Codon adaptation biases among sylvatic and urban genotypes of Dengue virus type 2.

Dengue virus (DENV) emerged from the sylvatic environment and colonized urban settings, being sustained in a human-Aedes-human transmission chain, mainly by the bites of females of the anthropophilic species Aedes aegypti. Herein, we sought evidence for fine-tuning in viral codon usage, possibly due to viral adaptation to human transmission. We compared the codon adaptation of DENV serotype 2 (DENV-2) genotypes from urban and sylvatic habitats and tried to correlate the findings with key evolutionary determinants. We found that DENV-2 codons of urban and sylvatic genotypes had a higher CAI to humans than to Ae. aegypti. Remarkably, we found no significant differences in codon adaptation to human between urban American/Asian and sylvatic DENV-2 genotypes. Moreover, CAI values were significantly different, when comparing all genotypes to Ae. aegypti codon preferences, with lower values for sylvatic than urban genotypes. In summary, our findings suggest the presence of a molecular signature among the genotypes that circulate in sylvatic and urban environments, and may help explain the trafficking of DENV-2 strains to an urban cycle.

Introduction

Dengue virus (DENV) is the world's most important mosquito-borne viral human pathogen. It is widespread throughout tropical and subtropical regions (Bhatt et al., 2013; Gubler, 1998). DENV has a natural sylvatic maintenance cycle, involving species of blood-feeding arboreal mosquitoes of the genus Aedes (Ae. taylori, Ae. furcifer, Ae. vitattus, and Ae. luteocephalus) and nonhuman primates, that develop viremic infections and serve to amplify transmission (Amarasinghe et al., 2011; Diallo et al., 2003). Urban cycle involve amplification hosts and peridomestic Aedes spp. vectors, mainly Ae. aegypti (Gubler, 1998; Vasilakis et al., 2011).

DENV is an enveloped virus of the genus Flavivirus, family Flaviviridae, classified in four phylogenetically and antigenically distinct serotypes (DENV-1-4). Its genome is made by a single-stranded, positive-sense RNA molecule of approximately 11 kb, which contains a single open reading frame encoding three structural [capsid (C), membrane (M) and envelope (E)] and seven non-structural (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5) proteins (Chambers et al., 1990). Based on their genetic diversity and geographical distribution, distinct groups have been also identified within each serotype, defined as genotypes (Cologna et al., 2005; Costa et al., 2012).

Six distinct DENV-2 genotypes are recognized (Costa et al., 2012; Twiddy et al., 2002; Wei and Li, 2017). The Sylvatic genotype circulates in a sylvatic cycle, while the American, American/Asian, Asian 1, Asian 2 and Cosmopolitan genotypes circulate in urban cycle (Wang et al., 2000). Inferred evolutionary relationships of DENV suggest that the urban genotypes evolved from sylvatic lineages due to several cross-species transmission events into humans, followed by the recent (i.e., early 20th century) evolution of urban forms (Messina et al., 2014; Wang et al., 2000; Wei and Li, 2017). Notably, environmental and social factors such as the introduction of the anthropophilic vector, urbanization, deforestation and the increased human travel and trade have facilitated the emergence, spread and evolution of DENV into human populations (Mayer et al., 2017; Parvez and Parveen, 2017). Elements such as genetic composition and pathogen-host interaction are believed to be involved in different patterns of transmissibility (Jenkins and Holmes, 2003; Lara-Ramírez et al., 2014; Lequime et al., 2016; Lobo et al., 2009).

The degeneracy of the genetic code refers to the fact that a single amino acid can be coded by different codons. The redundancy in the genetic code has an important role in controlling metabolic processes, but not all species use its built-in codon redundancy in the same way. The unequal preference of specific codons over other synonymous codons during the translation process creates a bias in codon usage. Codon usage biases are common throughout the Tree of Life, and for viruses, the balance between mutation and natural selection allows for changes in codon usage biases (Morton, 2003). Genetic plasticity and the capacity to adapt to new hosts facilitate the emergence of RNA viruses in novel, unexplored environments. This process entails both, (i) adapting to different types of cellular machinery involved in viral replication and (ii) the evasion from different types of cellular antiviral responses. However, the mechanisms used for virus for trafficking among hosts remain poorly understood (Bahir et al., 2009; Longdon et al., 2014; Pal et al., 2014). Because DENV must infect successfully and alternate between mosquito and human hosts, nucleotide analyses support the notion that arboviruses use a restricted and balanced nucleotide composition as a compromise to be able to infect both hosts, thus successfully ensuring several processes such as translational efficiency and replication (Shen et al., 2015). In this context, the Codon adaptation index (CAI) correlates with gene expression levels and adaptation of viral genes to their hosts (Gustafsson et al., 2012; Neame, 2009; Pal et al., 2014). In the present study, we performed comprehensive analyses of codon adaptation of DENV-2 from different habitat settings and host systems.

