We included 16 DENV-4 isolates in this study (Table). The viruses corresponded to low-passage virus strains (passage no. 1) from C6/36 cells obtained from the Department of Arbovirology and Hemorrhagic Fevers, Instituto Evandro Chagas, Brazilian Ministry of Health (Ananindeua, Brazil).

Nearly complete genome sequences were obtained by using high-throughput sequencing on a GS FLX+ System (454 Life Sciences, Branford, CT, USA) as described (11). The sequences of the 5′ and 3′ termini were obtained by using 5′ and 3′ rapid amplification of cDNA ends systems (Invitrogen, Carlsbad, CA, USA), according to the manufacturer’s instructions, with the following specific sets of primers: DENV-4 5′SP1 (5′-AKCCCTGTCTTGGGTCCAGC-3′), DENV-4 5′SP2 (5′- TTGAACGCCTCTTGAAGGTC-3′), DENV-4 5′SP3 (CTCTTGAAGGTCCAGGTCTA), DENV-4 3′SP1 (5′-ATATCTGAATGGCAGCCATC-3′), and DENV-4 3′SP2 (5′-TCACTGGCTGTTTCTTCTGCT-3′). All 5′ and 3′ rapid amplification of cDNA ends amplicons were cloned into a plasmid bacterial system by using the TOPO TA Cloning Kit (Invitrogen) and directly sequenced (in both directions) by using the ABI Prism BigDye Terminator v1.1 Cycle Sequencing Kit on an ABI Prism 3130 DNA analyzer (Applied Biosystems, Foster City, CA, USA) with the M13F/M13R set of primers (Invitrogen).

Data generated by the GS FLX+ System and ABI 3130 platforms were assembled by using the mapping reference method as implemented by the GS Reference Mapper Program (available in Newbler v.2.6 software, http://454.com/products/analysis-software/index.asp). The mapper program was used to rearrange the reads against a given reference sequence by using the following default parameters: input = 20 bp, all contig threshold = 100, large contig threshold = 200, minimum overlap length = 40, minimum overlap identity = 70%, k-mer = 12 (seed step), and k-mer = 16 (seed length). A total of 16 new full-length DENV-4 sequences were obtained and deposited in GenBank (accession nos. JQ513330–JQ513345).

We complemented our data with available full genome and envelope protein sequences for a total of 98 full genome (10,624 nt) and 314 full-length DENV-4 envelope protein (1,485 nt) sequences. These included 18 sequences from Brazil (3 from Belém, Pará State; 1 from Santarém, Pará State; 5 from Manaus, Amazonas State; 8 from Boa Vista, Roraima State; and 1 from Salvador, Bahia State); 26 full genome and 75 envelope protein sequences from the Caribbean region (Puerto Rico, Dominica, Dominican Republic, Jamaica, Puerto Rico, Trinidad and Tobago, Barbados, Montserrat, Martinique, and Bahamas); 8 full genome and 10 envelope protein sequences from Colombia; 34 full genome and 48 envelope protein sequences from Venezuela; 9 full genome and 88 envelope protein sequences from Mainland Southeast Asia (Thailand-Bangkok, Vietnam, Cambodia, Myanmar, and Peninsular Malaysia); and 3 full genome and 22 envelope protein sequences from Maritime Southeast Asia (Philippines, Indonesia, Singapore, and East Malaysia).

We aligned the sequences by using MAFFT (12). After we manually edited the resulting alignment by using Se-Al (13), the total lengths of the full genome and envelope protein alignments consisted of 10,624 bp and 1,485 bp, respectively. All sequences were screened for recombination by using the Phi-test (14), which is available in the SplitsTree4 program (15).

The parameters of a full probabilistic model of evolutionary history, including timed sequence evolution and spatial dispersal, were estimated by using a discrete Bayesian asymmetric diffusion approach (16) implemented in the BEAST software package (13,17). We used the Bayesian skyride model (18) as a flexible tree prior and a general time-reversible model with a discretized gamma distribution (general time-reversible + 4Γ) to account for among-site rate variation. To calibrate the time scale of the trees, we obtained isolation dates (in years) from the GenBank annotations, and to accommodate rate variation among lineages, we used a lognormal relaxed molecular clock approach (19).

This approach enabled us to estimate ancestral spatial locations throughout the phylogenetic history while accounting for uncertainty in the phylogenetic and diffusion process (17). For the full genome and envelope protein datasets, 3 Markov-chain Monte Carlo analyses were run for 50 million states and sampled once every 10,000 states. We used the BEAGLE library (13) together with BEAST (13) to augment the computational speed. After we removed 10% of the burn-in, we combined the runs by using LogCombiner (www.molecularevolution.org/software/phylogenetics/beast). Maximum clade credibility (MCC) trees were summarized by using TreeAnnotator and visualized by using FigTree (12). We use the SPREAD application (20) to visualize and convert the estimated divergence times and spatial estimates annotated in the MCC trees to a keyhole markup language file (the files are available from authors upon request). All evolutionary parameters are reported as posterior means along with their 95% Bayesian credibility intervals.

