Symptomatic patients with suspected Zika virus infection were enrolled in a research study approved by the Brazil Ministry of Health (Certificado de Apresentação para Apreciação Ética 45483115.0.0000.0046, no. 1159.184, Brazil). During April 2015–January 2016, acute Zika virus infection was diagnosed for 15 patients whose serum samples tested positive by a qualitative reverse transcription PCR (RT-PCR) by using primers targeting the nonstructural 5 gene (8). Clinical samples were retested for Zika virus positivity by using a separate quantitative RT-PCR (QuantiTect SYBR Green PCR kit; QIAGEN, Valencia, CA, USA) and primers targeting the envelope gene (9). Metagenomic next-generation sequencing libraries were constructed from serum RNA extracts, as described (10,11; Technical Appendix). Pathogen identification from metagenomic next-generation sequencing data was performed by using the Sequence-based Ultra-Rapid Pathogen Identification bioinformatics pipeline (12; http://chiulab.ucsf.edu/surpi/). Results of the metagenomic analyses and identification of co-infections with chikungunya virus are reported elsewhere (13).

For Zika virus genome sequencing, 2 isolates (Bahia07 and Bahia09; Table) with Zika virus titers >10⁴ copies/mL generated sufficient viral metagenomic data for complete genome assembly. For the remaining samples with lower titers, metagenomic next-generation sequencing libraries were enriched for Zika virus sequencing by using xGen biotinylated lockdown capture probes (Integrated DNA Technologies, Redwood, CA, USA) designed to tile across all sequenced Zika virus genomes >10,000 nt in GenBank (http://www.ncbi.nlm.nih.gov/genbank) as of March 1, 2016. Capture probes were curated for redundancy at a 99% nt similarity cutoff. Enrichment was performed on the metagenomic libraries in pools of 8 libraries (including Zika virus–negative serum sample controls) by using the xGen lockdown probe protocol and the SeqCap EZ Hybridization and Wash Kit (Roche, Indianapolis, IN, USA). Eleven Zika virus genomes with >40% genome recovery (mean 69.4% ± 2.0%) were assembled (Table). Distribution of single nucleotide variants across the 11 recovered genomes exhibited distinct patterns (Technical Appendix Figure 1), indicating that the assembled genomes were unlikely to result from cross-contamination by a single high-titer Zika virus sample.

Multiple sequence alignment was performed by using MAFFT version 7 (http://mafft.cbrc.jp/alignment/software/); maximum-likelihood (ML) and Bayesian phylogenetic inferences were determined by using PhyML version 3.0 (http://www.atgc-montpellier.fr/phyml/) and BEAST version 1.8.2 (http://beast.bio.ed.ac.uk/), respectively. The best-fit model was calculated by using jModelTest2 (https://github.com/ddarriba/jmodeltest2; details in Technical Appendix). Coding regions corresponding to the 11 complete or partial genomes from Bahia were aligned with all published and available near-complete Zika virus genomes and longer subgenomic regions (>1,500 nt) of the Asian genotype as of April 2016 (mean sequence size 8,402 nt with 1,652 distinct nucleotide site patterns). The ML phylogeny was reconstructed by using the best-fit general time-reversible nucleotide substitution model with a proportion of invariant sites (GTR+I). Statistical support for phylogenetic nodes was assessed by using a bootstrap approach with 1,000 bootstrap replicates. A Bayesian molecular clock phylogeny was estimated by using the best-fitting evolutionary model (2); specifically, a GTR+I substitution model with 3 components: a strict molecular clock, a Bayesian skyline coalescent prior, and a noninformative continuous time Markov chain reference prior for the molecular clock rate.

Figure

Figure. Timeframe of Zika virus outbreaks in the Americas. A molecular clock phylogeny is shown with the Zika virus outbreak lineage estimated from complete and partial (>1,500 nt) coding region sequences. For...

The isolates from patients in Salvador clustered together within 1 strongly supported clade (posterior probability 1.00, bootstrap support 100%, Bahia clade C) (Figure; Technical Appendix Figure 2). This support is notable; most Zika virus genomes in this clade are incomplete, and uncertainty is accounted for in phylogenetic inference. The tree topology accords with previous findings (2,4,5), and time to most recent common ancestor (TMRCA) of the epidemic in the Americas is similar to that previously estimated (2) (American epidemic clade A; Figure). The overall ML and molecular clock phylogenies exhibited many well-supported internal nodes with bootstrap support >60% and posterior probability >0.80 (Figure; Technical Appendix Figure 2), although several nodes near the ancestor of clade A were less well supported.

The updated phylogenetic analyses, including the newly identified clade C, suggest that Zika virus was introduced in Bahia during March–September 2014. An isolate from Maranhão in northeastern Brazil (≈1,000 km from Bahia) is ancestral to the Bahia clade (posterior probability 1.00, bootstrap support 74%, northeastern Brazil clade B) (Figure; Technical Appendix Figure 2). The TMRCA of clade B (comprising the Bahia clade and the Maranhão sequence) is estimated to be September 2013–April 2014, an early stage of the epidemic. This TMRCA is consistent with the hypothesis that Zika virus in the Americas originated in Brazil (2). A previously reported sequence from Bahia (6) clustered with an isolate from Belém in the state of Pará in northern Brazil, ≈3,000 km from Bahia (posterior probability 0.99, bootstrap support 81%) (Figure; Technical Appendix Figure 2). The patient denied history of travel, suggesting that multiple Zika virus lineages may circulate in Bahia.

Our results suggest an early introduction and presence (mid-2014) of Zika virus in the Salvador region in Bahia, Brazil. Given the size of the cluster and statistical support for it, this lineage likely represents a large and sustained chain of transmission within Bahia state. Most cases of this Zika virus lineage clustered closely to a sequence from Maranhão, and we found evidence for an additional potential introduction to Bahia from Pará state. Consequently, Zika virus in Salvador during mid-2014 was likely introduced from other regions in Brazil rather than from outside the country. Current findings of Zika virus emergence in Bahia state during mid-2014 are consistent with first-trimester viral infection in pregnant women corresponding to the initial reported cases of fetal microcephaly, which began in January 2015 (5) and peaked in November 2015.

Broader sampling across Bahia is needed to determine whether the Salvador lineage (clade C) identified in this article comprises most Zika virus cases in the state. Brazil currently faces a major public health challenge from co-circulation of Zika, dengue, and chikungunya viruses (2–4,14,15). Additional molecular surveillance in the Americas and beyond is urgently needed to trace and predict transmission of Zika virus.