Material and methods

Sequence datasets

The sequences used in this study were downloaded from the National Center for Biotechnology Information (NCBI) (https://www.ncbi.nlm.nih.gov/genbank/) website in GenBank format. The dataset included 877 sequences from all six genotypes as follows: American (16), American/Asian (435), Asian 1 (228), Asian 2 (20), Cosmopolitan (163) and Sylvatic (15). Comparisons of codon usage preferences were performed against reference sets from Homo sapiens (3803 CDSs) (https://github.com/CaioFreire/CUB) and Aedes aegypti (585 CDSs) (http://www.kazusa.or.jp/codon/), using the standard genetic code. All the complete genomic sequences available for DENV-2 with information regarding the location and year of isolation were recovered, and later converted into FASTA format.

Molecular phylogenetic approach

FASTA sequences were aligned using Clustal Omega (Larkin et al., 2007) and recombinant sequences were screened using all algorithms implemented in RDP4 program (RDP, GENECONV, BootScan, MaxChi, Chimaera, Siscan and 3Seq) using the standard settings (Martin et al., 2015). The alignment of recombinant free sequences was manually inspected and edited using the program AliView v.1.18 (Larsson, 2014), resulting in a final dataset with 877 coding sequences (Table S1). Viral phylogenies based on full-length coding sequences were estimated using Maximum Likelihood (ML) implemented in FastTree 2 (Price et al., 2010). We first evaluated the best transition model to be GTR + Γ using JModelTest v. 0.1.1. The final tree was then visualized and plotted using FigTree v.1.4.3 (http://tree.bio.ed.ac.uk). All sequences used in this work are presented in the format: genotype/accession number/country/year of isolation.

CAI of DENV to humans and Aedes aegypti mosquitoes

CAI is a measure of silent, synonymous codon usage bias based on the codon preference of a viral strain and a codon usage table for a given host (Sharp et al., 1986; Sharp and Li, 1987). We applied the CAI using a frequency table for housekeeping of human genes (Eisenberg and Levanon, 2013) (available at https://github.com/CaioFreire/CUB) and for Ae. aegypti using the table available in the Codon Usage Database (Nakamura, 2000). The CAI values were calculated to measure the synonymous codon usage bias using the CAIcal program (Puigbò et al., 2008a). To evaluate the statistical support of the CAI values, we define a threshold value or expected CAI (e-CAI) (Puigbò et al., 2008b) by generating random sequences with GC content, amino acid composition and sequence length similar to the DENV-2 query sequences. CAI values above the e-CAI are interpreted as statistically significant, meaning that codon similarity arises from codon preferences rather than from internal biases (Puigbò et al., 2008b).

Statistical analysis

Statistical significance was determined by the Mann-Whitney U test to compare the difference between CAI values for humans and the mosquito, assuming a significance level of 0.05. Multiple comparisons among genotypes were performed using a Kruskal-Wallis test implemented in the pgirmess v1.6.5 package for the R statistical environment. To evaluate levels of independence, we further carried out a Dunn's post hoc test. Further, to avoid Type I error inflation, we applied a Bonferroni adjustment of p-values, using the PMCMR package, at a significance level of 0.05. Perl and R scripts were used to analyze most of the data in this study and are available from the authors upon request.

Results

As has been well documented, we recovered six phylogenetically structured groups or genotypes for DENV-2, based on the open reading frame of 877 non-recombinant sequences. The most basal group encompassed lineages from West Africa and Asia of Sylvatic genotype, which reinforced the notion that the urban genotypes emerged from the sylvatic cycle, as previously described (Vasilakis et al., 2011; Vasilakis et al., 2007a). Urban genotypes were also identified as: (i) American, (ii) American/Asian, (iii) Asian 1, (iv) Asian 2 and (v) Cosmopolitan genotypes (Fig. 1A, Fig. S1). Comparison of the CAI values obtained from human and Ae. aegypti codon usage tables, shown in Fig. 1B, revealed significant differences between hosts (p-value <0.05), demonstrating that DENV-2 codon usage is more similar to that of humans than to Ae. aegypti.