To obtain the expectations for the location state transitions, we estimated Markov jump counts (18,21) along the branches of the posterior tree distribution (22). We applied kernel density estimation in R to summarize Markov jump densities through time from particular locations.

The analysis (Phi-test implemented in SplitsTree) of the 98 DENV-4 full genomes did not provide support for recombination. Therefore, all 98 complete nucleotide (10,624 nt) and 314 (1,485 nt) envelope protein sequences were included in the Bayesian phylogeographic analyses by using a discrete asymmetric diffusion model and a flexible demographic prior (Technical Appendix Table).

The complete genome and envelope protein MCC genealogies confirmed that 3 major phylogenetic groups were compatible with the genotypes currently established for DENV-4: genotypes I–III (Figure 1). In particular, genotype I (n = 5 full genome, n = 85 envelope protein sequences) included strains that were isolated in Southeast Asia and a unique strain (BHI 3581 Be H 781363) that was isolated from an autochthonous febrile patient in 2011 in Salvador, Bahia State. The dense sampling in the envelope protein dataset made it possible for us to establish a relatively recent common ancestor with an Asian genotype I dating back to 2004 (95% Bayesian credible interval [BCI] 2002–2007). In particular, results from the genomic and envelope protein phylogeographic analyses confirmed with high posterior probability (>0.99) that the Brazilian strain from Bahia was introduced from Mainland Southeast Asia.

Genotype II (n = 86 full genome, n = 223 envelope protein sequences) included the other 17 strains sampled in Brazil (Municipalities of Manaus, Belém, Santarém, and Boa Vista) and in Venezuela, Colombia, Central America, the Caribbean, and Southeast Asia. Genotype III (n = 2 full genome, 4 envelope protein sequences) included strains that were exclusively isolated from Southeast Asia (Thailand). On the basis of the envelope gene analysis, we reconstructed 3 introductions of genotype II DENV-4 in Brazil (Figure 1, left panel). First, the envelope protein MCC tree indicated a Brazilian cluster of strains identified in Boa Vista and Manaus as part of the major South American genotype II cluster. Second, the full genome MCC tree broke the cluster in Boa Vista and Manaus into 3 different lineages in the same genotype II cluster. Third, the large South American genotype II cluster, as part of the full genome MCC genealogy (Figure 1, right panel), indicates additional introductions of genotype II into Brazil. Thus, we identified a total of 6 introductions of DENV-4 in Brazil.

The most recent common ancestor of genotypes I–III is believed to have a Southeast Asian origin (7). In line with recent findings (23), we estimate that genotype II, the predominant genotype, was likely first introduced into the Americas (Caribbean region) around 1978 (95% BCI 1976–1979) (Figure 1; Technical Appendix Figure). Following a period of circulation in the Caribbean region, this genotype independently entered South American countries >4 times, as inferred respectively by the full genome and envelope protein analyses. Most of these introductions into South America seem to have led to relatively shallow phylogenetic clusters that probably reflect short-term survival of the virus in South America.

Nevertheless, our data provide support for a major genotype II cluster that reflects the successful establishment of this genotype in South American countries neighboring Brazil in or around 1994 (95% BCI 1992–1995). From Venezuela and Colombia, the genotype spread to Brazil >3 times in the last decade, as inferred by the full genome phylogeographic analyses (Figure 1, right panel). In contrast, the DENV-4 outbreak in Pará State (Santarém and Belém) was identified around 2008 (95% BCI 2006–2009) and is directly linked to strains from the Caribbean region. Overall, we found >3 distinct DENV-4 introductions into Roraima State: 1 from the Caribbean region ≈3 decades ago (indicated as 1 in Figure 1, left panel), and 2 from the strain that has become endemic in Venezuela within the last decade (indicated as 3 and 5 in Figure 1, right panel). One of these strains resulted in a well-supported cluster that dated back to about 2006 (95% BCI 2005–2007) and comprised 6 strains.

To identify the major export and import locations for DENV-4 at a global scale, we summarized the number of introductions from and to each of the locations that were estimated by using Markov jump count approaches as part of the full genome analysis (Figure 2, panel A). Our results pinpoint Maritime Southeast Asia, the Caribbean region, and Venezuela as the 3 most prominent sources of DENV-4 export. In addition, to investigate the temporal dynamics of viral dispersal from the 3 locations in the evolutionary history of DENV-4, we estimated the Markov jump density from these locations through time (Figure 2, panel B). We distinguished different peaks, which appear to reflect 3 distinct waves of DENV-4 migration; the most recent wave corresponded to exportations from South American countries bordering Brazil, particularly Venezuela, where DENV-4 has been established longer.

After an initial establishment in the Caribbean region, genotype II most likely first dispersed to Brazil (Roraima State), Venezuela, Colombia, and, more recently, to Brazil (Belém, Pará State) again (Figure 3). Locations such as Santarém, Pará State, and Manaus, Amazonas State, do not have a direct relation to the Caribbean locations; instead, the strain found in Santarém is linked to a cluster of sequences from other locations in Pará State. This finding further indicates that there is ongoing circulation of DENV-4 within Brazil. Nevertheless, most recent introductions of genotype II in Brazil in the last decade seem to have been fueled by strains originating from bordering countries.