2018 (including those in Figure 1). These sequences included complete genome (n = 45), full length or partial N (n = 819), glycoprotein (G; n = 208), phosphoprotein (P; n = 15), matrix protein (M; n = 25) and RNA-dependent RNA polymerase (L; n = 6) genes. Figure 3 shows the spatial–temporal trends of different RABV subclades in China, in which the 55 Chinese RABV sequences submitted to GenBank between 1969 (the earliest submission) and 2003 (therefore listed chronologically as “before 2004”) segregated within 3 SEA and 1 Cosmopolitan subclade. Following initiation of official rabies surveillance in 2004, numbers of sequences submitted to GenBank sharply increased and high numbers of submissions have been maintained thereafter. The resulting data showed clearly that most rabies outbreaks have been caused by strains of the Asian clade (93.3%), with limited involvement of strains of the other 3 clades. Within the Asian clade, the subclade SEA1 predominated in rabies endemics in China (70.1%), followed by SEA2 (16.7%). SEA1 is the most widely distributed of the subclades and continues to spread. The steppe-type subclade first emerged in 2011 and has rapidly become predominant among the livestock RABV strains found along border areas in Inner Mongolia and Xinjiang Province (Figure 2). AL2 was first detected in 2007 and has become a common subclade in recent years (10). The Indian Subcontinent clade caused an occasional case in 2017. The result showed that steppe-type, AL2, and Indian Subcontinent strains are emerging RABVs in China. Dog-transmitted Cosmopolitan strains have not been detected during the past decade.

Results of the Bayesian skyline model analysis showed that the mean rate of nucleotide substitution for the tested RABVs was 3.50 × 10⁻⁴ substitutions per site per year (95% highest posterior density 2.90–4.11 × 10⁻⁴ substitutions per site per year). This finding is consistent with the previous analyses of evolutionary change performed on dog-related RABV N genes (25). Differences in evolutionary rates among the clades and subclades were not significant. All representative RABVs in China and neighboring countries shared a tMRCA, predicted to merge 349–563 years ago (Figure 4). Further analysis of transmission routes of RABV subclades by BEAST revealed the transboundary transmissions of rabies in different border regions around China. The significant translocation pathway of SEA1 (Bayes factor 76.9) (Appendix Table 4) from China to Indonesia was identified in accordance with our previous analysis of the G gene (Figure 5) (26). Moreover, many SEA3 strains in Myanmar, Thailand, Laos, and Vietnam were genetically close to some strains circulating in Yunnan and Guangxi, China, indicating mutual transmission of SEA3 strains between China and bordering SEA countries (Bayes factor 4.3–85.0), as discussed previously (27). The same transmission was also found for steppe-type and AL2 subclades in border regions between China and Kazakhstan, Russia, Mongolia, and South Korea (Bayes factor 3.17–229.87). The most noteworthy event was the recent cross-border transmission of an Indian subcontinent strain from Nepal to the border region of Tibet, albeit with a lower Bayes factor (0.9), which caused a human rabies death in 2017 (24).