Fig. 1

DENV-2 genotypes. (A) Summary ML tree displaying the phylogenetic relationships between urban and sylvatic genotypes. (B) CAI values distribution of DENV-2 genotypes for human and vector hosts by using entire polyprotein sequences. The number of sequences for each genotype was distributed as follows: (i) American: 16 sequences; (ii) American/Asian: 435 sequences; (iii) Asian 1: 228 sequences; (iv) Asian 2: 20 sequences; (v) Cosmopolitan: 163 sequences; (vi) Sylvatic: 15 sequences. Genotypes are coded using different colors. The e-CAI is indicated for each host in the figure.

In addition, we assessed the values of CAI for each genotype in comparison to human and Ae. aegypti hosts. Because CAI values for the polyprotein sequences were not normally distributed, as assessed using Shapiro-Wilk and Kolmogorov-Smirnov tests, we determined statistical significance with the Mann-Whitney U test to compare among CAI estimates. The distribution of CAI values for different genotypes showed that all the genotypes fell above the e-CAI estimated for each host. Furthermore, statistical analyses showed significant differences (p-value < 0.05) among urban genotypes for both hosts (Table 1 ).

Table 1

Multiple comparison of CAI values among genotypes. CAI values estimated in human and Ae. aegypti cells are represented in gray and white cell background, respectively. Significant differences (*p-value <.05) between pairs of genotypes are indicated by asterisk symbols.

ns – Non statistical significance.

Significant differences between urban and sylvatic genotypes were identified in Ae. aegypti (Table 1). Likewise, the genotypes with the highest CAI were Asian 1, Asian 2 and the American genotypes. We also observed that the CAI values for the sylvatic genotype were the lowest when comparing with urban genotypes in A. aegypti (p-value < 0.05) (Fig. 1B and Table 1). This suggested that genotypes from urban setting could have experienced some fine-tuning process in codon optimization to translation in A. aegypti.

For humans, the only pairwise comparison that was not significantly different between urban and sylvatic genotypes was that between the American/Asian and the Sylvatic genotype (Fig. 1B and Table 1), suggesting that they share similar silent selective pressures. The American, Asian 2 and Cosmopolitan genotypes had the highest CAI values for humans (Fig. 1B). Finally, we observed that these values were similar and roughly constant over time (Fig. 2 ). Crucially, values were higher than the e-CAI for all the genotypes in human and mosquito cells, indicating that codon usage preferences were not random in DENV-2 genotypes (Fig. 2).

Fig. 2

Polyprotein CAI values distribution through time. Colors correspond to different genotypes. Symbols indicate strains circulating in urban (circle) and sylvatic (triangle) cycles. The color-filled symbols represent the CAI values in human cells while open symbols represent the adaptation in Ae. aegypti cells. Horizontal lines depict the expected CAI thresholds for both hosts.

Discussion

The potential emergence of sylvatic strains into urban cycles has become an increasing focus of research and public concern (Moncayo et al., 2004; Vasilakis et al., 2011; Vasilakis et al., 2007b). Recently, this concern become more tangible after some studies have suggested human infection cases with presumably sylvatic DENV-2 strains. Phylogenetic analyses showed that the highly divergent strains (QML22/2015, D2Sab2015) of DENV-2 was basal and strikingly different from all other previously isolated strains (Liu et al., 2016; Pyke et al., 2017). In both cases, we excluded these sequences from our analyses, because including highly divergent sequences without knowing which genotype belong, may impact on the estimates for the whole dataset due to sampling bias and differences in the evolutionary rates.

Previous data suggest that the nature of selective pressures is broadly equivalent in urban and sylvatic strains (Vasilakis et al., 2011; Vasilakis et al., 2007b). This could mean that the successful emergence of a novel sylvatic DENV strains in an urban cycle is unlikely to require a major adaptive challenge (Vasilakis et al., 2007b).

Insects and mammals are separated by about 1 billion years of evolutionary history, and they differ at the organismal level and in many biochemical processes (Lobo et al., 2009; Shen et al., 2015). Since DENV needs to negotiate with different hosts (i.e. human and Aedes mosquitoes), it is fair to assume that the efficiency in the viral translation process is a key factor sustaining the urban cycle.

