A total of 357 brain samples were collected from different species, mainly from cattle and horses between 1991 and 2009, from the Argentinean National Rabies Network. Sampling included vampire bats (n = 4) and human (n = 2) cases. RABV was detected mainly by direct immunofluorescence and by mouse inoculation test (Koprowski 1996). Later, RABV isolates were sent to two National Reference Laboratories for antigenic characterization (DILACOT, SENASA or Instituto de Zoonosis ‘Dr Luis Pasteur’). Rabies isolates were characterized as antigenic variant 3 (AgV3) using a panel of eight monoclonal antibodies directed against the viral nucleoprotein kindly provided by the Centers for Disease Control and Prevention (Atlanta, GA, USA; Delpietro et al. 1997).

Viral RNA was extracted from rabies isolates using TRI-zol^® (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions. Reverse transcription and PCR amplification of the complete nucleoprotein gene were carried out using 001-921B and 550F and 304 primers as described previously (Streicker et al. 2010). Amplicons were purified using ExoSAP-IT^® (USB Corporation). Sequencing was performed using an Applied Biosystems 3730 DNA Analyzer. Successful sequencing was performed on 227 samples (183 complete and 44 partial N gene; Table S1, Table S2, Fig. 2 and ‘SamplingSites.kml’, Supporting information).

Complete N sequences introduced in this work (n = 183) were analysed along with all vampire bat-associated N sequences longer than 1250 bp available from GenBank up to 1 May 2013 (n = 607).

Sequences were aligned with clustalX v2.1 (Larkin et al. 2007) and edited with bioedit v7.0 (Hall 1999). Maximum-likelihood (ML) phylogenetic tree and bootstrap analysis (1000 replicates) were obtained using the raxml software with rapid bootstrap algorithm (Stamatakis 2006; Stamatakis et al. 2008) implemented in the CIPRES Science Gateway server (Miller et al. 2010).

In addition, the 5′ end of the nucleoprotein gene (264 bp) was analysed. The data set included sequences introduced in this work (n = 227) and all vampire bat-associated sequences from GenBank up to 1 May 2013 (n = 1005). Analysis was performed as described above.

The spatial and temporal structure was studied for the complete N sequences data set and for the NWA and NEA groups, using the Parsimony Score (PS) and the Association Index (AI) statistics, estimated with the bats v0.1 software (Parker et al. 2008).

Two analyses were performed for the complete N sequences data set. On one hand, it was analysed labelling tips according to the country of origin or the year of isolation (n sequences = 790). On the other hand, only sequences with information about the state/province or year of isolation (labelling by state/province or by year) were included (n sequences = 547).

For both analyses, AI and PS statistics were computed for a null distribution with 1000 replicates (expected values) and for the posterior distribution of trees obtained from a Bayesian analysis (observed values). This analysis was performed using an appropriate substitution model (estimated with the jmodeltest v2.1 software (Darriba et al. 2012), according to the Akaike Information Criterion), the uncorrelated lognormal (UCLN) molecular clock model (Drummond et al. 2006) and a constant population size implemented in beast v1.8 (Drummond et al. 2012). Analyses were run using BEAGLE library (Suchard & Rambaut 2009; Ayres et al. 2012) at the CIPRES Science Gateway server (Miller et al. 2010). Convergence was assessed by effective sample size (ESS) values higher than 200 using the tracer v1.5 software (Rambaut et al. 2013), and 10% of the sampling was discarded as burn-in.

In addition, the spatial and temporal structure was also studied for the NWA and NEA groups. Data sets were analysed labelling tips according to the province of origin or the year of isolation. For both analyses, the posterior distribution of trees used as input for bats was obtained from the phylodynamic analyses.

bats analysis for temporal structure allow us to evaluate the association among phylogeny and year of isolation in NWA and NEA data sets (i.e. how samples from different years of isolation are distributed in the phylogeny) but do not allow us to study whether the temporal signal is adequate for phylodynamic studies (i.e. how informative the years of isolation are to calibrate the phylodynamic analysis).

Thus, to study the temporal signal of NWA and NEA data sets, a previously described strategy was used (Faria et al. 2012). Briefly, strict molecular clock models including and excluding temporal information (dated versus contemporaneous tips) were compared through Bayes factor using the marginal-likelihood values estimated by the path sampling and stepping stone methods (Baele et al. 2012).

