Follow up of a robust meta-signature to identify Zika virus infection in Aedes aegypti: another brick in the wall

The mosquito Aedes aegypti (L.) is the main vector of several globally distributed diseases (Lorenzo et al. 2014 ⁸ ). One of these diseases is dengue (DENV) that affects more than 2.5 billion people (WHO 2015 ²² ). Moreover, other illnesses such as yellow fever (YFV) are endemic to tropical regions (Bae et al. 2005 ² ) and have recently re-emerged in close proximity to major urban centres in Brazil (Paules and Fauci 2017 ¹³ ), while West Nile fever (WNF), which usually is associated with small outbreaks, presents high mortality rates (Pradier et al. 2012 ¹⁴ ). Chikungunya virus (CHIKV), once localised to parts of Africa, has now spread globally (Cauchemez et al. 2014 ⁴ ). Zika virus (ZIKV) emerged in 2015 in the Americas (Zanluca et al. 2015 ²³ , Faria et al. 2016 ⁶ , Slavov et al. 2016 ¹⁷ ) following sporadic outbreaks in the Pacific in 2007 (Micronesian Island Yap) and 2013-14 (French Polynesia). Initially, it was not considered dangerous as it is a self-limiting disease, but later, it was associated with a 'microcephaly outbreak' (Tang et al. 2016 ¹⁹ ), leading Brazilian authorities to declare a national health state of emergency and the World Health Organization to designate the Zika epidemic as a public health emergency of international concern (WHO 2016 ²¹ ).

Previous studies were performed to elucidate altered pathways of Ae. aegypti in response to viral infections, and our group recently identified a viral infection meta-signature for DENV, WNV and YFV by investigating the relationship between feeding and infection (Fukutani et al. 2018 ⁷ ). We identified a set of 110 genes highly correlated with viral infection, of which four genes (AAEL012128, AAEL014210, AAEL002477, and AAEL005350) were highly informative in identifying the infection-status classification in the vector. The role of the genes AAEL014210, AAEL002477 and AAEL005350 is regulatory based on the prediction of a zinc finger, DNA-binding domain (InterPro accession no. IPR013087), and basic-leucine zipper domain (InterPro accession no. IPR004827), and these genes harbour retinaldehyde binding and alpha-tocopherol transport domains (InterPro accession nos. IPR001071 and IPR001251, respectively). The gene AAEL012128 is described as having a 12-pass transmembrane protein with a cationic amino acid transporter and was previously reported as a retrovirus receptor (Wang et al. 1991 ²⁰ , Fukutani et al. 2018 ⁷ ).

The datasets used in the previous studies lacked ZIKV-infected samples, which were unavailable at the time. Recently, a next-generation sequencing study with Ae. aegypti infected with ZIKV was published (Etebari et al. 2017 ⁵ ) with a new data, allowing a test of our signature in this infection. Our hypothesis is that the expression of signature genes, as we have demonstrated for other viruses, also plays an important role in the classification of the ZIKV infection in the mosquito. To achieve this study, we assessed the NCBI/BioProject PRJNA399504 (https://www.ncbi.nlm.nih.gov/bioproject/399504) and downloaded all the raw data. In total, there are 18 paired-end replicates corresponding to six samples (three ZIKV-infected and three controls) (Etebari et al. 2017 ⁵ ). The fastq-dump from the SRA toolkit (NCBI 2011 ¹⁰ ) was used to obtain sequence files in FASTQ format. Sequences were filtered for low quality reads and adapters using Trimmomatic version 0.32 GNU (Bolger et al. 2014 ³ ), and transcripts were quantified using the Ae. aegypti reference transcriptome AaegL3.5 as a reference within Salmon v0.9.1 (Patro et al. 2017 ¹² ). Transcripts were summarised at the gene-level using the R package tximport (Soneson et al. 2015 ¹⁸ ), yielding a count table. The count table was filtered using the edgeR package (Robinson et al. 2010 ¹⁶ ), and only genes consistently expressed [counts per million (cpm) greater than 0.5] were kept. Each sample-expression value was determined by its sample-replicates mean, as specified by the BioProject metadata. To test our signature, a nonmetric multidimensional scaling (NMDS) was performed using the metaMDS function within the R package vegan v2.4.5 (Oksanen et al. 2017 ¹¹ ). metaMDS aims to represent the position of samples in a multidimensional space, as accurately as possible, using a reduced number of dimensions (axes). The axes resulting from metaMDS were used in a logistic regression model to access the predicted probability of a sample belonging to a given class (i.e., ZIKV- or mock-infected). To compare both models (the complete set of 110 genes or the restricted set of four genes), the area under the receiver operating characteristic (ROC) curve (AUC) was calculated using the pROC package (Robin et al. 2011 ¹⁵ ).

First, we tested our previously identified signature composed of 110 genes. This set consists of the previously correlated genes with infection without the blood-feeding influence (Fukutani et al. 2018 ⁷ ). The nMDS model was able to discriminate the groups (infected mosquitoes with ZIKV and uninfected samples) (Fig. 1A). The same approach was used with the smaller gene set (four genes) (Fig. 1B). Both the 110 and four gene sets were able to discriminate the ZIKV-infected from the uninfected samples with 7 x 10^-2 and 2 x 10^-4 stress values, respectively (lower stress values indicate a more reliable ordination of the dataset). To measure the accuracy of each gene set, we calculated the area under curve (AUC); using the 110 gene dataset, an AUC of 0.94 was found, whereas using a four gene dataset, the value was 0.83 (Fig. 2). However, there is no significant difference between both the AUC of both gene sets (D = -1.48, boot.n = 2000, p-value = 0.13). Despite this result, our original meta-signature was identified for the mosquito classification infected with YFV, WNF and DENV, and when applied in ZIKV-infected mosquitoes, these genes were able to discriminate the infected and healthy samples. In addition, irrespective of the set of genes selected (the 110 or four gene set), the signature remained robust with good predictive power for the classification of the Zika infection.

The limitation of the current approach is the relatively low number of samples re-analysed (n = 6). However, there are few datasets available of mosquitoes infected with the ZIKV. Another available dataset of Ae. aegypti (GSE96605) does not have an adequate number of samples (three samples: one DENV, one ZIKV and one control with three technical replicates) (Angleró-Rodríguez et al. 2017 ¹ ). Despite this limitation, our results showed that these gene sets are a powerful framework for future studies of vector infections. Applications of this meta-signature are promising and suggest a process that could be similar in other vector infections such as CHIKV and Oropouche virus. However, to date, there are no publicly available datasets of vectors infected with these viruses. These results improve the predictive power of our previously identified meta-signature as a consistent signature that is able to identify mosquitoes infected by DENV, YFV, WNF and now ZIKV. This result suggests common processes are involved in all infections. Moreover, these processes impact vector pathways related to the maintenance of virus replication, such as host protein machinery and amino acid transportation processes (Mosso et al. 2010 ⁹ ).

ACKNOWLEDGEMENTS

To Dr Cleyson Barros and Dr Herculano da Silva for their insights that greatly improved the manuscript. We thank Ms Raíza Tourinho and Mr Olival Rocha for their assistance.

AUTHORS' CONTRIBUTION

EF, MR, JIK, CFA and ARP downloaded the data and performed the analysis; PIR, KFF and ATLQ wrote the manuscript.