Our results suggested that codon adaptation was higher for human cells than for Ae. aegypti (Fig. 1, Fig. 2). This was also reported for other flavivirus (Butt et al., 2016; Freire et al., 2015; Moratorio et al., 2013; Pal et al., 2014). In addition, similar CAI values recovered for urban and sylvatic genotypes suggests their ability to possibly make similar use of human cellular machinery. This notion is consistent with kinetic experiments comparing the replication of sylvatic and urban DENV-2 strains, which could explain that the emergence of urban DENV-2 from sylvatic progenitors may not have required adaptation to replicate more efficiently in humans (Vasilakis et al., 2007b). Notwithstanding, when using the CAI tables from Ae. aegypti, we observed a conspicuous difference between urban and sylvatic genotypes with lower CAI values for the sylvatic genotype (Fig. 2). This finding suggested that the emergence of urban genotypes into the urban cycle may have required viral adaptation towards increased competence in Ae. aegypti, which should then entail higher efficiency in viral replication. This is in good agreement with susceptibility experiments reported by Moncayo et al. (2004). Furthermore, competence studies with Ae. aegypti from regions where sylvatic genotype circulation occurs indicate its low vector capacity for Dengue virus (Amarasinghe et al., 2011; Diallo et al., 2008). Despite this, there was no previous attempt to establish an association with codon usage biases.

Fluctuations in CAI distribution across genotypes (Fig. 1B) indicated possible differences in codon adaptation for both hosts (Table 1). Differential selection pressures exerted by the complex interplay of distinct virus factors, such as infection efficiency, population density and herd immunity (Plowright et al., 2017; Salazar et al., 2010). For example, a recent study reported by Salje et al. (2017) demonstrated the association between population density and the establishment of transmission chains. The increase in mosquito populations has also been identified as a potential determinant of the emergence and increased transmission of arbovirus, particularly in immunologically naïve human populations (Dudas et al., 2018; Pettersson et al., 2016). Our findings bring into question whether sylvatic strains could reach similar levels of transmission as the urban genotypes in Asian and American populations if the vector competence of Ae.aegypti is increased. In fact, recent outbreaks caused by other emerging viruses [e.g. Ebola virus, Influenza A virus (H1N1), Middle East respiratory syndrome coronavirus (MERS-CoV)], have occurred due to the zoonotic spillover and adaptation of the virus to replication in human cells (Dudas et al., 2018; Mänz et al., 2016; Plowright et al., 2017; Urbanowicz et al., 2016). In this context, as previous analyses have suggested (Moncayo et al., 2004), our findings support the notion that for the sylvatic strains to effectively colonize the urban environment, the virus needs a number of silent, adaptive nucleotide substitutions to optimize the codon usage to invertebrate host cells, while maintaining a compositional base balance suitable for efficient alternate spread among both human and insect hosts. Apparently, even when humans are susceptible to infection by DENV vectored by Ae. aegypti, its reduced vectorial competence ultimately constitutes a hurdle rather than an enabler of virus transmission.

Conclusions

In conclusion, our findings provide a comprehensive assessment of the codon adaptation of DENV-2 in different habitats (i.e. urban and sylvatic settings) and host systems (i.e. Homo sapiens and the mosquito vector Aedes aegypti). In this context, although the virus replicates in both human and mosquitoes, our analysis suggested that DENV-2 codon usage is better adapted to humans than to the main cosmopolitan vector (Ae. aegypti). This would imply that there is still room for adaptation and improved transmission among new DENV-2 strains for causing future pandemics.

The following are the supplementary data related to this article.

Table S1

GenBank accession numbers and isolation information for DENV-2 sequences used in phylogenetic and codon biases analysis.

Fig. S1

Maximum-likelihood tree (midpoint-rooted) inferred by using complete coding sequences recombinant free of DENV-2 to distinguish major virus genotypes; Numbers near nodes indicate percent SH-like values after 1000 replicates for major branches.

Acknowledgments and funding

This work was supported by the Brazilian National Council of Scientific and Technological Development (CNPq) process No. 441105/2016-5. MPC and ASOB received grants: No. 2016/08204-2 and No. 2013/25434-3, respectively. The funders had no role in the data collection or analysis, decision to publish, or preparation of the manuscript. We thank Felipe G. Naveca and Valdinete A. Nascimento for the support.