Sequences of Argentinean samples belonging to the two main groups described by the phylogenetic analysis were used in Bayesian coalescent analyses to study the spatio-temporal dynamics along with the substitution rate and the time to the most recent common ancestors (tMRCAs) for the different RABV lineages found in Argentina. Only sequences with complete sampling information (Department-Province and month-year) were included, and sequences from samples associated with documented or presumptive imported cases for each region were excluded (Table S1, Supporting information).

Analyses were carried out using an appropriate substitution model estimated with the jmodeltest v2.1 software, according to the Akaike Information Criterion, the uncorrelated lognormal (UCLN) molecular clock model and the GMRF Bayesian Skyride coalescent model (Minin et al. 2008) implemented in the beast v1.8 software package. A spatial diffusion process was modelled on time-measured genealogies over continuous sampling locations (geographical coordinates). Different diffusion models were tested: Homogenous Brownian, Cauchy and Gamma distributions (Lemey et al. 2010).

Convergence was assessed by effective sample size (ESS) values higher than 200 using the tracer v1.5 software, and 10% of the sampling was discarded as burn-in. The uncertainty in parameter estimates was evaluated in the 95% highest posterior density (HPD95%) interval. The diffusion model that best fits to the data was selected through Bayes factor, using the marginal-likelihood values estimated by the path sampling and stepping stone methods. Visualizations of the selected diffusion processes were generated with the spread v1.0 software (Bielejec et al. 2011).

In addition, to study the relation among the Ne x τ (Ne: effective population size; τ: generation time) as an approximation to the effective number of infections (median values for each year), the mean dispersal rates and rabies cases in cattle, a Pearson correlation test was performed using the graphpad instat software v3.1 (San Diego California USA, www.graphpad.com).

In spite of no changes in surveillance strategy were documented or epidemiologically linked cases were not processed, the distribution of sequences over the study period was not uniform. Thus, to determine whether results obtained with NEA data set could be influenced by a possible biased sampling, two analyses were performed. On one hand, to study a possible bias related to oversampling in some years, the median number of sequences per year was calculated (resulting 5.5 sequences per year). Based on that, the data set was randomly subsampled to obtain five sequences per year (the final data set contained 54 sequences), and analysis was carried out as previously described for the complete NEA data set.

On the other hand, to study a possible bias related to oversampling for hypothetical linked cases, data set was randomly subsampled to obtain only one sequence per department per year (the final data set contained 77 sequences), and analysis was carried out as previously described for the complete NEA data set.

The number of rabies cases was obtained from Departamento de Zoonosis y Vectores. Ministerio de Salud de la Nación and from Dirección Nacional de Sanidad Animal, SENASA Argentina.

Phylogenetic and statistical analyses on all available vampire bat-related complete N gene sequences showed both a geographical and a temporal structure (P < 0.001 for the AI and PS statistics; Fig. 1 and Table S3, Supporting information). Basal diversification events separated some sequences from Mexico (described as AgV8) and from Peru and Colombia (with an unavailable AgV pattern) from AgV3 sequences.

Two major groups were observed among AgV3 sequences. One of them comprised sequences from Central America and the north of South America, including sequences described as AgV11 from Mexico (Velasco-Villa et al. 2006). The other group was formed exclusively by South American sequences and presented six main clusters and subclusters named with roman numbers and letters, respectively, for description purposes only (Fig. 1).

Sequences from Argentina were located in four groups with highly supported bootstrap values (Fig. 1). The two largest groups were formed by sequences from northwestern (NWA, group V) and northeastern (NEA, within group VI) Argentina (Figs 2, S1 and S2, Supporting information). It is worth noting that three samples from NWA were found into the NEA group (Fig. S2, Supporting information). Although these samples were detected in NWA, animals presumably acquired RABV infection in the NEA region, considering documented cattle movements from the NEA to the NWA region (DNSA 2013a).

Furthermore, the NWA and NEA groups also showed a geographical and temporal structure when sequences were labelled according to its province of origin or the year of isolation (P < 0.001 for AI and PS statistics for both regions; Table S3, Supporting information).

In addition, some sequences from the Province of Misiones, which is part of the NEA region, bordering Paraguay and Brazil, were located in groups formed by sequences from Brazil [group II, Fig. S3(a), Supporting information], and from Brazil, Peru and Uruguay [group Ib, Fig. S3(b), Supporting information]. It is remarkable that group II included a high number of AgV3 sequences isolated from nonvampire bats.

