Genomic epidemiology supports multiple introductions and cryptic transmission of Zika virus in Colombia

The INS National Virological Reference Laboratory collected diagnostic specimens from over 32,000 suspected ZIKV cases over the course of the epidemic. Of these, roughly 800 serum specimens were positive for ZIKV by real time RT-PCR (rRT-PCR) and had cycle threshold (Ct) values less than 30. From this set we selected 176 serum specimens that were ZIKV-positive, but negative for dengue and chikungunya viruses, as per results from the Trioplex rRT-PCR assay [27]. Specimens were selected such that each Colombian department was represented over the entire time period in which specimens were submitted. We extracted RNA using the MagNA Pure 96 system (Roche Molecular Diagnostics, Pleasanton, CA, USA) according to manufacturer’s instructions. An extraction negative was used for each plate; positive controls were eschewed given the risk of cross-contaminating low titer clinical samples [20]. We attempted reverse transcription and PCR-amplification of ZIKV using the two-step multiplex PCR protocol developed by Quick and colleagues on all 176 extracted samples [20]. Briefly, cDNA was generated using random hexamer priming and the Protoscript II First Strand cDNA Synthesis Kit (New England Biolabs, Ipswich, MA, USA). We amplified cDNA using the Zika Asian V1 ZIKV-specific primer scheme [20], which amplifies 400bp long overlapping amplicons across the ZIKV genome, over 35 cycles of PCR. Amplicons were purified using 1 × AMPure XP beads (Beckman Coulter, Brea, CA, USA) and quantified using with the Qubit dsDNA High Sensitivity assay on the Qubit 3.0 instrument (Life Technologies, Carlsbad, CA, USA). Due to long storage periods and variable storage temperature, the vast majority of samples were too degraded to amplify. Of the 176 processed samples, only 15 amplified sufficiently to perform sequencing.

Universidad del Rosario (UR) collected and performed diagnostic testing on 23 human clinical samples from different geographic regions, and 38 Ae. aegypti samples from the Cordoba department of Colombia. RNA was extracted using the RNeasy kit (Qiagen, Hilden, Germany) and a single TaqMan assay (Applied Biosystems, Foster City, CA, USA) directed to ZIKV was employed [28] to confirm ZIKV presence. Approximately 60% of samples (14 clinical samples and 23 Ae. aegypti samples) were found to be ZIKV-positive by rRT-PCR. From these, we attempted amplification on 8 samples with sufficiently high viral copy numbers that they were likely to amplify. Amplification, purification, and quantification of ZIKV amplicons from UR samples were performed as described above. Of the eight samples that we performed PCR on, four samples amplified sufficiently to conduct sequencing; three samples were from human clinical specimens and one was from an Ae. aegypti sample.

We sequenced amplicons from 4 UR and 15 INS samples using the Oxford Nanopore MinION (Oxford Nanopore Technologies, Oxford, UK) according to the protocol described in Quick et al. [20]. Amplicons were barcoded using the Native Barcoding Kit EXP-NBD103 (Oxford Nanopore Technologies, Oxford, UK) and pooled in equimolar fashion. Sequencing libraries were prepared using the 1D Genomic DNA Sequencing kit SQK-LSK108 (Oxford Nanopore Technologies, Oxford, UK). We used AMPure XP beads (Beckman Coulter, Brea, CA, USA) for all purification steps performed as part of library preparation. Prepared libraries were sequenced on R9.4 flowcells (Oxford Nanopore Technologies, Oxford, UK) at the INS in Bogotá and at the Fred Hutchinson Cancer Research Center in Seattle.

Raw signal level data from the MinION were basecalled using Albacore version 2.0.2 (Oxford Nanopore Technologies, Oxford, UK) and demultiplexed using Porechop version 0.2.3_seqan2.1.1 (https://​github.​com/​rrwick/​Porechop). Primer binding sites were trimmed from reads using custom scripts, and trimmed reads were mapped to Zika reference strain H/PF/2013 (GenBank Accession KJ776791) using BWA v0.7.17 [29]. We used Nanopolish version 0.9.0 (https://​github.​com/​jts/​nanopolish) to determine single nucleotide variants from the event-level data, and used custom scripts to extract consensus genomes given the variant calls and the reference sequence. Coverage depth of at least 20x was required to call a SNP; sites with insufficient coverage were masked with N, denoting that the exact nucleotide at that site is unknown. After bioinformatic assembly, 8 samples produced sufficiently complete genomes to be informative for phylogenetic analysis.

All publicly available Asian lineage ZIKV genomes and their associated metadata were downloaded from ViPR [30] and NCBI GenBank. The full download contained both published and unpublished sequences; we sought written permission from submitting authors to include sequences that had not previously been published on. Any sequences for which we did not receive approval were removed. Additionally, we excluded sequences from the analysis if any of the following conditions were met: the sequence had ambiguous base calls at half or more sites in the alignment, the sequence was from a cultured clone for which a sequence from the original isolate was available, the sequence was sampled from countries outside the Americas or Oceania, or geographical sampling information was unknown. Finally, we also excluded viruses whose estimated clock rate was more than 4 times the interquartile distance of clock rates across all sequences in the analysis. Sequences that deviate this greatly from the average evolutionary rate either have far too many or too few mutations than expected given the date they were sampled. This deviation usually occurs if the given sampling date is incorrect, or if the sequence has been affected by contamination, lab adaptation, or sequencing error. After curation, the final dataset consisted of 360 sequences; 352 publicly available ZIKV full genomes from the the Americas (including Colombia) and Oceania, and the 8 Colombian genomes from the present study.

Data were cleaned and canonicalized using Nextstrain Fauna (https://​github.​com/​nextstrain/​fauna), a databasing tool that enforces a schema for organizing sequence data and sample metadata, thereby creating datasets compatible with the Nextstrain Augur analytic pipeline (https://​github.​com/​nextstrain/​augur) and the Nextstrain Auspice visualization platform (https://​github.​com/​nextstrain/​auspice). A full description of the Nextstrain pipelines can be found in Hadfield et al. [31]. Briefly, Nextstrain Augur performs a multiple sequence alignment with MAFFT [32], which is then trimmed to the reference sequence. A maximum likelihood phylogeny is inferred using IQ-TREE [33]. Augur then uses TreeTime [34] to estimate a molecular clock; rates inferred by TreeTime are comparable to BEAST [34], a program that infers temporally-resolved phylogenies in a Bayesian framework [35]. Given the inferred molecular clock, TreeTime then creates a temporally-resolved phylogeny, infers sequence states at internal nodes, and estimates the geographic migration history across the tree. These data are exported as JSON files that can be interactively visualized on the web using Nextstrain Auspice.

We generated a series of datasets in which ZIKV genomes from either Colombia or Mexico were subsampled. For each value from 1 to n, where n is the total number of available sequences for the country of interest, we generated 30 subsampled datasets in which n sequences were randomly sampled without replacement. Each subsampled set was then combined with all ZIKV genomes from other countries included in the main phylogeographic analysis described above, and the phylogeographic inference was rerun. For the Colombian rarefaction analysis, we have n=20 high quality whole genomes, so we inferred 20×30=600 phylogeographically labelled trees. For the Mexican rarefaction analysis, we have n=51 high quality whole genomes in total, and inferred 51×30=1530 phylogeographically labelled trees. For each labelled tree, we used custom scripts to traverse the resulting phylogeny and count the number of ZIKV introductions into the country of interest. We then plotted the inferred introduction count as a function of the number of sequences sampled.