Equivalent phylogenetic groups of sequences from Argentina were also observed in the analysis of partial N sequences, although lower bootstrap values were obtained (Fig. S4–S7, Supporting information). The lower phylogenetic resolution was also reflected in the loss of monophyly of the NEA group, owing to the introduction of sequences from Colombia and Ecuador. However, this analysis allowed observing crossing patterns of the RABV other than those evidenced in the analysis of the complete N gene finding sequences from NWA in the NEA group but also sequences from NEA in the NWA group.

First, to evaluate whether the temporal signal of sequences of the largest groups from Argentina (NWA and NEA) was adequate for calibrating phylodynamic studies, the marginal-likelihood estimates for models including and excluding sampling dates were compared through Bayes factor. For both data sets, BF favoured the model that incorporates sampling dates, suggesting a temporal signal enough for calibrating phylodynamic analysis. For NWA data set, it was obtained a log BF = 7.9 (for Path Sampling method) and a log BF = 8.5 (for Stepping Stone method), indicating a strong evidence against the model that excludes sampling dates (Kass & Raftery 1995). While for the NEA data set, BF resulted in log BF = 159.9 (for Path Sampling method) and log BF = 158.9 (for stepping stone method), strongly favouring the model that incorporates sampling dates.

Thus, the phylodynamic pattern of the NWA and NEA groups from Argentina was studied using complete nucleoprotein gene sequences under a Bayesian approach that considers the phylogenetic dispersion in time and space.

Homogeneous and relaxed diffusion models were tested for the NWA and NEA groups. For both cases, Gamma distribution was selected by the Bayes factor test as the best fitting model (Table S4, Supporting information).

Under this model, both groups showed similar median tMRCA dated in 1987 (HPD95% = 1959–1998) for NWA and in 1992 (HPD95% = 1991–1994) for NEA. Similarly, substitution rates were estimated as 1.9 × 10⁻⁴ substitutions/site/year (s/s/y) (HPD95% = 3.1 × 10⁻⁵ – 3.9 × 10⁻⁴) for NWA, and as 2.6 × 10⁻⁴ s/s/y (HPD95% = 1.9 × 10⁻⁴ – 3.4 × 10⁻⁴) for NEA.

Demographic reconstructions showed that viral diversity in NWA has remained almost constant since the tMRCA (Fig. 3). In contrast, in the NEA region, the effective number of infections has increased slightly since 1992, with a maximum in ~1998–2001, which seems to continue up to the present (Fig. 4). In addition, this increase in NEA was possibly characterized by a strong diversification process occurring soon after the tMRCA, preceded by an increase in the dispersal rate, suggesting that this diversification occurred simultaneously in different sites of NEA (Fig. 4 and ‘NEA_diffusion_process.kml’, Supporting information). The most recent cases would have derived from two main diversification foci located in the north of the NEA region (mainly in the Departments Pirané and Patiño, Province of Formosa) and in the south of the NEA region (mainly in the Department Laishi, Province of Formosa and in the Province of Chaco; Fig. 4, Fig. 5 and ‘NWA_diffusion_process.kml’, Supporting information).

In addition, the NWA group showed a median dispersal rate of 21.0 km/year (HPD95% = 0.9–40.6) with a maximum in 2007 (Fig. 3). Conversely, a median rate of 37.1 km/year (HPD95% = 28.9–44.1) was observed for the NEA group, which showed fluctuating patterns of high and low dispersal rates with marked increases in 1996, 2003 and 2007 (Fig. 4). It is worth noting that the NWA group showed higher uncertainty in all estimations performed, evidenced by wider HPD intervals.

Furthermore, increments in RABV dispersal rates preceded or accompanied the increments in the number of rabies cases in cattle that occurred in ~2008 in NWA, and in ~1996–1997 and ~2006–2008 in NEA. By visual inspection of the rate-through-time plot superimposed to the number of rabies cases in cattle for each region (Figs 3 and 4), a tendency of a positive correlation has been observed between dispersal rates and rabies cases in cattle but not for the effective number of infection and cases, particularly for NEA region.

However, the correlation analysis carried out to study the relation among these estimates and cases showed no significant correlation for neither the dispersal rates and cases nor the effective number of infections and cases for NWA and NEA regions (all P > 0.05).

Finally, to study the influence of a possible biased sampling, sensitivity analyses were performed on the NEA data set, showing consistent results with those obtained with the complete NEA data set, suggesting that results are not biased for an uneven distribution per year o per department (Table S5, Table S6, Fig. S8 and Fig. S9, Supporting information).